Taiwan Semiconductor Manufacturing (NYSE:TSM) is already worth about $2.2 trillion, with shares of the chip foundry trading at about $427 as of this writing.

My prediction: The company's market value passes the $3 trillion mark before 2029. To be specific, that means sometime before the end of 2028, about two years and four months away.

That may sound like a bold call. The stock would need to reach about $580 per share, about 21% above its 52-week high of $479.

But the yearly return the milestone requires is more ordinary than it sounds. And it's a fraction of the pace TSMC's business is growing at today.

Image source: TSMC.

TSMC needs about 14% a year to get there

Going from about $2.2 trillion to $3 trillion is a gain of about 35%. Spread over that stretch, it works out to about 14% compounded annually.

For a business growing the way TSMC is right now, that isn't a high bar.

I'm not assuming investors pay more for each dollar of TSMC's earnings than they do today, either. If the stock's price-to-earnings multiple simply holds steady, the share price should track earnings growth over time. In other words, earnings compounding at about 14% a year through 2028 could arguably get the company there on its own.

A 40% year

Highlighting how far ahead of that bar the business is running, TSMC's second-quarter revenue rose 36% year over year to NT$1.27 trillion ($40.2 billion in U.S. dollars), while net income surged 77%. Gross margin was 67.7%, a big step up from 58.6% a year before. And the momentum has carried into the second half of the year. July revenue rose about 45% year over year, putting revenue through the first seven months of 2026 up 37%.

Management expects more of the same. Guidance calls for third-quarter revenue of $44.6 billion to $45.8 billion. Against the year-ago quarter's $33.1 billion, the midpoint represents about 37% growth -- an acceleration from the second quarter's pace in dollar terms.

In July, management also raised its full-year outlook to revenue growth slightly above 40% in U.S. dollar terms.

"Moving into third quarter 2026, we expect our business to be supported by continued strong demand for our leading-edge process technologies, including the steep ramp-up of our 2-nanometer technology," said Wendell Huang, TSMC's chief financial officer, in the company's second-quarter earnings release.

The company is spending like it expects the demand to last, too. Management now plans $60 billion to $64 billion of capital spending in 2026, up from its earlier budget, and it announced an additional $100 billion investment in Arizona to build several more leading-edge chip fabs and advanced packaging plants.

What could go wrong?

The main risk is concentration.

High-performance computing accounted for 66% of TSMC's revenue in the second quarter, tying the company's growth closely to the artificial intelligence (AI) build-out. If the biggest spenders on AI infrastructure pull back, growth could slow quickly.

Of course, margins could give back some ground, too. Gross margin guidance of 65% to 67% for the third quarter sits below the 67.7% the company just posted. If profitability drifts lower from here, earnings could grow more slowly than revenue does -- and it's earnings growth, not revenue growth, that has to average about 14%.

But the prediction can absorb a lot of deceleration. Say revenue growth halves to 20% in 2027, then halves again to 10% in 2028.

Even that path compounds at about 15% a year over those two years, still above the requirement, assuming profit margins hold near current guidance and the price-to-earnings multiple stays put. And it leaves out the rest of 2026, when growth is running at about three times that pace.

The scenario I take more seriously, however, is a market that changes its mind. If investors sour on AI infrastructure spending, they could pay less for each dollar of TSMC's earnings even while those earnings keep growing. A compressing price-to-earnings multiple would likely raise the bar on the business -- possibly well past 14% a year.

Ultimately, though, a business guiding for revenue growth slightly above 40% this year clears a 14% hurdle with plenty of room to spare, even if growth fades hard through 2027 and 2028. I expect Taiwan Semiconductor's market value to top $3 trillion before the end of 2028.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Taiwan Semiconductor Manufacturing. The Motley Fool has a disclosure policy.

Anthropic, the company behind the Claude artificial intelligence (AI) models, plans to publish its initial public offering (IPO) prospectus after Monday's Labor Day holiday, The Information reported late last month. A listing may follow as soon as late September or in October. Amazon (NASDAQ:AMZN) shareholders have a more specific reason than most to open the document when it lands.

On April 20, Anthropic committed to spend "more than $100 billion over the next ten years" with Amazon Web Services (AWS), Amazon's cloud computing segment. That promise is equal to about a fifth of AWS's backlog of contracted work, which reached about $496 billion in June.

In other words, Amazon has already told investors how much one of its biggest cloud customers intends to spend. What no Amazon filing can show is whether the customer's own finances support it. That's what the prospectus is for.

Image source: Getty Images.

The contract is already in Amazon's filings

April's agreement covers up to 5 gigawatts of capacity on Amazon's own silicon -- Graviton processors and Trainium2 through Trainium4 AI chips, with an option on future generations.

Amazon's filings show what a deal that size does to the backlog. AWS's backlog (commitments in customer contracts with original terms longer than one year that haven't yet been recognized as revenue) had grown to about $496 billion by June 30. That was up from about $364 billion in March, and from $195 billion in the middle of 2025 -- growth of 154% year over year, including a $132 billion jump in a single quarter. And the Anthropic deal wasn't alone. The filing also discloses a $100 billion, eight-year expansion of AWS's existing $38 billion commitment from OpenAI, announced a quarter earlier.

Not only is AWS's contracted future far bigger than it was a year ago, but more of it also sits years away from becoming revenue. The weighted-average remaining life of the segment's long-term contracts stretched from 4.0 years to 6.4 years over those 12 months.

One half of the deal is easy to check

Of course, a backlog is signed work, not guaranteed revenue. Amazon says the amount and timing of what it recognizes "will be driven by customer usage and our performance in accordance with contractual obligations."

Amazon's half of that sentence looks strong. In the second quarter of 2026, AWS's revenue rose 37% year over year, to $42.2 billion -- the segment's fastest growth in 18 quarters and a $169 billion annualized pace. Segment operating income rose about 63% year over year to $16.6 billion. And the AI business inside AWS passed a $25 billion annualized revenue pace of its own, growing triple-digit percentages.

The customer's half is the part I can't verify yet. Anthropic is private, and its reported growth is extraordinary. In April, the company said its annualized revenue pace had passed $30 billion, more than triple its level entering the year. And by mid-August, CNBC reported, Anthropic was telling investors the pace had reached $65 billion by the end of July.

Spread evenly, the commitment works out to more than $10 billion a year, or about 6% of AWS's current annual revenue pace. That's arguably affordable if Anthropic's growth holds, and heavy if it doesn't.

And Amazon isn't just supplying the capacity. Its latest quarterly filing shows the company has put another $10 billion into Anthropic this year, with up to $15 billion more available under a financing arrangement tied to compute-delivery milestones. That means Amazon's interest in Anthropic's financial health goes beyond the contract itself.

What does a prospectus settle?

Nearly everything the market knows about Anthropic's finances today is reported, not filed. The company's only filing on record is the confidential draft it submitted to the Securities and Exchange Commission in June.

Its expected market value is a projection. People familiar with the matter told CNBC last month that the company could go public at a valuation of about $2 trillion, about double its private-market value. Its revenue pace is self-reported, and no audited numbers are public.

A prospectus replaces the estimates with audited financial statements: actual revenue, actual profit or losses, and actual cash. Even more useful for Amazon shareholders, it should carry Anthropic's own accounting of its purchase commitments, the other side of the agreements that swelled AWS's backlog.

At about $259 as of this writing, Amazon's stock trades at a forward price-to-earnings ratio of about 24. For a company whose cloud segment just accelerated to 37% growth, I think that's a reasonable price.

But AWS has now disclosed more than $200 billion of multi-year commitments from just two private AI companies, and until those companies file, investors can only judge them by reported figures.

Ultimately, Anthropic's prospectus is the first chance to check one of them. I'll be reading it closely.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Amazon. The Motley Fool has a disclosure policy.

Netflix (NASDAQ:NFLX) raised prices on every one of its U.K. plans in the past few days. The ad-supported standard plan took the biggest jump, moving from £5.99 to £7.99 a month (a third more), while the ad-free standard plan went to £13.99 and premium to £20.99. New members pay the new prices right away, and existing members typically get 30 days' notice before the change reaches their bills.

Shares of the streaming giant fell 5.4% on Friday to $78.25, the same day the increase made headlines.

Price increases are nothing new for this company, though. Netflix has been raising prices for 15 years, in markets all over the world, and its annual revenue has grown every single year through all of them.

But that streak is a low bar. The better measure, I'd argue, is what each increase did to the company's revenue growth rate -- and that record is more interesting than the streak itself.

Image source: Netflix.

The increases are coming faster

Netflix last raised U.K. prices in February 2025, when the ad-supported plan went from £4.99 to £5.99 a month. That makes this the second U.K. increase in about 19 months, and it leaves the ad tier costing 60% more than it did at the start of last year.

Netflix raised U.S. prices in March too, its second increase there in about 14 months, taking the standard plan from $17.99 to $19.99 a month.

Notably, the ad-supported tier (the plan built to catch price-sensitive members) is climbing fastest in both markets.

Revenue has grown through every increase

The worst increase Netflix ever made came in July 2011, when the company split its $9.99 streaming-plus-DVD plan into two $7.99 plans. Management acknowledged in its second-quarter 2011 shareholder letter that the change could be "as much as a 60% increase" for members who wanted to keep both services.

Hundreds of thousands of members canceled. Netflix ended the third quarter of 2011 with about 23.8 million U.S. subscribers, down about 805,000 in three months. And still, revenue rose 48% that year, and it grew another 13% in 2012.

The closest the streak has come to breaking was 2022. Netflix had raised U.S. prices that January, taking the standard plan from $13.99 to $15.49, and revenue for the year grew just 6.5% -- the company's slowest year of growth in at least a decade. A subscriber slump and a strong dollar contributed too. Even then, the top line grew. Growth stayed slow in 2023, then reaccelerated: revenue rose about 16% in both 2024 and 2025, reaching $45.2 billion last year, and 2025 opened with another round of U.S. price increases.

In short, no Netflix price increase has ever been followed by a down year of revenue. Where an increase can show up is in the growth rate, and even the clearest case took more than pricing to get there.

Will the ad tier change the pattern?

What's different this time is where the increase lands. Netflix's advertising business is its fastest-growing revenue line (ad revenue topped $1.5 billion in 2025, up more than 150%, and management is aiming to roughly double it this year), and that business depends on the ad-supported plan attracting members. Raising the plan's price by a third may test how much that audience is willing to pay.

The early evidence from the U.S. increase looks fine. In the shareholder letter accompanying its second-quarter results, Netflix said U.S. and Canada revenue grew 10% year over year, with what it described as only a partial quarter of impact from the March increase. The change, in management's words, "has gone well and as expected."

The companywide trend deserves more caution. Second-quarter revenue growth was 13% year over year, and the forecast for the third quarter is 11.7% -- a decelerating path. Full-year revenue guidance sits at $51.0 billion to $51.4 billion, or 13% to 14% growth, down from nearly 16% in 2025.

Meanwhile, engagement is nearly flat, with members watching only 2% more hours in this year's first half than in last year's. In other words, more members, higher prices, and advertising are carrying the growth, not more hours watched.

Ultimately, I expect the streak to survive this increase too. That kind of pricing power, I think, is rare, and Netflix has proved it over and over.

But the stock's valuation arguably already gives the company credit for it. At about $78, the price-to-earnings ratio is about 20 measured against expected 2027 earnings, a level that arguably assumes the pricing power continues.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Netflix. The Motley Fool has a disclosure policy.

Snowflake (NYSE:SNOW) gave investors a lot to like on Wednesday. The data cloud specialist's fiscal 2027 second-quarter report featured a third straight quarter of accelerating growth, with a bigger push from its artificial intelligence (AI) products.

Shares jumped more than 20% in extended trading on the news. As of this writing, they trade at about $338.

Management now expects about $6.07 billion of product revenue in fiscal 2027 (the year ending Jan. 31, 2027), or 36% year-over-year growth. It also lifted its full-year non-GAAP (adjusted) operating margin outlook to 14.5% from 13.5%. That's the second guidance raise this year. Snowflake opened the year forecasting 27% product revenue growth, raised the number in May, and now sits at 36%.

Put another way, the company keeps outgrowing its own forecasts. And that puts a bigger milestone within view -- $8 billion of product revenue in fiscal 2028, the following year.

Image source: Snowflake.

What would it take?

Snowflake's product revenue totaled $4.47 billion in fiscal 2026, up 29%. This year's guidance implies 36% growth on top of that.

Getting from this year's $6.07 billion to $8 billion the year after requires about 32% growth. In other words, Snowflake could decelerate by roughly four percentage points next year and still clear the mark.

The recent trend makes that bar look manageable. Product revenue came in at $1.49 billion for the fiscal second quarter (ended July 31), up 37% year over year, after 30% growth in the fiscal fourth quarter of 2026 and 34% the following quarter. Chief financial officer Brian Robins said the acceleration, the company's third quarter of it in a row, came from strength in the core data platform along with a meaningful pickup in AI revenue.

Retention is holding, RPO growth is cooling

The most important number behind that view, I'd argue, is Snowflake's net revenue retention rate (what existing customers spent over the past year compared with what the same group spent the year before). It came in at 126% for a second straight quarter, up from the 125% the company posted at the end of fiscal 2026. Remaining performance obligations (RPO), the contracted business Snowflake hasn't yet recognized as revenue, stood at $9.00 billion, up 30% year over year. That growth rate, though, is down from 42% at the end of fiscal 2026 and 38% last quarter.

At 126%, customers already on the platform are growing their spending fast enough to supply most of the roughly 32% the prediction needs. New business has to cover the rest.

The AI products are a newer source of support. CoCo, the company's AI coding agent, surpassed 9,100 accounts, up more than 2,000 in three months. Snowflake doesn't break out AI revenue in dollars, so investors can't size the contribution precisely. But the adoption numbers, and a forecast that keeps rising, suggest the spending is sticking.

A 27% year would fall short

The RPO trend is the one to watch. Snowflake runs a consumption model (customers buy capacity up front rather than paying a flat subscription fee, and draw it down as they use its cloud computing platform), so revenue follows actual usage. And slowing RPO growth can be an early sign of where that usage is headed.

If growth reverts to the 27% pace management originally guided for this year, fiscal 2028 product revenue lands around $7.7 billion, and the prediction misses.

Worth noting: Snowflake's first fiscal 2028 forecast, which should arrive when this year wraps up early next year, will probably start below 32%. After all, this year's guidance started at 27% and has been raised twice since.

A conservative opening forecast wouldn't kill the prediction. A sliding retention rate or another leg down in RPO growth would.

Ultimately, I expect Snowflake to clear the $8 billion mark. If retention holds, existing customers get the company most of the way there, and management has made a habit of guiding low and raising later. Sure, RPO growth is cooling, and a consumption business can decelerate quickly when customers pull back. But the prediction has room for that -- growth can come down four points from the full-year guide and still land above $8 billion.

Whether the growth stock is a buy at this price is a separate question. After the post-earnings jump, Snowflake is worth about $116 billion, or about 19 times this year's guided product revenue. That sales multiple arguably prices in a couple of years of strong execution already.

The prediction, though, is about the business, not the stock. And the business looks on track.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Snowflake. The Motley Fool has a disclosure policy.

Nvidia (NASDAQ:NVDA) closed Friday at $230.36, 2.6% below its 52-week high. Four of its biggest artificial intelligence (AI) chip peers ended the week nowhere near theirs. Advanced Micro Devices (NASDAQ:AMD) sits about 18% below its high, Micron Technology (NASDAQ:MU) about 19%, Broadcom (NASDAQ:AVGO) about 28%, and Marvell Technology (NASDAQ:MRVL) about 32%.

That spread is strange, because one wave of data center spending is paying all five companies. Nvidia expects capital spending by the five biggest hyperscalers (the biggest cloud and internet companies) to land near $800 billion this year and reach $1.3 trillion in 2027.

Do the discounts rank the opportunities? I don't think they do.

Image source: Nvidia.

Nvidia's small discount is earned

Nothing in Nvidia's business has cracked. Revenue in the fiscal second quarter of 2027 (the period ended July 26) was $96.2 billion, up 106% year over year. Growth accelerated from the prior quarter's 85%.

Data center revenue was $89 billion, up 117%. And management guided the fiscal third quarter to $108 billion.

Nvidia's chief financial officer, Colette Kress, told analysts in late August to expect fiscal 2028 revenue growth of about 70% -- a figure that reflects what the company can manufacture, not what customers want.

Shares cost about 15 times analysts' fiscal 2028 earnings estimates. For growth like that, the price still looks reasonable to me.

The two deepest discounts just raised their outlooks

Broadcom reported its fiscal third quarter of 2026 (the period ended Aug. 2) on Wednesday. AI semiconductor revenue reached $16.7 billion, up 221% year over year and 54% from the prior quarter, and management expects $21.7 billion in the current quarter.

Even more, CEO Hock Tan told analysts he is looking to double AI revenue to $115 billion next fiscal year, and in fiscal 2028 to double it again, to $230 billion. Those targets lean on a short list of customers (OpenAI and Anthropic among them) deploying on schedule.

In other words, the group's fastest guided AI growth belongs to its second-deepest discount. Analysts' fiscal 2027 estimates put the stock at about 19 times earnings.

Marvell's discount is the deepest of the four. Its late-August report covered the fiscal second quarter of 2027 (the period ended Aug. 1). Revenue was a record $2.7 billion, up 37% year over year. Data center revenue (now 79% of the total) grew 46%. And CEO Matt Murphy said the company was again raising its revenue outlook for fiscal 2027 and fiscal 2028.

But shares fell about 10% the next day. Management's non-GAAP (adjusted) gross margin forecast implies giving up about a point as lower-margin custom AI chips take a bigger slice of sales. That is a cost-of-winning problem, not a demand problem. Even at a price-to-earnings multiple near 33 on next fiscal year's estimates, a point of gross margin seems like a fair trade for bookings management calls exceptionally robust.

What about AMD and Micron?

AMD's numbers are excellent, too. Revenue rose 50% year over year to $11.5 billion in the second quarter of 2026, and data center revenue more than doubled to $6.7 billion, or 58% of the total. But even 18% below its high, the stock costs about 31 times next year's estimated earnings. That price already counts on a smooth ramp of the company's new Instinct GPUs, just as memory (a big slice of an accelerator's cost) could get more expensive. I'll watch this one from the sidelines.

Micron sits on the other side of that memory bill. Revenue more than quadrupled year over year to $41.5 billion in its fiscal third quarter of 2026 (the period ended May 28), and management's forecast for the fiscal fourth quarter (results due Sept. 30) calls for about $50 billion, with a gross margin around 86%.

Yet Micron's price-to-earnings multiple sits at about 6.5 on next fiscal year's estimates. The market is treating profits like these as a cyclical peak. Memory has always cycled, so I think some of that caution is fair. I'd hold Micron here, without adding to it.

The discounts don't rank the buys

Ultimately, these discounts measure the market's patience, not the companies' earnings paths. Broadcom and Marvell carry the group's two deepest discounts. Both just raised their outlooks anyway.

That mismatch is where I'd put new money: I'd buy Broadcom and Marvell at these prices, and I'd still buy Nvidia near its high. Broadcom and Marvell are priced for problems (deployment schedules at one, a point of gross margin at the other) that look affordable next to the growth they just guided for.

Of course, chip demand moves in cycles, and every discount here could get deeper before it closes. I'd size each position with that in mind.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices, Broadcom, Marvell Technology, Micron Technology, and Nvidia. The Motley Fool has a disclosure policy.

E-commerce platform Shopify (NASDAQ:SHOP) is doing something big companies rarely do: growing faster as it gets bigger. Gross merchandise volume (GMV), the dollar value of everything its merchants sell through the platform, grew 12% in 2022 and has accelerated every year since -- 20%, then 24%, then 29% in 2025. And 2026 is running faster still.

However, the stock hasn't followed the same line. It trades around $148 as of this writing, about 19% off its 52-week high of $182.19.

Where will Shopify stock be in five years? I think it hinges on a few numbers the company reports every quarter -- how fast volume grows, how much of it Shopify keeps, and how much of that turns into cash. It also hinges on how much of all that is already in the price.

Image source: Getty Images.

Faster every year

Shopify's second-quarter report, released in early August, extended the pattern. Revenue climbed 34% year over year to $3.6 billion, the second straight quarter of 34% growth, and GMV rose 32% to $115.6 billion.

For scale, Shopify estimates its merchants handled more than 14% of U.S. e-commerce in 2025.

"GMV growth accelerated on top of last year's already strong Q2 with solid results across all merchant sizes, channels, and geographies," said chief financial officer Jeff Hoffmeister in the second-quarter earnings release.

Of course, a five-year view also has to account for artificial intelligence (AI). If AI shopping tools help merchants sell more, volume per merchant can keep climbing. If they mostly make it easier for anyone to launch a competing storefront, they raise competition among Shopify's merchants instead.

The reported figures don't settle it yet.

Can Shopify keep more of each dollar?

Volume only matters to shareholders after Shopify takes its cut. The company's take rate, or revenue as a share of GMV, came to about 3.1% last quarter. That was a touch higher than a year earlier, as merchants adopted more of its services.

Merchant solutions revenue (payments and the other services merchants pay for as they sell) rose 37% year over year to $2.8 billion, while subscription revenue grew 22% to $802 million. Merchant solutions now make up about 78% of total revenue. Notably, those are lower-margin dollars. Gross margin there runs near 38%, versus about 80% on subscriptions. That mix is why gross profit rose 31% last quarter, trailing revenue's 34% growth -- a gap management expects again in the third quarter.

Meanwhile, cost discipline has more than made up for the cheaper revenue mix. Not only did operating income rise 68% year over year to $488 million, but free cash flow margin (free cash flow as a percent of revenue) also climbed to 18%, after 16% a year earlier and 15% in the prior quarter.

Investors are already paying for years of growth

The trouble is that none of it is a secret. At a market cap near $190 billion, Shopify trades at about 14 times its trailing-12-month sales, about 80 times its free cash flow over the same period, and about 60 times its expected adjusted 2027 earnings.

To justify those multiples of sales and cash flow, Shopify would need years of strong execution. If GMV compounds at 20% annually for five years (slower than today's pace), volume would reach about $1.1 trillion, from about $432 billion over the past year. A take rate near 3.1% turns that into revenue of around $33 billion. And if free cash flow margin climbs from 18% to 25%, Shopify would produce roughly $8 billion of cash in year five.

Today's market cap is still about 23 times that year-five cash flow. Five years of very good execution, in other words, gets a buyer to a valuation that is arguably just reasonable.

A materially higher stock needs more than that. GMV growth could hold near 30% for the full five years, which would put revenue around $50 billion and free cash flow above $12 billion at that same 25% margin. At today's price, that outcome would work out to about 15 times year-five cash flow, cheap enough to leave room for the stock to climb.

Additionally, the take rate may keep inching higher as merchants adopt more services, raising revenue without another dollar of volume. Both are possible. But neither is the kind of assumption I'd want my returns to depend on.

Ultimately, I expect Shopify to be a much bigger business in five years. But I don't expect the stock to climb nearly as fast as the business grows, because so much of that growth is already reflected in the price.

I'm not buying the stock at today's price. If shares pull back meaningfully, or a few more quarters show the take rate and free cash flow margin climbing together, I'd take another look.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Shopify. The Motley Fool has a disclosure policy.

SoundHound AI (NASDAQ:SOUN) is becoming a much bigger company. LivePerson (NASDAQ:LPSN) shareholders approved the sale of their company to the voice artificial intelligence (AI) specialist on Sept. 2, and the mostly stock deal closed on Friday, Sept. 4.

SoundHound's own second-quarter revenue grew 45% year over year to a record $61.9 million, and management raised the low end of its full-year outlook, which now calls for $230 million to $260 million of revenue in 2026.

Stack a large acquired business on top of that growth, and the top line could reach half a billion dollars surprisingly soon. The losses are moving much more slowly. Here's my prediction: SoundHound passes $500 million in annual revenue before it earns a sustained profit.

Image source: The Motley Fool.

Record revenue, a raised outlook

SoundHound's growth record is short but steep. Revenue nearly doubled in 2025, reaching $168.9 million. The first quarter of 2026 came in at $44.2 million, up 52% year over year, and the second quarter's growth was 45%, with revenue up 40% from the first quarter alone. The growth rate is decelerating as the numbers get bigger. But the dollars keep setting records.

This year's guidance of $230 million to $260 million implies 36% to 54% growth. Notably, the range excludes LivePerson. Management plans to update its guidance after the deal closes, likely before the end of the year.

From the $245 million midpoint, the compounding works fast. Another year of 45% growth would put 2027 revenue near $355 million, and a repeat in 2028 would land the company at about $515 million. Even if growth cooled to 30%, revenue would cross $500 million in 2029.

LivePerson pulls the timeline into 2028

The acquisition, a mostly stock deal with an enterprise value of about $250 million, brings a business nearly as large as SoundHound itself. LivePerson expects $195 million to $207 million of revenue this year.

That revenue is shrinking, though. Guidance calls for a 15% to 20% decline in 2026, and revenue retention among LivePerson's enterprise and mid-market customers fell to 78% last year.

SoundHound's own combined-company forecast reflects the erosion. Management projects at least $350 million to $400 million of revenue in 2027 (still below what the two would generate this year combined).

Still, from the top of that range, the combined company needs only about 25% growth in 2028 to pass $500 million -- a fraction of the rate SoundHound is delivering on its own. Management has said the combined business is expected to reach up to $500 million on its existing customers alone. I think revenue crosses in 2028, with 2029 as the slow case.

When do the profits show up?

Much later, on the current trend.

SoundHound's non-GAAP (adjusted) EBITDA loss narrowed to $9.6 million in the second quarter from $14.3 million a year earlier, a 33% improvement. (Adjusted EBITDA is earnings before interest, taxes, depreciation, and amortization, excluding items like stock-based compensation.)

The loss line is narrowing, but not in a straight line. The first quarter's adjusted loss widened year over year, to $26.7 million from $22.2 million.

The bigger problem is the gap between that measure and the company's full losses. SoundHound's second-quarter net loss was $42.8 million (about 69% of revenue), an improvement from $74.7 million a year earlier. Stock-based compensation plus depreciation and amortization alone separate the two figures by more than $30 million a quarter, and quarterly mark-to-market swings on acquisition-related liabilities add noise on top.

Of course, those swings can point the other way, too. SoundHound reported $40.1 million of net income in the fourth quarter of 2025, thanks to an $85 million non-cash mark-to-market gain. But that isn't the kind of profit this prediction is about.

And LivePerson won't speed things up. It lost $72.6 million last quarter, including a $51.8 million goodwill impairment, and the merger adds fresh acquisition accounting on top.

Ultimately, this isn't a close race. On the current pace, revenue passes $500 million in 2028. A sustained profit (money earned on operations, not a one-quarter accounting gain) looks unlikely before 2029 at the earliest. And even that requires the second quarter's operating leverage to hold through a big integration.

For the stock, that order matters. Shares trade at about $6.75 as of this writing, down about 70% from their 52-week high of $22.17, yet still about 12 times revenue, based on the midpoint of this year's guidance. That price already assumes the growth keeps coming, and I think it will.

But anyone buying the growth stock should expect red ink for years while the top line does the work. I'd avoid buying shares here until the loss line proves it can keep narrowing through the integration.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends SoundHound AI. The Motley Fool has a disclosure policy.

Michael Burry is going after Palantir Technologies (NASDAQ:PLTR) again. The investor of The Big Short fame laid out an accounting case against the artificial intelligence (AI) software specialist in a February post titled "Palantir: An Accounting."

This week he pressed the case again, arguing that Palantir's financial profile looks more like a consulting firm's than a software platform's. He says a company valued around $420 billion today could eventually be worth less than $100 billion.

He has had money behind the view. His Scion Asset Management disclosed put options on 5 million Palantir shares last fall. That was its final filing before he wound the fund down, and he told subscribers in April that he still holds Palantir puts.

Palantir stock, meanwhile, fell almost 6% on Wednesday, jumped 7.7% on Thursday (the same day the company and consulting giant PwC announced an expanded enterprise AI alliance), and traded near $174 as of this writing, down more than 4%.

Burry's case rests on numbers in Palantir's own filings, so that is where I checked it.

Image source: Getty Images.

The receivables are growing faster than sales

In nine of the last 12 quarters, Burry wrote in February, Palantir's accounts receivable (the money customers owe for work already billed) grew faster than its revenue. He argued that a pattern like that can point to channel stuffing, aggressive revenue recognition, or payment terms stretched to win deals.

The newest numbers don't break the pattern. Receivables stood at $1.49 billion at the end of June, up from $1.04 billion at the end of 2025 -- 43% growth in six months, against 38% growth in quarterly revenue over the same stretch. And the build is speeding up. It cut $434 million from operating cash flow in the first half, versus $164 million a year earlier.

Notably, Palantir itself offers an explanation. The company says in its June-quarter filing that it has been shifting away from collecting several years of payments up front and toward billing annually or even in arrears, meaning after the work is done.

That is a legitimate business choice. It is also exactly how a consulting firm gets paid.

One customer owes about $400 million

The filing also discloses that a single customer, identified only as Customer I, represented 27% of receivables at the end of June, up from 25% at the end of 2025. That works out to roughly $400 million owed by one customer.

However, no customer accounted for more than 10% of revenue in the first half -- about $357 million at most. In other words, one customer appears to owe Palantir more than it could have recognized in revenue from any customer all half.

That isn't proof of anything improper. A large government-related account may simply pay slowly, or billing may run ahead of schedule. But it is an unusual shape for a software company, with revenue spread across many customers and collection risk concentrated in one.

Does Palantir collect like a consultant?

Burry's sharpest comparison is a ratio. A subscription software company typically bills customers up front, so cash arrives before the revenue does and piles up on the balance sheet as deferred revenue. A consulting firm earns the revenue first and collects later. Salesforce, for example, carried $18.8 billion of unearned revenue in its most recent quarter -- more than one and a half times its $11.3 billion of quarterly revenue. At Accenture (NYSE:ACN), the consulting giant Burry measures Palantir against, deferred revenue of about $7.6 billion amounts to around 40% of quarterly revenue.

Palantir's deferred revenue of about $613 million comes to 32% of its $1.94 billion in second-quarter revenue, effectively the ratio Burry cites. Add the $453 million of customer deposits Palantir groups with it as contract liabilities, and the figure is still only about 55%. On either basis, Palantir collects like Accenture, not like Salesforce.

Of course, the rest of the filing hardly describes a company in trouble. Revenue grew 93% year over year in the second quarter, and operating cash flow more than doubled in the first half, to $2.1 billion.

Customers are paying. They are just paying later, and in a more concentrated way, than software investors might assume.

And that, I think, is where Burry's argument lands hardest. It isn't an accusation of fraud. Every number he cites is disclosed. It is a reclassification argument: If Palantir earns its revenue the way a consultant does, the stock may not deserve a software valuation.

At about 150 times earnings, shares have a long way to fall if the market ever agrees with him. His sub-$100 billion scenario is more than 75% below today's value. I was on the sidelines at this valuation before Burry wrote a word, and the second-quarter filing doesn't move me off them.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Accenture Plc, Palantir Technologies, and Salesforce. The Motley Fool recommends the following options: long January 2028 $260 calls on Accenture Plc and short January 2028 $280 calls on Accenture Plc. The Motley Fool has a disclosure policy.

Berkshire Hathaway (NYSE:BRKA)(NYSE:BRKB) CEO Greg Abel joined CNBC's "Squawk Box" from Tokyo on Wednesday, and the most detailed answers covered the business he knows best. Abel ran Berkshire's energy operation for years before succeeding Warren Buffett as CEO in January.

So when the conversation turned to artificial intelligence (AI) data centers and the power they need, he had specifics.

His message came with a condition attached. Berkshire wants the hyperscalers' business (the giant cloud companies building those data centers), but only on terms that leave its utilities' other customers unharmed. That bar matters because the energy business is one of Berkshire's better growers this year, earning about $2 billion in the first half. How much of that growth can Berkshire capture on those terms?

Image source: Getty Images.

Abel's conditions

Speaking with CNBC's Becky Quick, Abel framed Berkshire's role in the build-out as supplying power and electricity, not owning the data centers themselves. He said the company has operated by the same basic principles from the start and has shared them with the hyperscalers, governors, and state regulators.

"[W]e are interested in serving these hyperscalers ... if there was no impact to the rates of our other customers," Abel said. "And in fact, we've pretty much taken the approach. There has to be a net benefit to our customers."

The other conditions center on the communities. They have to understand a project's impact on water, which Abel said has become much more manageable as the industry limits water use.

And they have to want the facility there in the first place. In Abel's view, a data center has to be welcomed by its host community.

Iowa already shows what a yes looks like

Iowa, where Berkshire's MidAmerican Energy utility operates, is the place to look. Notably, data centers accounted for about 8% of the utility's load there last year, Abel said, and more is on the horizon.

Berkshire's utility results are climbing alongside that load. After-tax earnings at the company's U.S. utilities rose 38% in the second quarter from a year earlier, to $597 million, and 11% year over year in the first half of 2026. Retail volumes across those utilities were up about 3% through June, with MidAmerican, the Iowa utility, leading at 6%. Electric utility margin (revenue minus energy costs) expanded 8% year over year in the quarter, helped by higher retail volumes. The whole segment also accelerated as the year went on. After roughly flat earnings in the first quarter, Berkshire Hathaway Energy grew earnings 27% year over year in the second quarter, to $891 million.

That growth takes capital, and Berkshire is spending it. Of the company's $10.6 billion in first-half capital expenditures, $6.7 billion was attributable to the energy business and the BNSF railroad. The two units forecast about $8.6 billion more over the remainder of 2026.

For a regulated utility, that spending is the growth. After all, rates are largely set to recover costs plus a return on invested capital, so every data-center project that clears Abel's bar grows the base the company earns on.

Can the growth case survive the pushback?

Abel did flag one thing that could slow this down.

"There is a lot more pushback in the communities across the U.S.," he said.

He has long held the view that energy would be the constraint on the build-out, and he described the community reaction as a challenge on top of it.

So far, though, the pushback hasn't cost Berkshire a single energy-infrastructure site -- none has been rejected to date, Abel said, and construction continues.

And I'd argue the conditions are exactly why. A utility that can tell its regulators the hyperscalers won't be subsidized by everyone else's power bills is a harder target for that resistance. In Iowa, Abel added, the property taxes these projects pay are a substantial source of funding for schools and local services.

Of course, a moratorium in the wrong state or a community that says no could still stall a project regardless of the terms. And the utilities won't grow 38% every quarter -- the second quarter's jump got help from production tax credits, and comparisons may get tougher from here.

Ultimately, though, I think Abel's rate condition is less a limit on the growth case than the substance of it. Berkshire is qualifying load it can serve for decades in states that want it there. And it is putting billions of dollars behind projects regulators have little reason to fight.

Data centers were about 8% of Berkshire's Iowa load last year. On Abel's terms, that share can keep climbing.

Daniel Sparks and his clients have positions in Berkshire Hathaway. The Motley Fool has positions in and recommends Berkshire Hathaway. The Motley Fool has a disclosure policy.

By most measures, Cerebras Systems (NASDAQ:CBRS) delivered an outstanding second quarter.

The artificial intelligence (AI) computing specialist grew its non-GAAP (adjusted) revenue 103% year over year to $209.9 million. Its inference cloud business nearly quadrupled, and management raised its full-year outlook to a range of $880 million to $890 million in adjusted revenue.

But the most important number in the mid-August update wasn't on the income statement at all. Cerebras ended June with $25.4 billion in remaining performance obligations, the backlog of contracted work it hasn't yet delivered or recognized as revenue. That's nearly 29 times the revenue management expects for all of 2026, a figure lifted by data center costs passed through to OpenAI.

A figure that large deserves scrutiny. The company's own filings say where to look: at a single agreement with OpenAI.

Image source: Getty Images.

The backlog arrived almost all at once

In December 2025, Cerebras signed a master relationship agreement with the ChatGPT maker under which OpenAI committed to purchase 750 megawatts of computing capacity for AI inference -- a deal Cerebras has valued at more than $20 billion. OpenAI also holds an option to buy an additional 1.25 gigawatts of capacity by the end of 2030.

Remaining performance obligations were $24.6 billion at the close of 2025, then edged up to $25.0 billion in March and $25.4 billion in June. The balance grew only about 3% over the first half of 2026. Nearly all of it was on the books before 2026 began. And Cerebras says in its latest quarterly filing that a significant amount of the balance is attributable to its obligations under the OpenAI agreement.

Cerebras recognized $56.8 million of revenue under the arrangement in the second quarter, or about 32% of the company's $180.1 million in revenue under generally accepted accounting principles (GAAP), which grew 74% year over year.

When does the backlog become revenue?

The backlog converts slowly, by design. Cerebras expects to recognize only about 22% of the $25.4 billion (about $5.6 billion) over the 24 months ending June 30, 2028.

Another 43% should arrive between months 25 and 48, with the rest coming later. Of course, the timing can shift at the customer's request.

It's worth noting, though, that the near-term share has moved up. At the close of 2025, Cerebras expected about 15% of the balance to convert in the 24 months through 2027. The latest figure is 22%, though it covers a window ending six months later.

The conversion takes years partly because Cerebras is still building the thing it has sold. Capacity for OpenAI deploys in stages from 2026 through 2028. Cerebras says more than 600 megawatts of data center capacity is live or under contract for delivery by the end of 2027, with manufacturing capacity set to grow more than tenfold in 2026. And just this week, Cerebras announced a new 165-megawatt data center in Finland.

OpenAI is even helping to finance the build-out, advancing Cerebras a $1 billion working capital loan in January.

Concentration isn't new here

In 2025, Mohamed bin Zayed University of Artificial Intelligence accounted for 62% of the company's revenue, and Group 42 accounted for another 24%. Those figures are shares of last year's revenue, not of the backlog.

But the pattern held in the second quarter, when three customers each accounted for at least 10% of revenue, or 76% of it between them. The company doesn't say exactly how much of the $25.4 billion sits with OpenAI. Either way, a short list of buyers is doing most of the buying.

That matters because of the stock's valuation. With shares around $215 as of this writing (down about 44% from their 52-week high of $386.34), the whole company is valued near $51 billion -- nearly 58 times the adjusted revenue management expects this year, for a company still posting operating losses. Even if revenue more than triples in 2027, as management plans, the stock would trade at about 19 times those expected sales.

What, then, is the backlog worth to a shareholder? A lot, I think -- just not everything the headline number implies.

The $25.4 billion includes a customer's multiyear commitment, not revenue in hand. And most of it is scheduled to convert after mid-2028, by a company that must build enormous capacity on time, much of it for one buyer whose needs could change.

The business itself is executing well. Adjusted gross margin improved about nine percentage points from a year ago, and Cerebras holds about $8.6 billion in cash and investments after May's initial public offering.

Ultimately, the backlog is evidence of extraordinary demand and arguably the best reason to keep watching Cerebras closely. But I'd want to see the OpenAI revenue step up for a few more quarters before paying today's price.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

When Advanced Micro Devices (NASDAQ:AMD) announced its Anthropic partnership in late July, two commitments stood out. Anthropic agreed to deploy up to 2 gigawatts of AMD Instinct MI450 series graphics processing units (GPUs), with deployment of the first gigawatt set to begin in the first half of 2027. And AMD committed to invest up to $5 billion in the artificial intelligence (AI) company behind the Claude models.

The second commitment is about to get easier to measure. Anthropic plans to publish its initial public offering (IPO) prospectus after the Labor Day holiday on Monday, with a listing as soon as late September or early October, The Information reported late last month.

What does AMD hold today, then? Not a stake, at least not yet.

Image source: AMD.

Conditions attached

AMD's press release put it carefully: The company "has committed to make a strategic equity investment of up to $5 billion in Anthropic in the future."

AMD's early August quarterly filing added structure. It describes investment commitments of up to $5 billion entered after the quarter ended, "subject to certain contingencies," with the money expected to go out through fiscal year 2028.

Neither company has said what the contingencies are. And no valuation for the investment has been disclosed.

That shape has become standard among Anthropic's backers. Alphabet agreed in April to invest up to $40 billion -- $10 billion immediately, the remaining $30 billion contingent on performance milestones.

For scale, AMD held $1.7 billion of investments in private companies at the end of the second quarter. This one commitment could grow to nearly triple that.

What would a listing change?

Anthropic itself has confirmed very little. The only filing on record is a confidential draft registration statement submitted in June.

However, the reported figures are staggering. CNBC has reported that Anthropic is valued at close to $1 trillion in the private markets, and that investors project it could float at about a $2 trillion valuation. The growth underneath, I think, explains the excitement. Anthropic's annualized revenue run rate (a full-year projection of its recent revenue pace) topped $30 billion in April and passed $65 billion by the end of July. The company has reportedly raised at least $130 billion, and its offering is expected to surpass the June IPO of SpaceX, which raised about $86 billion, the largest on record.

Every one of those figures is reported, not filed. And at the reported valuations, AMD's up-to-$5 billion would buy no more than about half of 1% of the company.

Still, a listing would give whatever stake AMD may eventually hold a daily price that flows straight into its reported results. The company ended the second quarter with $425 million of net unrealized gains on marketable equity securities, mostly from holdings that went public during the quarter.

AMD is helping finance a customer

The part I'd watch most closely isn't the stake at all. AMD has committed money to a company that agreed to deploy its chips. AMD's OpenAI arrangement runs in the opposite direction. That deal handed OpenAI a warrant for up to 160 million AMD shares, vesting as deployment and stock-price milestones are hit.

Showing what those deals feed, AMD's data center segment revenue more than doubled year over year to $6.7 billion in the second quarter, or 58% of record companywide revenue of $11.5 billion, up 50% year over year. Management guided third-quarter revenue to about $13 billion, up about 41%. That guided rate marks a deceleration, at a much larger scale.

When a customer AMD helps finance commits to up to 2 gigawatts of deployments, some of the dollars moving through the system could be AMD's own. In effect, a slice of the industry's demand could end up self-financed.

That, I'd argue, is the strongest reason the IPO matters to AMD shareholders. An offering that surpasses SpaceX's could pay for a chunk of the buildout with public investors' money instead of suppliers' commitments. Even more, a public Anthropic would have to show, quarter after quarter, how much revenue it's actually producing.

In short, the two commitments aren't equal. The up-to-$5 billion investment is conditional, unpriced, and small next to Anthropic's reported valuations. The deployments are what can become revenue, and they aren't set to begin until 2027.

Meanwhile, AMD shares trade near $474 as of this writing (about 19% below their 52-week high), at about 30 times next year's expected earnings -- a price with a lot of chip demand already baked in. I think the Anthropic deal makes that demand more likely to show up on schedule. But what AMD holds from it today is still a promise.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices and Alphabet. The Motley Fool has a disclosure policy.

Shares of NuScale Power (NYSE:SMR) trade around $9.63 as of this writing, down about 83% from their $57.42 52-week high.

But the demand that sent nuclear stocks soaring in the first place hasn't reversed. Constellation Energy (NASDAQ:CEG), for instance, reported an additional 920 megawatts of long-term power purchase agreements alongside its second-quarter results.

In other words, the crash and the demand can both be true at once, because these companies do very different jobs. NuScale designs reactors, Cameco (NYSE:CCJ) sells the fuel, and Constellation already sells the power. A $2,000 investment split across the three buys three different claims on the same electricity demand.

Image source: The Motley Fool.

1. NuScale: approved designs, almost no revenue

NuScale's reactor is the first and only small modular reactor design certified by the U.S. Nuclear Regulatory Commission. Then in May 2025, the agency approved the company's uprated US460 plant design -- six modules of 77 megawatts each, or 462 megawatts in total.

What the approvals haven't produced yet is a paying customer base. Revenue went from $8.1 million in last year's second quarter to $565,000 in this year's first quarter to $75,000 in the second. The engineering work behind the older figures, on a project in Romania, wrapped up in late 2025, and nothing has replaced it. Meanwhile, the company's net loss widened to $50.1 million in the quarter, versus $37.6 million in the same period of 2025.

Buying the growth stock today is mostly a bet that the approvals turn into orders. The company does have a $1.9 billion war chest of cash and investments, which buys it time. And its partner ENTRA1 Energy is in discussions with the Tennessee Valley Authority about what management calls potentially the largest nuclear power deployment program in U.S. history -- though a definitive agreement hasn't been signed.

2. Cameco: the pounds are already sold

Cameco mines and refines uranium, and much of what it will deliver for years to come is already sold. The company has contracts in place for average annual deliveries of more than 28 million pounds of uranium over the next five years. Commitments run above that average from 2026 through 2028, and below it in 2029 and 2030. Notably, Cameco expects its own share of this year's production to total 19.5 million to 21.5 million pounds.

The company also said the long-term uranium price strengthened further, supported by contracting activity in the year's first half as customers increasingly focus on security of supply.

The second quarter itself was quieter. Cameco delivered 7.1 million pounds of uranium, and earnings fell from a year earlier, mostly because earnings from its stake in nuclear services company Westinghouse dropped after an unusually strong year-ago quarter.

Cameco's stock isn't the beaten-down one here. Shares trade about 25% off their 52-week high as of this writing, a far smaller discount than NuScale's. An investor buying today is paying for deliveries already under contract, not for a turnaround.

3. Constellation: the only one selling power today

Constellation is the largest nuclear energy company in the U.S. Its overall fleet has 55 gigawatts of generating capacity, a figure that includes Calpine, the natural gas and geothermal power producer it bought in January. Its nuclear fleet produced 44,160 gigawatt-hours of electricity in the second quarter.

Profits are rising, too. Second-quarter non-GAAP (adjusted) operating earnings climbed 34% year over year to $2.55 per share, and management now expects $11.50 to $12.50 in full-year adjusted earnings per share, a raised outlook. The new power purchase agreements run 15 to 20 years, and they include a 176-megawatt deal with Walmart supporting a capacity expansion at its Dresden plant in Illinois.

The stock isn't cheap, however. At about $292 per share, and using the midpoint of management's raised guidance, the stock's price-to-earnings multiple sits near 24.

But unlike NuScale, Constellation's valuation is measured against profits that already exist.

How I'd split the $2,000

Ultimately, all three stocks are claims on the same electricity demand. They just sit at different distances from the cash.

Constellation collects it today. Cameco has years of deliveries under contract. NuScale is still waiting for its first real order.

Sure, NuScale likely offers the most upside if orders finally land. But it's also the only one that could still be years away from meaningful revenue, and its losses are widening in the meantime.

So if I were putting $2,000 into nuclear power today, I think the split should favor proof: about half in Constellation, most of the rest in Cameco, and the smallest slice in NuScale. That way, most of the money sits with companies already getting paid.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Cameco, Constellation Energy, and Walmart. The Motley Fool recommends NuScale Power. The Motley Fool has a disclosure policy.

One quick way to test whether a boom has peaked is to look at what buyers are still willing to pay. A 36-gigabyte chip of HBM3E, the high-bandwidth memory (HBM) that feeds artificial intelligence (AI) processors, sells for about $2,100 on the spot market, where chips trade for immediate delivery.

Long-term supply agreements price the same product at roughly 500,000 to 700,000 won (about $370 to $510), according to reporting this week from the Seoul Economic Daily. In other words, buyers who need memory today are paying four to five times the contract price to get it.

Memory is a famously cyclical business. But cycles turn when supply catches up with demand, and prices are where that shows up first. And from the spot market to Korea's export data, the prices are pointing the other way.

No company has more riding on this than memory specialist SK Hynix (NASDAQ:SKHY). The growth stock trades around $170 as of this writing, about 13% below its 52-week high.

Image source: The Motley Fool.

A four-to-five-times premium

Most of the industry's output is already spoken for. SK Hynix said in its late-July second-quarter report that it has finalized long-term agreements with about 10 customers, including key strategic partners, with discussions ongoing with other major clients. Those deals lock up much of the tech company's output for years at negotiated prices.

After all, a buyer who believed memory prices were about to roll over wouldn't pay four to five times the contract rate for chips today. They'd wait. That spot buyers keep paying up instead says the chips simply aren't available at anything close to contract prices.

Sure, the spot market is a thin slice of overall memory volume, and thin markets can overshoot. But when memory cycles have rolled over in the past, spot prices have tended to crack first, sliding below contract levels as buyers step back. A premium this wide is arguably the opposite signal.

Korea is shipping fewer chips for more money

Korea's export data tells the same story from a different angle. In May, the country exported about 682 million DRAM chips worth $11.4 billion, according to the same report.

By July, volume had fallen about 13% to about 592 million units, while the value of those shipments rose about 19% to $13.6 billion. The average unit price jumped about 37% in two months, from $16.76 to $22.90.

At a supply driven peak, new supply would flood in, volume would climb, and unit prices would flatten or fall as competition returned.

Instead, producers are shipping fewer chips and collecting more money for them, which is what I'd expect as production lines steer toward pricier AI memory and the remaining supply gets rationed by price.

That combination of falling volume and rising value isn't what a market coming back into balance looks like.

Has the memory cycle peaked?

The boom is showing up in SK Hynix's own results, too. Capturing how fast this cycle is still compounding, SK Hynix's second-quarter revenue came in at 79.3 trillion won (about $58 billion), up 257% year over year. Operating profit did even better, surging 557% year over year to a record 60.5 trillion won (about $45 billion), a 76% operating margin. And the trajectory is still climbing, not rolling over: operating profit went from 9.2 trillion won a year ago to 37.6 trillion won in the first quarter to 60.5 trillion won in the second, a 61% jump in one quarter.

The pricing pressure hasn't let up since, either. Market researcher TrendForce expects conventional DRAM contract prices to rise 13% to 18% in the third quarter from the second, with NAND flash prices up 10% to 15%.

Of course, this cycle will end the way memory cycles usually end -- with too much supply. Prices like today's eventually invite a wave of new capital expenditures across the industry. And a slowdown in AI spending could turn the data quickly.

SK Hynix carries a company-specific risk, too. Samsung Electronics (OTC:SSNLF) is ramping up rival HBM4 shipments, and a stronger second supplier could cut into SK Hynix's share of the boom even if memory prices stay high. Counterpoint Research reported Thursday that SK Hynix's HBM revenue share fell from 58% to 50% in the second quarter, while Samsung's rose from 21% to 33%.

Memory investors should expect volatility along the way. But a peak should be visible somewhere in the numbers -- a shrinking spot premium, unit prices flattening, export volumes recovering. None of that is happening yet, so I think the peak calls are early.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

Quantum computing company IonQ (NYSE:IONQ) hosts an investor day on Tuesday, Sept. 8. And I expect the event to produce something the company has never put in its guidance: a revenue forecast that reaches beyond the current year.

Second-quarter revenue rose 287% year over year to about $80 million, the company's fifth consecutive quarter of record results, up from about $65 million in the first quarter of this year. And management has raised its 2026 revenue guidance twice, from an initial range of $225 million to $245 million in February to $280 million to $290 million today.

But every one of those figures stops at Dec. 31. Meanwhile, shares trade near $39 as of this writing, giving IonQ a market value of about $15.5 billion -- about 54 times the midpoint of this year's guided sales.

In other words, the years actually supporting the price are years management has never guided to. My prediction is that this changes on Tuesday.

Image source: The Motley Fool.

Good guidance, short horizon

IonQ's record as a forecaster is strong. The company delivered $130 million of revenue in 2025, up 202% year over year and 20% above the midpoint of its own guidance. That result, the company says, made it the first public quantum company to top $100 million in annual GAAP revenue.

What management has never done is put a year beyond the current one into its guidance. Even at the second-quarter report in early August, with the company's $1.8 billion acquisition of chipmaker SkyWater Technology closed just days earlier, management went no further than the current year.

"Because we have operated as a combined company for less than a week, we need to integrate our operations before providing combined company revenue or EBITDA guidance," said Inder Singh, IonQ's chief operating officer and chief financial officer, on the earnings call.

Singh did sketch a timeline, to be fair. Investors, he suggested, could get "color at Analyst Day perhaps, but certainly at the close of quarter."

The full combined-company numbers, in short, may wait for the third-quarter report.

Why Tuesday?

The SkyWater deal practically demands a longer view. SkyWater generated $442 million of revenue in fiscal 2025, more than three times what IonQ itself produced that year. A multiyear frame lets management describe the much larger combined business on its own terms.

The technical roadmap already reaches years out, too. Not only has IonQ committed publicly to 800 logical qubits in 2027, up from a 2026 milestone of just 12, but it has also promised 2 million physical qubits supporting 80,000 logical qubits by 2030.

A company willing to publish engineering milestones four years ahead, while never guiding to what those machines could earn, has left an obvious gap. An investor day seems like the natural place to fill it.

The market has already picked a number

IonQ's market value sits near $15.5 billion. At 15 times sales (a sales multiple usually reserved for the fastest-growing software companies), supporting today's price takes about $1 billion of annual revenue. IonQ's 2026 guidance midpoint is $285 million. Double that in 2027 and again in 2028, and revenue reaches about $1.1 billion -- at which point the stock, at today's value, would still trade at roughly 14 times sales. Put another way, a multiyear range that satisfies this market has to promise the doubling continues.

And profits can't fill the gap in the meantime. IonQ's non-GAAP (adjusted) EBITDA loss, a rough measure of underlying operating losses, more than tripled year over year in the second quarter, widening to about $120 million -- larger than the quarter's entire revenue.

Until that swings, the top line is what investors have to go on.

Ultimately, I expect a formal IonQ revenue number beyond 2026 to land on Tuesday, likely a multiyear frame for the combined company instead of a single 2027 figure. Of course, the honest risk to that call is Singh's own timeline. But management has spent years beating its own forecasts, and its published roadmap already runs to 2030.

A multiyear number would be welcome. It would hand investors a guidance yardstick beyond the current year, something a $15.5 billion valuation arguably should have had all along.

It wouldn't make me a buyer, though. At about 54 times this year's guided sales, the growth stock's price already assumes something spectacular. I'd stay on the sidelines for now.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends IonQ. The Motley Fool has a disclosure policy.

At an invite-only event in downtown Austin on Thursday, Tesla (NASDAQ:TSLA) put the Cybercab into service. The two-seat robotaxi has no steering wheel and no pedals. And riders in the city can now hail one through the company's Robotaxi app, joining the driverless Model Ys that have carried paying passengers there since June 2025.

Shares rose 5.7% on Thursday ahead of the event, closing at about $376.

CEO Elon Musk spent the run-up teasing the launch, pinning a post on X that read "A storm of Cybercabs." The storm, for now, is modest: Texas has authorized 45 Cybercabs for driverless operation statewide.

Six weeks before Thursday's launch, though, Tesla itself removed the Cybercab from the list of products it expects to reach volume production in 2026. The event settled where riders can find a Cybercab. It didn't settle when Tesla can build them at a rate that matters.

Image source: The Motley Fool.

A launch, not a ramp

Tesla first showed the Cybercab as a concept in October 2024, a two-seater with butterfly doors and no driver controls. For months, the company tested versions of the vehicle with human drivers and traditional controls in several U.S. markets. What arrived Thursday is the production version. Tesla began producing the vehicle earlier this year, according to its latest quarterly filing.

Sure, the event itself was small, with five winners selected at random through a Robotaxi rider sweepstakes. But the deployment behind it is commercial. A paying customer in Austin can now hail a vehicle that was a concept on a stage almost two years ago.

What Tesla didn't attach to the launch was a production rate or a new volume timeline.

Tesla pulled the volume promise in July

Tesla's first-quarter update told investors that "Cybercab, Tesla Semi and Megapack 3 are on schedule for volume production starting in 2026." The second-quarter update, published July 22, no longer makes that promise. The Semi and Megapack 3 are still slated to start production this year, though the promise is now production, not volume production. The Cybercab is out of the sentence altogether. And the update stopped promising volume production of the Optimus robot, too.

Tesla didn't leave the reason to guesswork. The July letter calls battery pack capacity expansion "the main limiting factor to near-term vehicle production volume increase." And it says the company is increasing output of its 4680 battery cells to support production ramps of the Cybercab, the Tesla Semi, and the Model Y.

Notably, the factory itself isn't the constraint.

Tesla's installed-capacity table shows the Cybercab line at Gigafactory Texas is built to make more than 125,000 vehicles a year and is already producing. The 45 Cybercabs registered in Texas so far are a tiny fraction of that.

I think the letter matters more than the launch, because it names the thing that has to change before the Cybercab can become a meaningful business.

What turns 45 cars into revenue?

Tesla has already told investors where this is supposed to go. In the first-quarter update, the company said it expects the Cybercab "will begin to replace the existing Model Y fleet and will be the largest volume vehicle in the fleet over time."

That ambition sits a long way from 45 vehicles. And the path to it runs through the battery constraint Tesla named in July.

The spending to get there is well underway, however. Tesla raised its 2026 capital-spending plan to more than $25 billion, nearly triple its recent annual levels. Second-quarter capital expenditures were more than double the year-ago figure, and the quarter's free cash flow was negative.

Put another way, the spending is accelerating ahead of the revenue it's meant to produce -- and that revenue still waits on batteries.

The order of events matters because of what the stock costs. Tesla carries a forward price-to-earnings ratio of about 155, based on what the company is expected to earn next year -- a price that arguably assumes the storm of Cybercabs arrives without much delay.

Ultimately, Thursday's event did what a launch can do. It proved the product, and it put paying riders in the seats. What it couldn't do is move the constraint Tesla named in July.

For now, the pace of the Cybercab business likely rests on battery output. That is the number I'd watch.

Daniel Sparks has clients with positions in Tesla. The Motley Fool has positions in and recommends Tesla. The Motley Fool has a disclosure policy.

SpaceX (NASDAQ:SPCX) held its first earnings call as a publicly traded company on Aug. 4, and CEO Elon Musk used it to place the company's largest capital outlay, artificial intelligence (AI) computing, in the hands of a single supplier.

Musk said, "going forward, we've decided to build exclusively on Nvidia because we think the Vera Rubin architecture is the best architecture. We think it's the best AI computer, and we greatly value our close cooperation and partnership on many levels with Nvidia. So, we're exclusive to Nvidia."

Nvidia (NASDAQ:NVDA) shares closed up 3.4% the next day. Advanced Micro Devices, the closest alternative supplier of graphics processing units (GPUs) for AI, closed down 7% after reporting its own quarterly results that same afternoon.

Nvidia is also a shareholder, with about $21 billion in SpaceX shares at the end of June.

For SpaceX shareholders, the most interesting figure is the bill. What gets locked in by building it with a single supplier?

Image source: The White House.

How much computing capacity does Musk promise?

SpaceX's capital expenditures were $18.4 billion in the second quarter, and $15.8 billion of that was allocated to AI computing infrastructure. The AI figure was $7.7 billion in the first quarter and $749 million a year earlier. In other words, the computing line item grew more than 20 times year over year and now absorbs 86 cents of every capital dollar.

CFO Bret Johnsen told analysts to expect the next two quarters to be "very similar to the current quarter" in terms of capital expenditures, probably about $37 billion more this year.

SpaceX ended June with 1.4 gigawatts of installed computing capacity, compared to 1 gigawatt in March and 0.4 gigawatts a year earlier. Musk expects the company to end 2026 with more than 2 gigawatts. And by the end of 2027, he said, the total "may, let's say, be closer to 10 gigawatts of compute than 5 gigawatts of compute."

Under its commitment, every gigawatt built from now on will use Nvidia hardware.

SpaceX has not filed any contract

The 10-Q SpaceX filed on the day of the conference does not mention Nvidia, nor has any subsequent filing.

What it does show is $28 billion in noncancelable purchase commitments at the end of June, of which $22.2 billion mature in 2027, described mostly as AI infrastructure, cloud capacity, and its spectrum purchase.

During the conference, when asked how much confidence he had regarding the chips, Musk said, "our understanding with NVIDIA is that we will receive a very significant percent of their GPUs next year."

Customer contracts, on the other hand, specify Nvidia chips, and I would argue they say more about SpaceX's tie to Nvidia than the commitment does. SpaceX's cloud service agreements with Anthropic cover about 325,000 Nvidia GPUs at $1.25 billion monthly through May 2029. Its agreement with Google, of Alphabet, covers about 110,000 Nvidia GPUs at $920 million monthly from October 2026 through June 2029. Each can be terminated with 90 days' notice after an initial period. And if SpaceX does not deliver the committed GPUs by Sept. 30, Google could walk away after a one-month grace period or pay only for the GPUs delivered.

So SpaceX has sold Nvidia capacity it has not yet finished buying, with delivery dates.

What SpaceX gives up without a second bid

A buyer of this size gives up two things.

The first is price. Nvidia's gross margin was 75% in its quarter ended July 26: on average, three-quarters of what customers pay Nvidia is gross profit.

The second is the timeline. Nvidia said in its quarterly report that it is "currently experiencing certain supply constraints," and Vera Rubin did not begin production shipments until the quarter that started on July 27. SpaceX's 2-gigawatt and 10-gigawatt targets depend on how much a single supplier ships of a product with limited supply.

Terafab, the chip plant that SpaceX is planning with partners, is its hedge against shortages, but the prospectus says there are still no definitive agreements.

Of course, management's answer is that profitability arrives quickly. Johnsen said current cloud economics provide SpaceX with "less than a one-year payback" on new capital allocated to computing, and the company signed contracts for another $6.7 billion in cloud service revenue during the first weeks of the third quarter. If that holds, paying more for the best computer could be the right decision.

But the stock arguably already assumes it will hold. SpaceX's market value sits near $1.9 trillion, with shares around $142 at the time of writing, more than 60 times the revenue a full year would produce at the second-quarter run rate. That price leaves no room for the bill to be larger or arrive later than planned, and SpaceX has committed to building it all on a single supplier's hardware.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices, Alphabet, and Nvidia. The Motley Fool has a disclosure policy.

David Tepper's hedge fund, Appaloosa Management, sold 690,000 shares of Micron Technology (NASDAQ:MU) during the second quarter, cutting its stake in the memory specialist by 41%, according to the fund's latest 13F filing. On its own, that looks like a manager heading for the exit.

But the same filing shows the opposite. The stake Appaloosa kept was worth about $1.13 billion at the end of June -- about 15% of the fund's equity portfolio, and its second-biggest position, behind only a $1.19 billion Amazon stake.

Micron stock gained about 242% during the quarter, which is how both things can be true at once.

What is a manager doing, then, when he sells that much of a stock and ends up more concentrated in it? I'd argue the size of what he kept is the more telling number. And what it reflects is Micron -- the cycle, and the earnings underneath it.

Image source: Getty Images.

Selling didn't shrink the bet

At the end of March, Appaloosa held 1.665 million Micron shares worth about $563 million. By the end of June, the 975,000 remaining shares were worth about $1.13 billion, in a portfolio totaling about $7.7 billion. Not only was the trimmed position worth twice what the bigger one had been in March, but it also took up more of the fund (nearly 15% of the portfolio, up from about 9.5%).

The stock did that work. After all, Micron climbed from about $338 at the end of March to about $1,154 at the end of June. Had Appaloosa sold nothing, Micron would have grown to more than a fifth of the fund.

In other words, the sale didn't so much shrink the bet as keep it from getting even bigger.

Is Tepper getting out of memory?

A 13F deserves one caveat, though. It is a snapshot of a single day (June 30, in this case), filed 45 days after the fact, and it says nothing about what a fund has done since.

Since then, Micron stock has pulled back to around $1,000 as of this writing, about 14% below where it ended the quarter.

And Appaloosa reportedly kept moving. In August, CNBC reported, citing a person familiar with the matter, that the fund had bought a bigger position in memory stocks since the quarter ended than it sold during it. The buying came as the group slumped.

The same filing also showed Appaloosa selling out of Sandisk (NASDAQ:SNDK), its smaller memory position. But set beside the buying reported since, even that exit arguably looks like profit-taking after a huge run.

The boom doesn't erase the cycle

Why leave 15% of a fund in one memory stock? Because the earnings have become enormous.

In the fiscal third quarter of 2026 (the period ended May 28, 2026), Micron's revenue reached $41.5 billion, more than quadruple the year-ago period's $9.3 billion. That was up from $23.9 billion just one quarter earlier, too. Net income came in at $28.2 billion, up about 15-fold year over year. And management guided the fiscal fourth quarter, which ended this week, to about $50 billion of revenue at a gross margin of about 86%.

Demand from artificial intelligence data centers is doing most of the work. Micron's cloud memory unit alone produced $13.8 billion of fiscal Q3 revenue, about four times its year-ago total.

Also worth noting: the guided step up in revenue, about $8.5 billion, would be smaller than either of the last two sequential jumps. And that is with an extra, 14th week in the quarter. Put another way, the growth is decelerating.

Of course, memory has always moved in cycles, and the down half is brutal. Three years ago, in fiscal 2023, Micron lost $5.8 billion as revenue roughly halved.

Investors haven't forgotten. Micron trades at about 6.5 times expected earnings for its next fiscal year -- and a price-to-earnings multiple that low, on earnings still climbing, usually means the market expects those earnings to fall.

Ultimately, I see the same opinion in Tepper's positioning and in Micron's valuation. The profits are enormous. How long they last is the question.

My own stance lands close to his. I view Micron stock as a hold here. I wouldn't sell a business earning like this, but this deep into the cycle's good half, I wouldn't put new money in at today's price either. And this cyclicality is risky. So keep that in mind.

Sure, holding without adding could mean missing more upside if this boom is still in its early innings. I'm comfortable with that.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Amazon and Micron Technology. The Motley Fool has a disclosure policy.

Shares of Joby Aviation (NYSE:JOBY) trade below $7 as of this writing, near their 52-week low, having lost about two-thirds of their value from a 52-week high of nearly $20. Investors, it seems, may be tired of waiting for electric air taxis to turn into meaningful revenue.

The company, meanwhile, isn't waiting. On Aug. 11, Joby announced an agreement to acquire Resonant Sciences, a defense technology company, for about $500 million -- about $450 million in cash plus $50 million in stock. It's a purchase big enough to roughly double Joby's revenue base.

Half a billion dollars is serious money for a company that still spends far more than it takes in. Here's a closer look at what the deal costs -- and what shareholders get.

Image source: Joby Aviation.

Fast growth in defense

Resonant, based in Dayton, Ohio, builds radio frequency (RF) and mission systems for U.S. national security customers. It also specializes in low-observability technology. In simpler terms, its systems help military aircraft sense their surroundings and avoid detection.

Not only did Resonant generate more than $100 million of revenue over its trailing twelve months, up about 40% year over year, but the business also produces positive adjusted earnings before interest, taxes, depreciation, and amortization (EBITDA). And demand is accelerating. In the first half of 2026, Resonant booked more than three times as much new business as it did a year earlier, and its backlog more than doubled year over year.

Joby's own outlook, raised in August, calls for full-year 2026 revenue of $115 million to $125 million. Resonant's trailing-twelve-month revenue, in other words, is nearly as large as everything Joby expects to book this year.

However, the deal isn't expected to close until the first half of 2027, subject to regulatory reviews. None of Resonant's results are in Joby's numbers yet.

Can Joby afford it?

Joby can afford the deal, I think, at least on today's balance sheet.

Joby's cash and short-term investments stood at about $2.3 billion at the end of June. Management expects to use between $385 million and $415 million of it in the second half of 2026 alone. The $450 million going to Resonant works out to about a fifth of the war chest.

In February, Joby raised about $576 million in net proceeds from a stock offering and another $670 million from an offering of convertible notes. The company, in other words, is spending cash investors handed it months ago, not cash the business generated.

Between the guided second-half cash use and the Resonant payment, about $850 million of the June 30 balance is already spoken for. The $50 million of stock barely registers, adding less than 1% to the share count. However, on the same day it announced the deal, Joby also put a program in place to sell up to $750 million in new stock over time.

Joby stock is still an air taxi bet

Almost none of Joby's revenue today comes from electric air taxis.

Of the $38.6 million the company reported for the second quarter, $36.2 million came from passenger flights booked through Blade (the passenger business Joby acquired in August 2025). Blade's demand peaks in the summer, and the second quarter's $38.6 million was up from about $24 million in the first. And the full-year outlook implies a second half no bigger than the first, not an acceleration.

But the air taxi business itself isn't generating revenue yet. Joby said in its August update that it made its strongest quarterly progress yet in the fifth and final stage of FAA type certification. The company is still targeting its first passenger flights before the end of 2026, with the first flights under a federal pilot program expected in Texas this month.

The price of the deal also looks reasonable next to Joby's own valuation. At a market capitalization of about $6.7 billion, Joby trades at more than 50 times the midpoint of its 2026 revenue outlook. Resonant, by comparison, is being bought for less than 5 times its trailing sales -- a modest price, I'd argue, for a business growing about 40%.

Investors, in short, aren't paying for the revenue Joby has today. They're paying for the air taxi business it hopes to build.

Ultimately, the acquisition strikes me as a sensible use of Joby's cash. It buys a business that could keep growing whether or not air taxis arrive on schedule. But the deal doesn't change what this growth stock is: a bet that electric air taxis become a big business before the cash runs low.

Of course, the certification work isn't finished, and the first paying passengers haven't flown. I would avoid buying shares here. If those passengers arrive on schedule and spending starts to fall, I would consider changing my mind.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

Vertiv (NYSE:VRT) said Wednesday that it agreed to acquire UtilityInnovation Group (UIG), a designer of on-site power systems for data centers, in a deal worth up to $2.6 billion.

But only about $1.45 billion of that is payable in cash at closing. The remaining $1.15 billion is contingent on the acquired business hitting earnings targets.

That split is the most telling part of the announcement. Vertiv is spending big on the idea that power availability is becoming the thing that most limits how fast artificial intelligence (AI) data centers get built. But it structured the deal so the seller has to prove nearly half the price before collecting it.

Image source: The Motley Fool.

A bet on faster power

UIG, founded in 2020, designs and delivers microgrids (self-contained power systems that can combine on-site generation, energy storage, and utility power) for data center operators in the United States and Europe. Its products include a controls platform and switchgear that coordinate multiple power sources in real time.

Vertiv already sells much of the power and cooling equipment inside a data center. Grid constraints increasingly limit how fast AI infrastructure can be deployed, the company said. UIG extends that portfolio upstream to the point where a facility connects to the grid.

Not only does that put Vertiv in the conversation earlier, when a site's power design is being decided, but it also keeps customers from being tied to any single power generation technology or supplier.

"For AI data center operators, competitive advantage increasingly depends on how quickly they can move from site selection to first token," Vertiv CEO Gio Albertazzi said in the announcement.

In other words, the race is to get new capacity powered on and producing.

The deal is expected to close in the fourth quarter of 2026, subject to regulatory approval. Vertiv also expects the acquisition to boost adjusted earnings per share in its first year.

The other $1.15 billion must be earned

The $1.45 billion base price represents about 13 times UIG's expected 2027 earnings before interest, taxes, depreciation, and amortization (EBITDA), according to Vertiv. The additional payments are tied to UIG hitting EBITDA targets over 12- and 24-month measurement periods. Vertiv also said the EBITDA multiple it ends up paying should be "significantly lower" if the full earnout is paid.

Work backward, and Vertiv is effectively saying it expects UIG to produce about $110 million of EBITDA in 2027. And for the full $2.6 billion price to work out to less than 13 times EBITDA, UIG's earnings would need to clear about $200 million -- nearly double that expectation.

In short, Vertiv pays full price only for growth that shows up. Even at the base price, the valuation isn't cheap for a business founded in 2020. But I'd rather see part of the risk of those growth hopes sit with the sellers than all of it with Vertiv shareholders -- and this structure puts it there.

Can Vertiv afford it?

Easily. Vertiv said it expects to fund the acquisition from existing resources -- and it can. The company ended the second quarter of 2026 with $5.6 billion of liquidity and a net cash position, up from $5.0 billion three months earlier. Second-quarter adjusted free cash flow was $925 million, up 234% year over year, and management guided for adjusted free cash flow of $2.4 billion to $2.6 billion this year. The $1.45 billion closing payment, then, amounts to about seven months of cash generation at the midpoint of guidance.

And Vertiv isn't buying revenue growth to mask a slowdown at home. Organic sales rose 23% year over year in the first quarter of 2026, and the second quarter's 18% was a step down that management attributed to timing shifts.

Guidance calls for 34% to 36% organic growth in the third quarter, with about 31% expected for the full year. In other words, management expects growth to reaccelerate, not cool.

Shares of the growth stock trade around $269 as of this writing, about 29% below their 52-week high.

Ultimately, I like the way this deal is built. Vertiv is paying up front for the business UIG is expected to have next year, and the other $1.15 billion depends on what UIG delivers.

Sure, the deal still needs regulatory approval to close. But the money is aimed at arguably the biggest constraint in AI infrastructure today, and it's coming from a company generating more cash than it needs. That seems like a sensible use of it to me.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Vertiv. The Motley Fool has a disclosure policy.

Robinhood (NASDAQ:HOOD) shares were up about 15% as of this writing Thursday, at about $123.

The jump followed a wave of analyst notes and a record day on its own new blockchain network. Morgan Stanley upgraded the stock Tuesday to overweight from equal weight and lifted its price target to $150 from $124, and more bullish notes followed this week.

Morgan Stanley analyst Michael Cyprys argued that Robinhood's expanding product lineup is producing more activity and more revenue per customer. In plain terms, they're arguing Robinhood is no longer just a trading app.

And Robinhood itself put a number on that idea in late July: 13 business lines that have each reached $100 million or more in annualized revenue.

Since then, network data suggests a 14th has joined the list, and it didn't exist three months ago.

Image source: Getty Images.

The count holds up

Robinhood's second-quarter report, released in late July, showed record revenue of $1.31 billion, up 32% year over year, and net income up 48% (helped by one-time investment gains). Chief financial officer Shiv Verma said the results reflected the company's product pace, with "Robinhood Legend and the Credit Card business joining our growing roster of now thirteen different business lines that have reached $100 million-plus in annualized revenues."

I count 13 lines in the 10-Q's revenue table that annualize above $100 million (anything above $25 million in the quarter). They span options, event contracts, cryptocurrencies, and equities, five interest-based lines led by margin lending, Gold subscriptions, proxy services, and two catch-all "other" buckets. The company's list is built on products rather than filing line items (Robinhood Legend doesn't get its own row), but both counts land at 13.

Lines can fall off the list, too. Securities lending was above the bar a year ago, at $54 million in the quarter, and produced just $10 million in this one.

How big is the newest line?

The 14th didn't appear in any of those documents. It barely existed when they were filed.

Robinhood Chain, the company's own blockchain network built for real-world assets such as tokenized stocks, went live on July 1 -- one day after the second quarter ended.

Not only did the network set a fee record of about $3.8 million on Tuesday, but it also collected more than the Ethereum and Base networks that day. It broke that record Wednesday at about $4.5 million, according to DefiLlama data. Its average daily fee pace over the past 30 days now annualizes to about $179 million.

That $179 million needs two adjustments. Robinhood sends about 10% of the network's revenue after costs back to the Arbitrum ecosystem, whose technology the chain runs on. And annualizing the hottest stretch of a two-month-old network is generous math -- the chain's lifetime revenue through the start of this week was only about $10 million, and daily fees that spike may fade just as quickly.

Even with those adjustments, the pace arguably clears $100 million. Zoom out, though, and it amounts to about 2% of Robinhood's revenue pace of roughly $5.2 billion. Big enough to make the list, and far too small to carry the company.

Order flow still supplies a third of revenue

How much of the company still runs on its best-known business, routing customers' stock and options trades to market makers?

In the second quarter, equities produced $129 million of transaction revenue and options $342 million. Together, the two lines produced about 36% of total revenue.

But that share isn't shrinking. A year earlier, the two supplied about a third of revenue as well, and both are still growing. Equities transaction revenue nearly doubled year over year, while options revenue rose 29%.

The diversification is happening elsewhere. Cryptocurrency trading revenue was $160 million a year ago, $134 million in the first quarter, and $100 million in the second -- a steady step down. Meanwhile, event contracts (Robinhood's prediction-markets business) went from $10 million a year ago to $104 million in the first quarter and then $156 million in the latest one, and margin interest nearly doubled to $215 million.

Ultimately, the case under this week's upgrades mostly checks out against Robinhood's own disclosures. The valuation is where I hesitate.

After Thursday's jump, shares cost about 43 times the earnings analysts expect the company to generate next year, while brokerage peer Charles Schwab costs about 14 times its own next-year forecast. Of course, some premium is deserved. After all, Schwab isn't growing revenue 32% or adding two new $100 million lines in a single quarter.

However, higher price targets aren't a reason to buy a stock, and neither is a 15% pop. I wouldn't sell a business that keeps adding $100 million lines. But I wouldn't chase the growth stock here, either. I view it as a hold for now.

Charles Schwab is an advertising partner of Motley Fool Money. Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool recommends Charles Schwab and recommends the following options: short September 2026 $95 calls on Charles Schwab. The Motley Fool has a disclosure policy.

Alphabet (NASDAQ:GOOG)(NASDAQ:GOOGL) dodged a breakup on Wednesday. U.S. District Judge Leonie Brinkema declined the government's request to force a sale of the company's ad exchange and its publisher ad server, instead accepting most of the parties' proposed behavioral remedies, with modifications of her own. The decision lands about 16 months after the same judge found Google had illegally monopolized key advertising technology markets, and it closes off the most severe outcome the case could have produced.

But the ad tech Google just won the right to keep sits in the one piece of Alphabet that is already shrinking. Revenue in the company's Google Network business, where the contested products live, has fallen for three straight years. And it slipped again in the second quarter while every other revenue line at the company grew.

In other words, Alphabet spent years of legal effort defending what is arguably its least important business. That context matters, I think, before assuming this week's ruling changes much for shareholders.

Image source: Alphabet.

New rules, same owner

The ruling, entered Wednesday in the Eastern District of Virginia, stops short of the structural remedy the Justice Department wanted. Google won't have to sell AdX, its ad exchange, or DFP, its publisher ad server -- the two products it bundles together as Google Ad Manager.

Instead, Brinkema accepted most of the behavioral remedies the two sides had proposed, reshaping them where she saw fit. The proposals on the table included requiring Google to make real-time AdX bid data available to rival ad servers and letting publishers set different price floors for individual bidders. They also included ending the first look and last look privileges that gave its exchange the first or final opportunity to win an ad sale.

Worth noting: the judge's full written opinion is sealed for about two weeks while both sides review it for confidential material, so the finer details of the remedies aren't public yet.

The business it kept is shrinking

Google Network includes the revenue Alphabet generates from AdSense, AdMob, and Google Ad Manager -- the money it makes selling ads on other companies' websites and apps instead of on its own properties.

And the decline there isn't new. Network revenue slipped from $31.3 billion in 2023 to $30.4 billion in 2024, then $29.8 billion last year. In its most recent annual report, Alphabet attributed last year's drop primarily to AdSense, and Google Network ad impressions fell 7% for the year.

The slide has continued into 2026. Network revenue fell about 4% year over year in the first quarter and slipped again in the second, coming in at $7.3 billion.

Compare that to the rest of the company. Second-quarter revenue from Google Search & other grew 17% year over year, YouTube ads grew 13%, subscriptions, platforms, and devices grew 15%, and Google Cloud surged 82%, led by demand for artificial intelligence infrastructure. Google Network was the only revenue line that shrank.

The business now accounts for about 6% of Alphabet's total revenue. And the new rules, which aim to open Google's auctions to more competition, could pressure that line further.

Does the ruling change the investment case?

Not much, I'd argue. The remedies land on tools in a small and fading corner of an otherwise thriving business. The tech company's second-quarter revenue rose 24% year over year, reaching $119.8 billion (the company's 12th straight quarter of double-digit revenue growth). Further, operating income rose 30%, and the company's operating margin expanded 2 percentage points to 34%.

The ruling mostly removes a tail risk. After all, a forced sale would have meant years of appeals and a messy separation. Instead, Alphabet gets compliance obligations in a business that matters less to its results with every passing quarter.

Of course, the Justice Department could still appeal, so the case may not be over. Still, the worst case is off the table for now.

Meanwhile, the stock trades around $342 as of this writing, well below its 52-week high of $408.61. With a price-to-earnings ratio of about 23 on the earnings analysts project for next year, shares arguably look reasonably priced for a company growing this fast with an expanding operating margin.

Ultimately, this case was never the reason to buy or avoid Alphabet stock. The growth story runs through Search and Google Cloud -- and the court just confirmed the contested ad tech stays put, with new rules attached. I wouldn't buy or sell shares over this ruling.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet. The Motley Fool has a disclosure policy.

For the past three months, the biggest number attached to Broadcom (NASDAQ:AVGO) was CEO Hock Tan's forecast of more than $100 billion in artificial intelligence (AI) semiconductor revenue for fiscal 2027. On Wednesday evening's earnings call, he replaced it.

The new fiscal 2027 target is about $115 billion. And for the first time, Tan put a number on fiscal 2028: $230 billion, about four times the $58 billion of AI revenue Broadcom now expects for the current fiscal year.

Both years, he told analysts, come with the supply already secured.

The forecasts landed on top of a record quarter. In the fiscal third quarter of 2026 (the period ended Aug. 2), revenue rose 86% year over year to $29.6 billion, and net income more than tripled to $13.1 billion. Still, doubling AI revenue twice more in two years is a promise about manufacturing as much as about demand.

Image source: The Motley Fool.

Two more doublings

Broadcom now expects $58 billion of AI semiconductor revenue in fiscal 2026, up 186% from about $20 billion last year. "In 2027, we have secured the supply to again double AI revenue to approximately $115 billion," Tan said on the call. And demand, he added, "actually exceeds this outlook." He extended the same line to fiscal 2028, with what he called line of sight to $230 billion. "Here again, we have secured the supply to meet this outlook," he said.

AI semiconductor revenue was $10.8 billion in the fiscal second quarter. Not only did it jump to a record $16.7 billion in the fiscal third quarter, up 221% year over year, but guidance also calls for $21.7 billion in the fiscal fourth -- each quarter $5 billion to $6 billion bigger than the one before it.

However, the bulk of the outlook rests on six custom accelerator (XPU) customers. That is a short list for a number this size, I'd argue.

What would delivering $230 billion take?

Zoom out, and $230 billion of AI revenue in fiscal 2028 works out to an average of nearly $58 billion a quarter. For perspective, Broadcom as a whole (semiconductors and software combined) just reported a record $29.6 billion quarter. Two years from now, the AI line alone would need to average almost double that.

The physical version of that math is measured in gigawatts of data center capacity. Anthropic alone is expected to deploy 5 gigawatts of TPU chips in 2027, with line of sight to another 10 gigawatts after that. Broadcom designs those custom accelerators with Google parent Alphabet.

Deliveries like those are why Broadcom has been locking up manufacturing capacity years in advance, down to building its own chip substrate plant in Singapore.

Secured supply isn't deployed supply

What "secured the supply" commits is the part Broadcom controls. The company has lined up the leading-edge wafers, high-bandwidth memory, and substrates needed to build the chips. What it can't commit is everything after those chips ship. "[E]ven as we ship the chips, are they going to be deployed on a timely basis?" Tan said. Land and power dictate when a customer's data center capacity turns on, he said, and any piece of the chain may become the bottleneck.

Of course, an outlook is not revenue in hand, either. Tan's fiscal 2027 target itself just moved, three months after he last reiterated it. But with demand running ahead of the forecast, the main risk isn't the orders -- it's whether everything gets built and deployed on time.

Shares slipped in extended trading Wednesday, then recovered as the call went on. At about $345 as of this writing, down sharply Thursday morning, the stock trades at about 18 times the earnings analysts expect for fiscal 2027, the $115 billion year. That risk looks worth taking at this price, I think.

Tan also said the company is on target to exceed $30 in earnings per share in fiscal 2028. If he is right, today's buyer is paying about 12 times those earnings. For a company expected to keep doubling its largest business, that arguably looks cheap.

Ultimately, I've viewed the growth stock as a hold in recent weeks because the big forecast still had to show up in reported numbers. Wednesday's report moved the forecast up instead.

Sure, deployment timing could make a quarter or two look ordinary along the way, and six customers is still a short list. But with supply secured and demand running ahead of the fiscal 2027 outlook, I'd buy Broadcom stock here.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet and Broadcom. The Motley Fool has a disclosure policy.

Nvidia (NASDAQ:NVDA) reported its fiscal second-quarter results on Aug. 26, and the figures were extraordinary. Quarterly revenue rose 106% year over year to $96.2 billion, accelerating from the 85% growth recorded in the fiscal first quarter. Data center revenue rose 117% to $89.0 billion.

But the figures that caught my eye came out of the earnings call, from chief financial officer Colette Kress.

"With cloud industry backlog now greater than $2 trillion, [capital expenditures] by the top 5 hyperscalers is expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027," Kress said.

That backlog is the pipeline behind both spending figures: cloud customers turn it into data centers, and a meaningful share of every data center dollar goes to Nvidia. So the way to size Nvidia's cut is to pin down that share.

Image source: Nvidia.

How much of hyperscaler spending goes to Nvidia?

Nvidia divides its data center revenue into two categories. The hyperscaler category takes in the public clouds plus the world's biggest consumer internet companies. The rest (AI clouds, industrial and enterprise customers, which the company abbreviates as ACIE) covers everyone else.

Revenue from hyperscalers reached $48.7 billion in the fiscal second quarter. That was a 13% rise from the $43.1 billion in the fiscal first quarter, and was more than double the $24.2 billion Nvidia recorded a year earlier (Nvidia recast prior periods after moving a customer to the hyperscaler category).

Multiply the $48.7 billion from the second quarter by four, and revenue from hyperscalers reaches a run rate of about $195 billion a year. If you compare that figure with the nearly $800 billion in capital expenditures Kress says the top five hyperscalers are expected to make in 2026, Nvidia's share comes out to about 24%.

The comparison is loose, to be sure: Nvidia's fiscal year ends in late January, so its fiscal 2027 aligns only approximately with calendar 2026, and its hyperscaler category includes more customers than those five -- which means the true share of those five companies' spending runs somewhat lower. Even so, the last two quarters come to about $92 billion against half of this year's $800 billion -- about $400 billion, if that spending were distributed evenly throughout the year -- or about 23%.

Nvidia's share of the $1.3 trillion is about $315 billion

If that share holds, 24% of $1.3 trillion equals about $315 billion in revenue from hyperscalers in calendar 2027, most of which falls into Nvidia's fiscal 2028. That single category would be larger than the $215.9 billion Nvidia brought in for all of fiscal 2026.

And hyperscalers represent only about half of Nvidia's data center business. ACIE revenue was $40.3 billion in the second quarter, a 25% quarter-over-quarter increase and a 138% year-over-year increase. Kress said that non-hyperscaler business should continue to represent about half of data center revenue.

If that distribution holds and the $315 billion is doubled, data center revenue in fiscal 2028 comes out to about $630 billion. Use the second quarter's actual split instead (hyperscalers were about 55% of the data center total) and the figure comes out closer to $575 billion.

Nvidia cannot manufacture everything its customers want

Wherever demand for Nvidia's products lands, there's a holdup: manufacturing.

Kress said the company's preliminary expectation is that fiscal 2028 revenue will grow about 70%, and that the figure reflects supply constraints.

CEO Jensen Huang put it more directly, saying "even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%."

What does 70% equal in dollars?

Nvidia's revenue during the first half of fiscal 2027 was $177.8 billion, and the company forecast $108 billion for the third quarter. And a fourth quarter that matched the third would put fiscal 2027 near $394 billion. If that figure grows by 70%, fiscal 2028 revenue comes to about $670 billion.

Data center revenue accounted for more than 92% of Nvidia's total last quarter, so $670 billion in total revenue implies about $620 billion for the data center business -- right in the middle of the $575 billion to $630 billion the demand math yields. That is what you would expect if supply is the real limit: revenue can only reach what Nvidia can build, and the demand Huang says runs well past 70% shows up in the backlog instead of the income statement.

One risk is how much it costs to manufacture all that. Memory prices are rising, and the company now expects its gross margin to bottom out in the fiscal fourth quarter between 71% and 72%, compared with 75% in the second quarter.

The other risk is the share itself. Capital spending also buys land, buildings, power, and networking gear, and the big cloud companies design some chips of their own -- so Nvidia's quarter of the total is an observation, not a guarantee.

As for the stock, it trades at about $217 as of this writing, up about 4% since the report and about 8% below its 52-week high. The stock trades at about 27 times earnings. Relative to the earnings analysts expect for fiscal 2028, the price-to-earnings multiple drops to about 14, which seems reasonable to me for a company expecting 70% growth.

The semiconductor industry is cyclical, of course, and a $2 trillion backlog could shrink just as fast as it was built. But Nvidia has already told the market how much it expects to grow next year, and said demand is higher than that figure. With this in mind, I do think shares look attractive here. But I would simply maintain a modest position, given how cyclical chips have always been.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Nvidia. The Motley Fool has a disclosure policy.

Intel (NASDAQ:INTC) closed at $24 a share a year ago. As of this writing, it trades near $89, about 3.7 times the price a year ago. The stock has also risen around 141% in 2026 alone.

However, the stock hit a high of $142.35 in late June and has fallen around 38% since then. It also trades below the $95 a share that Intel got in August, when it sold about 242 million new shares for approximately $23 billion.

The stock's direction from here depends on three things: whether the foundry wins external customers, how quickly earnings grow under more than $20 billion in capital expenditures, and across how many shares those earnings are split. Here is how I would turn those three into a range.

Image source: Intel.

Growth is back

Intel's revenue in the second quarter rose 25% year over year, to $16.1 billion -- an acceleration from the first quarter's 7% and the fourth quarter of 2025's 4% decline. Non-GAAP (adjusted) gross margin reached 41.8%, 12 percentage points wider than a year earlier, and adjusted earnings per share were $0.42, versus a loss of $0.10 in the year-ago quarter.

The data center and artificial intelligence (AI) segment did most of the work, with revenue rising 59% year over year, to $6.3 billion, and operating income of $2.5 billion.

And management forecasts third-quarter revenue between $15.8 billion and $16.8 billion, implying about 19% growth at the midpoint -- slower, but well above anything Intel posted in 2025.

Will the foundry win any big customers?

Intel's foundry revenue grew 31% year over year, to $5.8 billion, and its operating loss narrowed to $2.1 billion, from $3.2 billion a year earlier and $2.4 billion three months prior. But almost all of that revenue comes from Intel making chips for itself. Revenue from external customers was $293 million.

The company is spending as if that could change. Chief financial officer David Zinsner said on the second-quarter earnings call that Intel now expects capital expenditures of more than $20 billion in 2026 and that 2027 spending should be "significantly above the 2026 levels."

None of this has a big external name attached yet. Fortinet joined in July for a security processor, but the grand prize is Intel 14A, the next manufacturing process.

Version 0.9 of the 14A design kit (the toolset external chip designers work on) is scheduled for October. And CEO Lip-Bu Tan said in January that he expected customers to start making firm supplier decisions in the second half of this year and during the first half of 2027.

Those customer decisions, I believe, are what drive both ends of the range. Zinsner said in January that Intel would not spend on 14A capacity until it had secured customers. But Tan said on the second-quarter earnings call that Intel decided during the quarter to fully commit to high-volume 14A production in 2028, citing demand for its own products along with customer conversations. So the money will be spent either way. An external commitment determines whether customers help pay for it.

A larger share count

Intel had 5.04 billion shares outstanding at the end of June, and the August sale added about 242 million. That puts the number near 5.3 billion, approximately 21% above the year-ago quarter's average of 4.37 billion.

Of course, the balance sheet strengthened. Intel had about $30 billion in cash and short-term investments at the end of June, before the sale. But every dollar the company earns will be split across a fifth more shares than a year ago.

Where will Intel stock be in 5 years?

Analysts expect around $2 in adjusted earnings per share next year. At $89, that equals about 44 times next year's earnings.

At the low end, no big 14A customer emerges and the foundry continues to lose money on Intel's own chips. That leaves a products company earning about $2 a share, which, at 15 times earnings, could put the stock near $30.

At the midpoint, the foundry reaches breakeven by the end of the decade, earnings rise to about $3.50 a share, and a 25-times-earnings multiple puts the stock near $90.

At the high end, 14A wins a couple of big customers, the foundry turns profitable, and earnings reach about $6 a share by 2031. At between 25 and 28 times earnings, that equals between $150 and $170, or an annual return of between 11% and 14% from here.

In other words, the current price already assumes the middle scenario. It could be said that the business is in its best shape in a decade. But, at this price, the reward for being right on the foundry is approximately the same size as the penalty for being wrong. I would stay on the sidelines for now. An identified 14A customer with volume to back it up would change my mind.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Fortinet and Intel. The Motley Fool has a disclosure policy.

Netflix's (NASDAQ:NFLX) most-watched English-language film during the week of Aug. 24 to Aug. 30 was not at all a Netflix film. It was Grand Theft Auto VI: An Extended Look, a 27-minute trailer for the video game that Rockstar Games will release in November. The trailer earned 31.1 million views and topped the list in 87 of the 93 countries that Netflix monitors.

Netflix premiered it on Thursday, Aug. 27, and Rockstar posted the same footage on YouTube, for free, six hours later.

It's strange to see at the top of a film list. But it fits with what Netflix has been telling investors it wants: moments, not hours.

Image source: Netflix.

Why a 27-minute trailer wins on views

Netflix counts a view by dividing total hours watched by a title's runtime, so a 27-minute trailer needs much less watch time than a two-hour film to register a high figure.

Measured in hours, the trailer accumulated 14 million during its four days in the measurement period, while the film that finished second, the crime thriller The Whisper Man, earned 23.2 million views, but 44.1 million hours. Even the most-watched film from the previous week, Don't Say Good Luck, generated 20.3 million hours with only 12.8 million views.

And even by views, it's not a record. The Rip earned 41.6 million views during its premiere week in January. The Grand Theft Auto trailer fell about 25% short.

Of course, 31 million views for a trailer in four days is a high figure. But it's not the same as accumulating the most hours on the streaming service.

How much did it cost Netflix?

Nobody wants to say. Both Netflix and Take-Two Interactive (NASDAQ:TTWO), Rockstar's parent, declined to discuss the financial terms.

When asked during Take-Two's Aug. 7 earnings call whether Netflix paid a premium for exclusivity, CEO Strauss Zelnick said he "wouldn't normally give detail on sort of the nature of the back and forth or the terms of the arrangement."

At another point, he called Netflix "a great marketing partner for us and distribution partner," and added that the premiere "is part of Rockstar Games marketing strategy."

In other words, this wasn't Netflix buying a film. It was Rockstar using Netflix as a launch platform for an announcement, with a six-hour head start.

It's also not the first deal between them. Rockstar's Grand Theft Auto trilogy came to Netflix as mobile games in December 2023, and the last of the three, San Andreas, left the service in December 2025.

For its part, Netflix called the presentation a cultural moment. "It's a reflection of what we hope Netflix is becoming: a place where the most ambitious storytelling, from any medium, can find the biggest possible audience," said Brandon Riegg, the company's vice president of nonfiction series, in Netflix's announcement.

Netflix chases moments, not hours

The company's second-quarter shareholder letter, published in July, shows why a deal like this might be attractive. Members watched 2% more hours during the first half of 2026 than a year earlier, a slight acceleration from the 1.5% growth in 2025. Meanwhile, second-quarter revenue was $12.6 billion, up 13% year over year. So the growth comes from more members, higher prices, and advertising, not from people watching much more content. Advertising revenue grew more than 150% in 2025, surpassing $1.5 billion, and management expects it to about double again this year.

Management has also been telling investors hours aren't what matters. Live programming, the letter noted, will account for just over 5% of content spending this year, but only about 1% of hours watched. However, live events represent six of the 10 biggest sign-up days in Netflix's last five years.

"[T]here is not a linear relationship between view hours and revenue and profit because all hours are not created equal," said co-CEO Greg Peters during the second-quarter earnings call.

A video game premiere fits with that way of thinking. From a broader perspective, 14 million hours barely register on a service whose members watched 97 billion hours in six months -- about 2 billion in any four-day period.

Ultimately, I think it's a smart trade for Netflix, but also one that doesn't change the important numbers. Management expects revenue growth to slow to 12% in the third quarter, and engagement is growing at a low single-digit rate.

But many moments like this, compounded over time, could be a game changer.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Netflix and Take-Two Interactive Software. The Motley Fool has a disclosure policy.

Micron Technology's (NASDAQ:MU) labor unions in Taiwan said Tuesday they are moving toward a possible strike unless the memory maker overhauls how it pays bonuses. For fiscal 2026, they want a one-off bonus the unions size at about 83 months of salary per Taiwan-based employee -- nearly seven years of pay, delivered at once. And starting in fiscal 2027, they want Micron's incentive plan replaced with a system that sets bonuses at 15% of operating profit, paid quarterly rather than annually.

Both numbers sound impossible. But Micron is earning profits in this artificial intelligence (AI) memory cycle that make even demands like these payable.

The question for investors, I'd argue, is what a strike at the company's largest manufacturing base could do to the supply of memory feeding the AI build-out. And a lot rides on the answer: the growth stock trades around $933 as of this writing, and the company's market value is about $1.05 trillion.

Image source: Micron.

The gap behind the demand

The two unions, based in Taoyuan and Taichung, have about 10,000 members between them. More than 80% of members who took part in an internal survey in August backed strike action. A formal strike vote could follow as early as this month if mediation fails.

The workers' case leans on precedent. Samsung Electronics' (OTC:SSNLF) semiconductor division and rival SK Hynix reportedly pay bonuses worth 10.5% and 10% of operating profit, respectively. Samsung's arrangement grew out of a deal in May that averted a strike of its own.

Micron's Taiwan bonuses, by contrast, have reportedly run about 2.6 months of salary, under a plan capped near five months. That gap is what the unions want closed.

For its part, Micron's Taiwan office has said this year's performance-bonus payout will be the highest in the company's history.

What would the demands cost?

The one-off bonus is hard to size from outside the company, since it turns on each worker's salary. The recurring demand is easier. Micron's fiscal third quarter of 2026 (the period ended May 28, 2026) produced operating income of $33.3 billion on revenue of $41.5 billion -- an 80% operating margin.

That profit line has been climbing at an extraordinary rate. After all, revenue in the quarter more than quadrupled year over year. Operating income was just $2.2 billion in the year-ago period, at a 23% operating margin, and $16.1 billion in the fiscal second quarter, at a 68% margin.

So 15% of the latest quarter's operating profit works out to about $5 billion -- for a single quarter. That's more than double the operating profit the whole company generated in the same period a year ago.

And the base the unions want a share of is still growing. Micron's guidance for the fiscal fourth quarter (the current period) calls for revenue of about $50 billion at a gross margin around 86%. At profitability like that, the union formula would produce a quarterly bonus pool of more than $6 billion.

Sure, those are staggering sums. But even paying the full 15%, Micron would keep 85% of an operating profit line that ran $33 billion last quarter. The money, arguably, is there.

Taiwan is the base the boom runs on

A majority of Micron's DRAM output in fiscal 2025 came from the company's fabrication facilities in Taiwan. Micron's own annual filing says any loss of that output could have a material adverse effect on the business.

Taiwan also holds about $19 billion of Micron's long-lived assets, such as plants and equipment -- more than any other country -- and it's where the tech company is modernizing DRAM and high-bandwidth memory capacity to meet rising demand.

In other words, the workers weighing a strike sit at the center of a business running an 80% operating margin. At margins like these, almost every dollar of lost revenue would come straight out of profit.

Of course, the dispute is headed to mediation for now, not a walkout. Mediation sessions were reportedly expected in late August and mid-September, and a strike vote could follow as early as this month if the talks fail. No vote has been scheduled yet.

And fiscal 2026, the year whose bonus is in dispute, ends this week. So the profit behind that bonus is almost fully earned.

Ultimately, I don't think the 83-month figure is the number that matters most for Micron's valuation. Samsung's standoff ended in a deal, and that seems the more likely ending here, too. Micron has already signaled a record payout is on the way, and it can close much of the bonus gap with money it is already generating.

What the company can't replace is lost output from the fabs that make most of its DRAM. A deal, even an expensive one, looks like the better outcome for shareholders.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Micron Technology. The Motley Fool has a disclosure policy.

Sandisk (NASDAQ:SNDK) built its name on memory cards and flash drives. But in its fiscal fourth quarter of 2026, which ended July 3, the company sold $2.98 billion of storage to datacenter customers -- about a third of its $8.97 billion in total revenue. A year earlier, that datacenter business generated just $213 million in quarterly sales.

The scale of the change goes beyond one quarter. Sandisk separated from Western Digital in February 2025, and in fiscal 2026, its first full year on its own, it generated $20.25 billion of revenue, up 175%, with the datacenter piece up 437%.

But the bigger change isn't who is buying the company's storage. It's how they're buying it.

Image source: Getty Images.

A steep mix shift

Showing just how fast the customer base is moving, datacenter revenue has climbed for three straight quarters. It was $440 million in the fiscal second quarter, about 15% of the company's revenue. By the fiscal third quarter, it had grown to $1.47 billion, about 25%. And it hit $2.98 billion in the fourth, about a third of the total.

That said, the edge business, which sells flash storage to makers of PCs, smartphones, gaming consoles, and cars, is still the biggest piece of the company, at $5.43 billion of fiscal fourth-quarter revenue.

Consumer products, however, contributed just $556 million, about 6% of the quarter and down 5% year over year. In other words, the retail cards and drives Sandisk is named for are now its smallest business.

What do the contracts guarantee?

Memory pricing is famously boom-and-bust, and Sandisk's answer is what it calls the New Business Model (NBM) -- multiyear supply agreements signed directly with large datacenter and edge customers.

The terms are what make the shift structural. Chief financial officer Luis Visoso said on the company's August earnings call that Sandisk now has 10 of these agreements across eight customers, five of them signed since April. The agreements run as long as five years, with a weighted average duration of more than four years. Pricing includes fixed and variable elements, with the variable portion subject to floors and ceilings. In total, the NBMs Sandisk has signed represent a minimum of $93.9 billion in expected revenue, assuming every variable price settles at its floor. The deals are also backed by $16.5 billion of customer cash deposits and financial instruments.

The contracted share is still growing, too. Management expects NBMs to cover about half of Sandisk's bit shipments in fiscal 2027, and about two-thirds in fiscal 2028.

Of course, contracted volume isn't the same thing as guaranteed revenue, and the ceilings may cap Sandisk's upside if spot prices keep climbing. But I'd argue the floors matter more than the $93.9 billion headline number. Minimum prices under a growing share of shipments change the downside math in an industry known for brutal crashes.

Higher prices did most of the work

For all that structure, fiscal 2026 was mostly a pricing story. Sandisk's total products sold rose by a mid-teens percentage on an exabyte basis (a measure of raw storage volume shipped), while revenue rose 175%. And management said about two-thirds of the fiscal fourth quarter's sequential revenue growth came from higher pricing, with one-third from higher volumes.

That pricing boom shows up most clearly in profitability. Gross margin reached 84.6%, up from 26.2% in the year-ago period.

The company also swung to $6.9 billion of quarterly net income from a small loss a year earlier. And free cash flow for the full year went from a $120 million outflow in fiscal 2025 to $11.5 billion.

Management doesn't expect a cooldown yet, either. It guided fiscal first-quarter 2027 revenue between $10.3 billion and $10.8 billion, up 15% to 20% sequentially, with gross margin expected to stay at 83% to 85%.

The market remains skeptical, though. Shares trade around $1,537 as of this writing, down about 35% from a 52-week high, at about 21 times fiscal 2026 earnings.

Measured against expected earnings for fiscal 2027, the price-to-earnings multiple falls to about 7. A steep decline in memory pricing, in other words, is arguably already priced in.

Is Sandisk a different company now? On the customer side, I think it clearly is. A third of revenue comes from data centers, about half of this fiscal year's shipments are already committed under contract, and there are price floors where prices used to float freely.

However, the new model hasn't been tested by a downturn yet. And even Sandisk's own long-term financial model, laid out at its August investor day, calls for non-GAAP (adjusted) gross margins of about 80% for fiscal 2028 through 2030 -- below the 84.6% it just reported. The floors cushion a fall in contracted pricing. They don't make fiscal 2026's boom prices permanent.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Western Digital. The Motley Fool has a disclosure policy.

Micron Technology's (NASDAQ:MU) fiscal third quarter of 2026 (the period ended May 28, 2026) produced $41.5 billion of revenue. The company's entire fiscal 2025, its biggest year to that point, produced $37.4 billion. The memory specialist collected more revenue in 13 weeks than in its whole previous year.

Alongside that late-June report, Micron guided the fiscal fourth quarter to $50.0 billion of revenue, give or take $1.0 billion, with gross margin around 86%. The earnings guide is $31.00 per share, give or take a dollar, on a non-GAAP (adjusted) basis.

That implies nearly $36 billion of adjusted profit in a single quarter.

But the stock hasn't followed the numbers. The share price is around $950 as of this writing, and the 52-week high is $1,255, so the stock has given back about 24%.

A gap that wide, with results this strong, suggests investors doubt the earnings can hold.

Image source: Micron.

Three quarters of acceleration

The fiscal year opened with $13.6 billion of revenue in the first quarter. The second quarter brought $23.9 billion and the third $41.5 billion -- a period that produced just $9.3 billion a year earlier. Each revenue step has been bigger than the one before. Gross margin climbed alongside, from 57% to 75% to about 85% on an adjusted basis. And net income reached $28.2 billion in the latest quarter, up about 15-fold year over year.

Most of the demand is coming from artificial intelligence (AI) data centers. Micron's cloud memory unit generated $13.8 billion of fiscal third-quarter sales, about four times its year-ago total, and its core data center unit brought in $11.5 billion, up from $1.5 billion a year earlier. Together, that is more than half of the company's sales.

Even management has been guiding too low. In March, Micron guided the fiscal third quarter to about $33.5 billion of revenue at an 81% adjusted gross margin. The quarter finished more than $7 billion past the top of that range, at an 84.9% gross margin.

What does $50 billion assume?

The guided quarter is longer than the one it follows. Fiscal 2026 is a 53-week year, and the extra week falls in the fiscal fourth quarter (14 weeks against the usual 13).

The calendar alone accounts for about $3.5 billion of the step-up. Even stripping that out, the underlying weekly pace of revenue rises about 12%.

The rest is pricing. Not only would gross margin, at about 86%, sit about a point above the level just reported, but adjusted operating expenses are guided to only $1.65 billion. With expenses that small, most of each additional dollar of memory Micron sells falls through to profit.

As for how long that can continue, management points to its supply contracts.

"We believe our multi-year Strategic Customer Agreements will significantly enhance the durability and predictability of Micron's strong financial performance," CEO Sanjay Mehrotra said in the June earnings release.

Those strategic customer agreements are take-or-pay contracts: customers commit to set volumes for years and pay for them whether they end up needing them or not. The contracts are about how long the boom might last. The guide is about how big it has already become.

Investors are already pricing the peak

If the fiscal fourth quarter lands at the guide's midpoint, fiscal 2026 will close with about $129 billion of revenue, nearly 3.5 times fiscal 2025's total. GAAP earnings per share would land near $72, up from $7.59 the year before.

Growth like this usually commands a premium valuation. But Micron trades at about 21 times earnings. Measured against a full year at the guided quarter's pace, the stock costs about 8 times earnings. I think the second number is the more telling one. Investors are treating these profits as a cyclical peak -- and arguably with reason.

After all, this is the same business that lost $5.8 billion just three years ago. In fiscal 2023, the bottom of the last memory downturn, revenue fell by about half, to $15.5 billion.

Sure, nothing reported so far has turned. And the latest quarter finished well above the company's own forecast. The first official look at the 14-week quarter arrives on Sept. 30, when Micron reports results and should guide its first fiscal 2027 quarter.

Of course, stock prices look ahead, and memory pricing has always moved in cycles. Prices that soared this fast could fall fast, too, and the skepticism is aimed at next year, not at the quarter Micron is about to report.

Is a 24% discount on numbers like these a buying opportunity? I'd call Micron stock a hold at today's price.

If you already own shares, results like these are no reason to sell. But buying more here means believing this cycle winds down more gently than the last one did -- and I'm not there yet.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Micron Technology. The Motley Fool has a disclosure policy.

Tesla (NASDAQ:TSLA) has ended the Solar Roof, according to reporting from Electrek. The company stopped taking orders for the glass solar tiles as of Aug. 20, told its network of certified installers it will no longer supply the product, and redirected the Solar Roof page on its website to conventional solar panels.

The product Tesla unveiled in October 2016 was pitched as the reinvention of the roof -- shingles that generate power while looking better than ordinary tiles. CEO Elon Musk put a number on the ambition. "I'm confident that, let's say, within the next, I don't know, year or -- maybe even by end of year, we should be installing at a rate of 1,000 a week," Musk said on Tesla's first-quarter 2020 earnings call.

For a company that rarely retreats in public, the shutdown is unusual. So what does it say about the energy business that investors in the growth stock actually own?

Image source: Getty Images.

The economics never worked

The Solar Roof's problem seems to have been less about demand for the idea than about the economics of the product. Tesla marketed the tiles as costing less than a new roof plus traditional solar panels, but real quotes ran far higher. TechCrunch reported quotes reaching $200,000 for a single installation. And in 2021, Tesla sharply raised prices, in some cases on customers who had already signed contracts. The tiles were unique to the Solar Roof system, too, requiring custom manufacturing equipment whose cost per unit climbed as volumes disappointed.

Those volumes showed up in Tesla's own quarterly updates, which reported the solar business as one combined line (megawatts of solar deployed, panels and tiles together). By the fourth quarter of 2023, that figure had shrunk to 41 megawatts, down 59% year over year and lower for a fourth straight quarter.

Even more telling: Electrek reported, citing a source close to the program, that Tesla internally concluded the product is not financially viable. Tesla itself hasn't publicly explained the decision, and the company could say more when it next reports results, likely in October.

The energy business never needed it

For shareholders, the Solar Roof's death says very little about the electric vehicle maker's energy segment, because the segment's growth was never coming from it.

In the second quarter of 2026, Tesla's energy generation and storage revenue rose 13% year over year to about $3.1 billion, or about 11% of the company's total revenue. In its quarterly filing, the company attributed the increase to higher Megapack deployments (the utility-scale batteries), partially offset by lower Megapack prices and a decline in Powerwall deployments.

Whatever the mix, the volumes keep building. Tesla deployed 13.5 gigawatt-hours of energy storage in the second quarter, its second-best quarter ever on that measure.

Solar, meanwhile, has disappeared from Tesla's reporting altogether. The 41 megawatts deployed in the fourth quarter of 2023 turned out to be the last solar deployment figure the company has disclosed to date. The line item vanished from Tesla's first-quarter 2024 update, and 10 straight quarterly updates have now gone by without one.

Storage gets a deployment figure every quarter. Solar gets none.

That doesn't mean Tesla is done with solar. The company began manufacturing a new retrofit solar panel in 2025, according to its quarterly filing. And in July it applied for Texas tax incentives on a proposed $10.1 billion solar cell factory, with commercial operations targeted for 2029. The energy pitch still centers on storage, though, with Megapack, the newer Megablock, and a new Megafactory under construction near Houston.

The shutdown looks like discipline

The decision looks like discipline to me, and arguably overdue discipline. Tesla kept the Solar Roof alive for nearly a decade after unveiling it, through pricing resets and production experiments, while the product that actually scaled (the Megapack) drove the segment's most recent growth and helped push its revenue to about $3.1 billion a quarter.

Killing a product this publicly associated with the company's image, and with Musk's own promises for it, is not a small step. But it could free up resources for the parts of the energy business that have proven they can grow.

There is a risk worth acknowledging, though. The energy segment's growth rate has cooled to 13%, Powerwall deployments are falling, and Megapack prices are coming down. Storage is a competitive business, and it now carries the whole segment -- for a company whose stock still costs more than 150 times next year's expected earnings.

The Solar Roof was supposed to make every rooftop a Tesla product. More than six years after Musk said Tesla should be installing 1,000 a week, the company is moving on. Judged by where the energy segment's money comes from, it arguably should have moved on sooner.

Daniel Sparks has clients with positions in Tesla. The Motley Fool has positions in and recommends Tesla. The Motley Fool has a disclosure policy.

Broadcom (NASDAQ:AVGO) reports its fiscal third-quarter results after the close on Wednesday, Sept. 2. Heading into the report, the chip and software giant's trailing-12-month revenue is up 32% to $75.5 billion, extraordinary at this scale.

But the bottom line is moving far faster. Net income over the same period climbed 127% to $29.3 billion -- nearly four times the pace of revenue growth.

About 39 cents of every revenue dollar now lands as profit, up from about 23 cents a year earlier.

Broadcom's disclosures show where that spread comes from, and what Wednesday can and can't settle.

Image source: Getty Images.

Is the profit growth overstated?

Part of the 127% is inherited. In the fiscal third quarter of 2024, Broadcom reported a rare $1.9 billion net loss under generally accepted accounting principles (GAAP). The cause was a one-time $4.5 billion noncash tax charge tied to an intellectual property transfer to the United States. That loss sits in the year-ago window and flatters the trailing growth rate.

Strip that charge out, and profit still grew about twice as fast as revenue.

The most recent quarter needed no such help. In the fiscal second quarter, which ended May 3, revenue rose 48% year over year to $22.2 billion while net income climbed 88% to $9.3 billion.

The trend is the stronger evidence, I think. Broadcom's GAAP operating margin has expanded from about 39% of revenue in the year-ago quarter to 44% in this year's fiscal first quarter to nearly 49% in fiscal Q2. That is almost 10 percentage points in a year.

Costs are barely moving

The spread comes from the expense lines. While fiscal Q2 revenue jumped 48%, total operating expenses rose about 6% year over year to $4.6 billion. Research and development spending grew 11%. Selling, general and administrative costs fell. And the noncash amortization from past deals (about $2 billion a quarter) didn't grow at all. A charge that took more than 13% of revenue a year earlier now takes about 9%.

Both segments are contributing. Semiconductor operating income nearly doubled year over year in fiscal Q2, lifting its operating margin from 57% to about 62% on 79% revenue growth.

And infrastructure software (built around VMware) turned 9% revenue growth into 13% profit growth because its costs fell. Its operating margin now sits near 79%, up from about 76% a year earlier.

Notably, the extra profit isn't coming from richer margins on each product sold.

In fact, custom artificial intelligence (AI) accelerators and networking brought in $10.8 billion in fiscal Q2, CEO Hock Tan said, up 143% year over year -- nearly three-quarters of the chip segment's revenue. On the June 3 earnings call, then-chief financial officer Kirsten Spears said consolidated gross margin should decline in fiscal Q3 as AI grows as a share of sales. That is a product-mix effect, she said, not a structural change in chip margins.

In short, the profit surge comes from selling much more without spending much more.

Wednesday will test the spread at $29.4 billion

"In Q3 we expect consolidated revenue growth to increase 84% year-over-year to $29.4 billion, with non-GAAP operating margin stable at 67% reflecting our strong operating leverage," Spears said in the company's June 3 earnings release.

Non-GAAP (adjusted) results strip out items like stock-based compensation and deal-related amortization. Even on that friendlier basis, the guide asks a lot: costs stay in check while revenue steps up by about $7 billion from the quarter just reported. Tan expects $16 billion of the quarter's revenue to come from AI, up more than 200% year over year.

Wednesday can settle that much. Does the cost discipline hold at $29.4 billion?

However, one report can't settle the longer arc. Gross margin pressure from the AI mix could eventually outrun the cost discipline. Expenses may not stay near $4.6 billion forever as its AI revenue keeps doubling. Those answers play out over years.

Meanwhile, at around $369 as of this writing, down about 25% from its 52-week high of $495, the stock trades at about 60 times earnings -- arguably steep, even for growth this fast. But the earnings under that price aren't standing still. Trailing earnings per share more than doubled in a year. If the spread between profit growth and revenue growth holds, that price-to-earnings ratio shrinks quickly.

Ultimately, the spread is disclosed line by line, it has widened for a year, and management's guide calls for a stable non-GAAP operating margin. But at about 60 times earnings, it is also the thing shareholders are paying for. If costs start climbing alongside revenue, profit growth falls back toward revenue growth, and today's price-to-earnings ratio gets hard to defend.

If I owned shares, I'd hold them through Wednesday's report. But I wouldn't buy at this price, and for now I'd call the stock a hold.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Broadcom. The Motley Fool has a disclosure policy.

Shares of artificial intelligence (AI) cloud infrastructure provider CoreWeave (NASDAQ:CRWV) trade around $82 as of this writing, down about 47% from their 52-week high. But the business keeps growing at an extraordinary pace. Second-quarter revenue rose 112% year over year to about $2.6 billion, and the company's revenue backlog reached about $104 billion (a figure that excludes more than $25 billion of new commitments added early in the third quarter).

The cost of financing that growth is climbing even faster. CoreWeave's interest expense was $640 million in the second quarter -- 2.4 times the $267 million it recorded a year earlier.

And the bond market isn't helping. The 30-year Treasury yield has closed above 5% on 55 days since the start of January, the most closes above that mark in any year since 2006.

To be fair, CoreWeave doesn't borrow at 30-year maturities, and its debt doesn't price anywhere near Treasury yields. But in a bond market like that, I think borrowed money could stay expensive for a while. And CoreWeave needs a lot more of it.

Image source: The Motley Fool.

More debt, cheaper debt

CoreWeave's interest expense has climbed every quarter for the past year, from $267 million in the second quarter of 2025 to $311 million, $388 million, $536 million, and now $640 million. The driver is the balance, not the rate. Total debt reached about $35 billion as of June 30, up from about $21 billion at the end of 2025. That is a lot of debt for a company that completed its initial public offering (IPO) less than 18 months ago.

The rate, in fact, has moved in CoreWeave's favor.

"Over the past year, we have reduced our weighted average cost of debt by almost 300 basis points, representing approximately $1.1 billion of annualized interest saving based on our end of Q2 debt load," chief financial officer Nitin Agrawal said in the company's second-quarter earnings call.

Those savings are real. Low-rate convertible notes and bigger credit facilities have replaced some of the expensive borrowing from earlier in its cloud build-out. The bill more than doubled anyway, because the balance grew far faster than the rate fell.

How expensive is all that debt?

CoreWeave's latest quarterly filing lists effective interest rates for its borrowings, and the range is wide: 2% on its convertible notes, mostly 9% to 11% on its term loans and senior notes, and 15% on its oldest term loan.

Weight each rate by its balance, and the blended cost works out to about 8.4%. On a balance this size, each percentage point costs more than $350 million a year.

New money is still arriving above that average. CoreWeave issued senior notes at 9.75% in April and 9.625% in June, plus euro-denominated notes at 8.5% -- effective rates of 9% to 10% once fees and discounts are folded in.

And the $2.6 billion term loan facility it added in August prices at 5.5 percentage points over the benchmark short-term lending rate.

The broader bond market offers little sign of relief coming. The 30-year yield touched 5.34% in mid-August, its highest since 2007, and sits at about 5.27% as of this writing.

The bill keeps climbing

Management expects third-quarter interest expense of $860 million to $940 million, a step up of about 41% at the midpoint, against $200 million to $260 million of adjusted operating income.

Operating profit was already far behind. Adjusted operating income was $128 million in the second quarter, down from $200 million a year earlier even as revenue more than doubled.

But the maturity schedule, at least, looks manageable. About $4.4 billion of principal comes due through year-end and $6.2 billion in 2027, while nearly $15 billion isn't due until after 2030. Refinancing isn't the near-term problem, in my opinion. New borrowing is.

That's because the spending isn't slowing down. CoreWeave spent $16.1 billion on capital expenditures in the first half, and its full-year guidance of $35 billion to $39 billion implies roughly $19 billion to $23 billion more in the second half.

Against that, CoreWeave held about $5.5 billion of cash at the end of June -- arguably not much next to spending plans that size.

Ultimately, the second quarter showed a company getting better at borrowing while needing more of it than ever. Sure, the spending builds the AI infrastructure behind the $104 billion of contracted revenue already on the books. But the interest bill is climbing faster than the operating profit that is supposed to carry it.

That gap is the number I'd watch. Interest expense ran about $500 million ahead of adjusted operating income in the second quarter, and guidance implies the distance widens in the third.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

In February, Nebius Group (NASDAQ:NBIS) told investors to expect more than 3 gigawatts (GW) of contracted power by the end of 2026. In May, the target became more than 4 GW. In August, alongside second-quarter results, the artificial intelligence (AI) cloud provider raised it again, to 5 GW.

Contracted power is the raw material of Nebius' business. It's the electricity capacity the company has secured for data centers that rent out graphics processing units (GPUs). Three raises in six months say the company keeps finding more of it, faster than it expected. That escalation has my attention.

Nebius carries a market value of about $56 billion, with shares just above $200 as of this writing. Its revenue over the past 12 months was about $1.4 billion.

What does 5 GW of power have to earn to justify a price like that?

Image source: Getty Images.

Three raises in six months

The escalation is the company's own, laid out in its August shareholder letter. A year ago the target was more than 1 GW. It became more than 2.5 GW in November, more than 3 GW in February, more than 4 GW in May, and 5 GW now.

The business underneath is scaling almost as fast.

Second-quarter revenue grew 454% year over year to $582.3 million, with the core AI cloud business contributing about 98% of the total. Annualized run-rate revenue reached $3.0 billion at the end of June -- up 598% year over year, and up 56% from $1.9 billion just three months earlier.

Profitability is arriving with scale, too. The AI cloud business produced an adjusted EBITDA margin of 50% in the quarter, up from 45% in the first quarter and 24% in the fourth quarter of 2025. Companywide, adjusted EBITDA swung to a positive $236 million from a loss a year earlier. (EBITDA is earnings before interest, taxes, depreciation, and amortization.)

Each megawatt is worth more than it used to be

Nebius closed four landmark deals in the second quarter, averaging more than $1 billion in total contract value, with AI developers Reflection and Cohere among the customers. The company said those deals carry annual contract value of $20 million to $25 million per megawatt.

That is up from about $12 million per megawatt on its 2026 base of business. And early third-quarter short-term capacity deals are pricing above $40 million per megawatt. All told, the company counts $40 billion in customer commitments.

So what could the full target earn? If Nebius eventually deployed all 5 GW (5,000 megawatts) and sold it at even the older $12 million rate, the implied revenue would be about $60 billion a year. At the second quarter's deal prices, the figure could be far higher.

Against a $56 billion market value, that is the bull case in one calculation.

The capacity can't arrive all at once

However, contracted power is not deployed power, and deployed power is what generates revenue. Nebius says it plans to bring more than 1 GW of capacity online per year starting in 2027. At that pace, turning 5 GW of contracts into running data centers is a project that can stretch toward the end of the decade.

The spending, meanwhile, is immediate. Nebius spent $5.7 billion on property and equipment in the second quarter alone, and management expects $20 billion to $25 billion of capital expenditures for the full year. The company still runs at an operating loss ($176 million in the second quarter), and its quarterly depreciation and interest costs are climbing fast as the build-out compounds.

Of course, customer prepayments help. The company expects more than $9 billion of them in 2026, and prepayments covered 50% to 60% of the capital spending tied to recent deals. The capital markets supply much of the rest, including a convertible note sale that closed in August with about $5.75 billion of gross proceeds. But the model still consumes enormous amounts of money before it returns any.

Even so, management reaffirmed its full-year guidance, including revenue of $3 billion to $3.4 billion and a year-end run-rate target of $7 billion to $9 billion. Hit the top of that range, and today's market value works out to about six times year-end run-rate revenue. For growth like this, that's arguably a fair price. But it leaves no room for deployment delays, softer GPU pricing, or a pause in AI spending.

I believe the contracts will become revenue -- the customers are signed, and the price per megawatt keeps rising. Still, most of those megawatts won't produce a dollar until 2027 or later, and the building costs land now. I'm watching Nebius closely, but I'm not buying shares yet.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

For most of its life, Palantir Technologies (NASDAQ:PLTR) has been a government software company.

Even now, with shares near $186 as of this writing, the artificial intelligence (AI) software specialist's biggest business at home is still the one built on government contracts. U.S. government revenue was $809 million in the second quarter, against $764 million for U.S. commercial.

But the gap is down to $45 million, and the two lines are not growing at the same speed. Last quarter, the U.S. commercial side grew 149% year over year. The government side grew 90%.

Roll those curves forward and they cross almost immediately. My prediction: U.S. commercial revenue passes U.S. government revenue in the third quarter of 2026 (the quarter that ends this month), and well before 2027 even if the timing slips.

That crossover would be more than a milestone, because when the faster-growing business becomes the bigger one, the whole company's growth rate starts bending toward it.

Image source: Getty Images.

Two lines, $45 million apart

The second quarter is the closest the race has been all year. U.S. commercial revenue reached $764 million, up 149% year over year and 28% from the first quarter. U.S. government revenue reached $809 million, up 90% year over year and 18% sequentially.

Both rates are extraordinary at this scale. The government side's 90% alone would be a standout result for most software companies. The commercial side has simply been faster, and consistently so. In the first quarter, the same race ran 133% against 84%.

Notably, though, the quarter-to-quarter race is tighter. In the first quarter, the government business grew faster sequentially, 21% versus 18%. That flip came from a customer program moving out of the commercial segment and into the government one. Management said commercial growth would have reached 143% year over year without the transition.

Still, the year-over-year gap is the durable pattern -- 49 percentage points in the first quarter, 59 in the second.

Management credits the surge to demand for what CEO Alex Karp calls "AI sovereignty," meaning customers want control over their own operations, data, and decisions.

When do the lines cross?

Take the second quarter's sequential rates and roll them one quarter forward. Commercial revenue growing 28% from $764 million lands at about $980 million. Government revenue growing 18% from $809 million lands at about $955 million. On that math, the lines cross in the third quarter, the period ending Sept. 30.

And the bar, I think, is lower than it sounds. After all, closing a $45 million gap from a $764 million base only takes a sequential growth edge of about 7 percentage points. The commercial side's edge in the second quarter was 10 points.

Management's own numbers lean the same way. Palantir raised its full-year U.S. commercial revenue guidance to more than $3.424 billion, which implies growth of at least 134%. It raised its adjusted free cash flow outlook, too, to between $4.5 billion and $4.7 billion for the year.

Could the timing slip a quarter? Of course. One large government deal landing in September may hold the old order for another period.

But for the crossover to miss 2026 entirely, commercial's sequential growth would need to slow to about 21% for two straight quarters while government held its 18% pace. And a slowdown that sustained seems unlikely: U.S. commercial remaining deal value (the value left on signed contracts, assuming customers exercise every option and cancel none) climbed 124% year over year to $6.2 billion.

A bigger commercial business lifts the whole growth rate

The order of the two revenue lines matters because the company's blended growth rate is a weighted average, and the weights are about to flip.

Today, the slower-growing government business carries more weight in U.S. revenue. Once commercial is the bigger line, its 149% growth counts for more than the government side's 90%, and the blended rate drifts higher before anyone signs an extra contract.

To be fair, U.S. revenue growth accelerated from 104% in the first quarter to 115% in the second, but that was mostly both segments speeding up. The shifting weights added only a fraction of a point. That contribution grows as commercial's share of the revenue base rises.

A majority-commercial Palantir would also get judged the way a commercial growth stock is judged -- on the size of its market, not on federal budget cycles.

Investors won't have to wait long to see whether my prediction is right. Palantir's third-quarter report, likely in early November, will print both numbers. If commercial lands on top, the crossover arrives with a quarter to spare.

If the sequential steps flip again the way they did in the first quarter, the date may slide one period out. But with commercial growing 59 points faster year over year, I don't see it slipping past 2026.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Palantir Technologies. The Motley Fool has a disclosure policy.

Advanced Micro Devices (NASDAQ:AMD), Cisco (NASDAQ:CSCO) and HUMAIN, the artificial intelligence (AI) company backed by Saudi Arabia's sovereign wealth fund, said Monday that the AI infrastructure they have been building in the Kingdom is live. Systems built on AMD Instinct MI355X graphics processing units (GPUs) are serving HUMAIN's customers, selling computing power for training AI models and for running them once trained.

Of course, chip partnerships get announced almost weekly in this market. But capacity that is switched on and serving customers is much rarer. And for AMD, whose AI story rests heavily on multiyear commitments that stretch far into the future, I think the difference between the two is what Monday's news is really about.

"Together, we are building an open, high-performance AI platform that is serving customers today and will scale significantly in the coming years," AMD CEO Lisa Su said in the announcement.

Image source: AMD.

Serving customers today

Monday's milestone grew out of a partnership that began in May 2025, when AMD and HUMAIN agreed to build a network of AI computing centers stretching from Saudi Arabia to the United States. Cisco joined, and the three companies formed a joint venture to build AI infrastructure in the Kingdom.

The systems now running pair those Instinct GPUs with AMD EPYC processors and Cisco networking. HUMAIN sells the computing they produce as a service, to customers in Saudi Arabia and beyond.

And the next step is already planned. The partners plan to deploy up to 250 megawatts (MW) of additional AI infrastructure, built on AMD's coming Instinct MI400 series chips. Deployment is planned to begin in 2027, and the joint venture remains on track to reach up to 1 gigawatt (1,000 megawatts) by 2030.

Sizing the build-out

What is a megawatt of AI infrastructure worth? The original agreement offers a rough benchmark. When AMD and HUMAIN formed their collaboration, they said the parties would invest up to $10 billion to deploy 500 megawatts of AI compute capacity over five years.

That works out to about $20 million per megawatt. At that rate, the planned phase of up to 250 MW implies an investment of up to about $5 billion. And the full gigawatt would be about four times that -- roughly $20 billion.

Of course, AMD doesn't collect all of that money. Data centers, power systems and networking take their share. But AMD's GPUs and processors are the heart of what those dollars buy.

For scale, consider what AMD's data center business already produces. Data center segment revenue reached $6.7 billion in the second quarter, up 107% from about $3.2 billion a year earlier. The segment's share of total revenue keeps expanding, too -- 58% last quarter, up from about 42% in the year-ago quarter. Companywide revenue grew 50% to a record $11.5 billion, and management guided third-quarter revenue to about $13 billion, up about 41% year over year.

Against numbers moving that fast, a build-out spread over several years can't transform AMD's income statement on its own. What it does instead is show the model working end to end -- chips shipped, systems stood up, customers served.

After all, so much of AMD's AI opportunity still sits in the future, in announced commitments that have yet to become revenue. The Saudi joint venture's gigawatt is one slice of a much larger pipeline.

Deployments like this one are how commitments turn into sales. This sovereign-scale example is now up and running, and the next phase already has a size and a start date.

The stock already assumes a big ramp

However, investors have already given AMD plenty of credit for what's coming. Shares trade near $467 as of this writing, at about 30 times next year's expected earnings. For a business that just doubled its data center revenue, that valuation is arguably fair. But it assumes the ramps keep arriving on schedule, generation after generation, without the margin stumbles that big hardware transitions can bring.

So I'd treat Monday's news as proof of execution, not a reason to rush in. Revenue from this first phase won't move a $13 billion quarter much. The proof matters more: AMD and its partners took a sovereign AI project from announcement to running systems, and that makes the rest of its order book easier to believe.

Ultimately, though, the build-out is still in its early innings, and the stock already prices in a smooth ramp. I'd wait for a better entry point before buying shares.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices and Cisco Systems. The Motley Fool has a disclosure policy.

A company that grows revenue 92% and still loses half a billion dollars is doing two very different things at once. SpaceX (NASDAQ:SPCX) did exactly that in the second quarter.

The rocket maker booked $7.8 billion of revenue across its three businesses, up 92% from a year earlier, and still lost $541 million.

The segment tables explain how. One business, the connectivity segment built around Starlink, earned a $1.66 billion operating profit. The other two, the original space business and the artificial intelligence (AI) unit, lost a combined $1.8 billion. And the AI unit absorbed $15.8 billion of the quarter's $18.4 billion in capital spending.

In other words, one segment pays the bills while another spends at a pace the earner can't come close to covering. That split, more than the growth, is what the stock's nearly $1.9 trillion valuation turns on, I'd argue.

Image source: Getty Images.

The profitable segment is bigger than Starlink

The headline profit belongs to Starlink, but the segment earning it is broader than the consumer satellite internet service.

SpaceX's connectivity segment combines consumer Starlink ($2.5 billion of second-quarter revenue, up 44% year over year) with an enterprise and government business that grew 108% to $1.8 billion. The latter got help from new airline agreements and more than $6 billion in multi-year U.S. government contracts for Starshield, the company's secure satellite network for government customers.

Add it up, and connectivity revenue rose 66% year over year to $4.3 billion, while the segment's operating profit climbed 79% to $1.66 billion.

Notably, the growth is coming from subscribers. Starlink ended June with 12 million subscribers, double a year earlier, while average revenue per user held at $66 a month for a second straight quarter, down from $85 a year ago.

What the segment doesn't spend matters just as much. Connectivity's capital expenditures were $1.4 billion in the quarter, less than its operating profit. This is the one SpaceX business that funds itself.

The $15.8 billion quarter

The AI unit is the opposite case. The segment spent $15.8 billion on capital expenditures in three months, 86% of the company's total. It spent $7.7 billion in the first quarter and $749 million in the year-ago period. That is a roughly 20-fold increase in four quarters, with the money going into the Colossus II data center build-out that pushed the company's compute capacity to 1.4 gigawatts, up from 0.4 a year ago.

The spending is buying growth, to be fair. AI revenue more than tripled year over year to $2.6 billion, driven by cloud computing agreements ($14.1 billion in contracted sales signed during the quarter), and the segment's operating loss narrowed to $1.3 billion from $2.5 billion in the first quarter. But the segment still loses money.

And the space segment added a $542 million operating loss of its own on $962 million of revenue, up 29%, as the company accelerated research and development spending on Starship -- spending it plans to extend with a Louisiana launch complex, announced Aug. 25, that could cost up to $100 billion.

But who pays for all this? Not the profitable segment, at least not alone. Connectivity's $1.66 billion quarterly operating profit covers about a tenth of the AI unit's quarterly capital bill.

The rest comes from the balance sheet. SpaceX ended June with $100 billion of cash, cash equivalents, and marketable securities, built largely from about $85.7 billion in net proceeds from its June initial public offering (IPO). A $25 billion bond sale mostly refinanced a bridge loan, at a weighted average rate of about 5.9%.

What is the stock priced for?

Shares sit near $142 as of this writing, close to the $135 offering price, and SpaceX's market value is roughly $1.9 trillion -- about 60 times sales, annualizing the second quarter.

A sales multiple like that is not a bet on the business that already works. Starlink's economics are impressive, and arguably proven. But a segment producing about $6.6 billion in annualized operating profit doesn't support a $1.9 trillion price on its own.

The valuation is mostly a claim about the money-losing unit. It assumes today's $15.8 billion quarters convert into an AI infrastructure business big enough to justify them.

Maybe they will. The AI segment's revenue is scaling fast, its losses are narrowing, and on segment adjusted EBITDA (non-GAAP) the unit even earned $1.1 billion in the quarter. But shareholders are paying for that conversion up front, while the segment that reliably earns money could, by itself, justify only a fraction of the price.

Ultimately, the second quarter answered which business pays for the others. Starlink's segment pays, the AI unit spends, and the stock trades on the spender's future. The profitable business is excellent. But at about 60 times annualized sales, I'd stay on the sidelines for now.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

Dell Technologies (NYSE:DELL) reports its fiscal 2027 second-quarter results on Tuesday, Sept. 1, with a conference call set for 3:30 p.m. Central time. One line in that report interests me more than the revenue number, the earnings number, or the size of the artificial intelligence (AI) order backlog. It's the operating margin of Dell's infrastructure solutions group, the segment that builds the servers powering the AI build-out.

That's because memory prices have been climbing across the chip industry, and the companies that design AI chips have spent recent weeks describing what those costs are doing to their own margins.

Dell sits further down the same supply chain. It buys memory in huge volumes and assembles it into finished servers. If rising component costs are going to squeeze anyone's margins, the assembler is where the squeeze should show up first.

Tuesday's report gives investors their first good look at the answer.

Image source: Getty Images.

The margin already stepped down once

Dell's infrastructure solutions group posted record first-quarter revenue of $29 billion, up 181% year over year. AI-optimized servers (machines built around graphics processing units and high-end memory) drove it, contributing $16.1 billion of revenue, nearly double the fiscal fourth quarter's $9 billion. And the company booked $24.4 billion of new AI server orders during the quarter. The rest of the segment grew, too -- traditional servers and networking revenue rose 92% year over year to $8.5 billion, while storage grew 8% to $4.3 billion.

The profitability was more complicated. Segment operating income was $3.1 billion, up 206% year over year, and the segment's operating margin of 10.5% was actually higher than the year-ago quarter's. However, it was down sharply from 14.8% in the fiscal fourth quarter.

Part of that step-down is seasonal. The segment's margin also fell sharply between the same two fiscal quarters a year earlier, from about 18% to under 10%, back when AI servers were less than a fifth of the segment. Much of the rest is mix, not memory. AI servers carry much thinner margins than Dell's traditional servers and storage, and chief financial officer David Kennedy said the AI server business is running in line with its target of a mid-single-digit operating margin.

In other words, when a low-margin product line grows from a sliver of the segment into more than half of it, the blended margin falls even if nothing is going wrong.

That's why Tuesday's number is so useful. The mix effect is known, and management has set the bar itself: Kennedy guided to a sequential improvement in the segment's operating margin this quarter. A margin that rises from the first quarter's 10.5% says Dell is passing its higher memory costs through. One that merely holds, or slips, says some of the bill is landing on Dell.

Management is already repricing

Dell hasn't been shy about naming the pressure. On the company's fiscal first-quarter earnings call in late May, chief operating officer Jeff Clarke described an inflationary environment across memory and other components, and said the company has been adjusting prices frequently in response.

Clarke also named notable commodity constraints, particularly in DRAM and NAND (the two main types of memory chips), as part of a challenging demand and supply environment.

The demand side looks fine. Dell guided second-quarter revenue to $44 billion to $45 billion, up about 50% at the midpoint. The infrastructure segment is expected to grow roughly 75%, including about $15.5 billion of AI server revenue. And adjusted earnings per share guidance of $4.80, plus or minus $0.10, implies growth of more than 100% year over year.

Growth, then, isn't in doubt on Tuesday. What the report settles is how much of it Dell keeps while one of its most important inputs gets more expensive by the quarter.

What would a good answer look like?

I'd watch three things. First is the segment margin itself. A number above 10.5% says pricing power is holding, and one at or below it says it isn't. Second is the companywide gross margin, which fell to 17.8% in the first quarter from 21.1% a year earlier, largely on the AI mix. Another sharp drop there suggests costs are outrunning prices. And third is any updated commentary on memory, because Dell's guidance for the rest of the year assumes the repricing keeps working.

Shares trade near $461 as of this writing, at about 26 times the adjusted earnings management has guided to for this fiscal year -- arguably a full price for a hardware business, and one that assumes the AI growth stays profitable.

I think Dell probably passes the test. Management saw the memory problem early and started repricing months ago. But the margin line is the test, and the answer arrives Tuesday. I see no reason to guess a day early.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

Here are two numbers that shouldn't normally sit next to each other. Oracle (NYSE:ORCL) ended fiscal 2026 with $638 billion of remaining performance obligations, which is contracted work its customers have signed up for that hasn't yet become revenue. The company's market value, meanwhile, is about $430 billion, with the stock near $150 as of this writing -- down about 57% from its high of $345.72.

In other words, Oracle's stock is worth roughly $200 billion less than the revenue its customers have already contracted to hand over. And even adding the company's roughly $98 billion of net debt to the price, the market values the whole business about $110 billion short of the backlog.

When Oracle revealed the $638 billion figure in June, alongside its fiscal 2026 fourth-quarter results (the fiscal year ended May 31), the company's market value stood near $580 billion. The stock has slid since, and the gap has only widened.

What is the market saying with a price like this? I think the answer comes down to two of Oracle's own disclosures. One is how slowly the backlog converts. The other is what serving it costs.

Image source: Getty Images.

A backlog that converts slowly

The backlog itself is astonishing. Remaining performance obligations grew 363% year over year and rose $85 billion in the fiscal fourth quarter alone, driven by demand for cloud infrastructure to train and run artificial intelligence (AI) models.

And the revenue behind it is showing up: Oracle's cloud infrastructure revenue grew 55%, 68%, 84%, and then 93% year over year across fiscal 2026's four quarters. The business accelerated all year.

But contracted is not the same as soon. Management said on the June earnings call that it expects 12% of the backlog to be recognized as revenue over the next 12 months, and another 34% between 13 and 36 months. That works out to about $77 billion arriving within a year, and roughly $290 billion inside three years. More than half of the total sits further out than that.

For context, Oracle confirmed guidance for about $90 billion of total revenue in fiscal 2027, up from $67.4 billion in fiscal 2026, with fiscal first-quarter revenue expected to grow 27% to 29%. The backlog supports years of growth like that. It just can't be pulled forward.

Serving it costs real money

The second disclosure is what those contracts require. Oracle generated a record $32 billion of operating cash flow in fiscal 2026, up 54%. It spent all of that on data centers, and then some. Free cash flow came in at negative $23.7 billion.

So the company raised $43 billion in debt and $5 billion in equity during the fiscal year, and it expects to raise approximately $40 billion more in fiscal 2027 through a combination of debt and equity, including a previously announced $20 billion at-the-market stock program.

To the company's credit, its customers are helping carry the load. Oracle said the prepaid and customer-supplied hardware portions of its large AI contracts now total $75 billion, which "substantially reduces the amount of capital Oracle must raise to build out our AI datacenters."

Still, the shape of the business has changed. A company that used to throw off cash now consumes it. And each contracted dollar of AI infrastructure revenue arrives with heavy costs attached -- the graphics processing units, the buildings, and the electricity, plus the interest on the borrowing that funds them.

So is the stock cheap?

A backlog bigger than the market cap sounds like an obvious bargain. It isn't, necessarily. Backlog is revenue, not profit, and the market's judgment is about what that revenue will be worth after Oracle pays for the infrastructure that produces it.

The stock trades at about 26 times earnings, and at about 19 times the $8.05 of adjusted earnings per share management has guided to for fiscal 2027. For a company guiding revenue up 34% this fiscal year, that isn't an expensive price. Arguably, it reflects doubt about the margins on AI contracts and about the years of heavy borrowing and stock sales still ahead.

Ultimately, I'd stay on the sidelines here. The contracted demand is enormous, but the economics of serving it are still being proven, and the balance sheet is absorbing tens of billions of dollars of strain in the meantime.

What nobody can see yet is how much profit all that contracted revenue leaves behind once the data centers are paid for. I'd want to see some of it first.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Oracle. The Motley Fool has a disclosure policy.

Alphabet (NASDAQ:GOOG)(NASDAQ:GOOGL) said something striking on its second-quarter earnings call in July. Even after committing to as much as $205 billion of capital spending this year, the company still can't build artificial intelligence (AI) computing capacity as fast as customers want it.

"[W]e continue to be supply constrained -- a sign of momentum and rapid adoption," CEO Sundar Pichai said in his remarks on the quarter.

Yet Alphabet has agreed to hand multi-gigawatt blocks of that scarce capacity to a fast-growing outside customer: Anthropic, the AI company behind the Claude models.

And a look at Alphabet's underlying business performance shows why the company is racing to sell its capacity to major customers like Anthropic -- even if it's scarce.

Image source: Alphabet Inc.

Selling scarce capacity is a great business

Google Cloud, the segment that sells cloud computing to outside customers, grew revenue 82% year over year to $24.8 billion in the second quarter. That was up from 63% growth in the first quarter.

The profit is growing even faster than the revenue. Google Cloud's operating income more than tripled year over year, from $2.8 billion to $8.8 billion -- after reaching $6.6 billion in the first quarter. The segment's operating margin came in at about 36%, versus about 21% in the year-ago quarter and 33% in the first quarter of this year.

And the contracted work keeps piling up. Pichai said cloud backlog (future revenue from signed contracts) grew to about $514 billion in the second quarter, up from $462 billion at the end of the first quarter. That backlog is now more than four times the revenue Alphabet's entire business produced last quarter.

How much of it is Anthropic?

Alphabet doesn't break out the number, but the disclosed pieces -- even if they lack financial details -- are big. Last October, Anthropic agreed to expand its use of Google Cloud in a deal giving it access to up to 1 million of Google's tensor processing units (TPUs), the AI chips Google designs in-house, with well over a gigawatt of capacity coming online in 2026. Google Cloud said the agreement was worth tens of billions of dollars.

This spring, the relationship got much bigger. In early April, Anthropic secured multiple gigawatts of next-generation TPU capacity from Google and chip partner Broadcom, coming online starting in 2027 -- about 5 gigawatts in all, CNBC reported. Anthropic will access that capacity through Broadcom, according to a Broadcom securities filing. Weeks later, Google agreed to invest up to $40 billion in Anthropic itself, putting in $10 billion right away with as much as $30 billion more tied to performance milestones.

Worth noting from that Broadcom filing, though, is that Anthropic's use of the expanded capacity "is dependent on Anthropic's continued commercial success."

That is the honest risk in this arrangement.

To be fair, Anthropic said in April that its run rate revenue (its recent revenue pace, annualized) had surpassed $30 billion, up from about $9 billion at the end of 2025. Growth like that is extraordinary. But it means a meaningful slice of Alphabet's contracted future rests on one young AI developer growing into its commitments, and Alphabet is now an investor in that developer on top of being its supplier.

The build-out still has to be paid for

Of course, Alphabet has to build all of this capacity before anyone can rent it. The company raised its 2026 capital expenditures guidance in July to $195 billion to $205 billion.

In the second quarter, capital spending of $44.9 billion exceeded the $39.1 billion of cash its operations produced. And the funding has gone well beyond cash on hand. Alphabet collected $49.6 billion from stock sales in June and issued senior notes (a form of debt) for another $20.3 billion of proceeds during the quarter.

In other words, the company is financing enormous capacity ahead of the revenue it will carry, and pre-selling chunks of it profitably.

What's in it for Alphabet? Probably more of the incredible momentum it's already seeing: Faster cloud revenue growth, a segment margin up from about 21% to about 36% in a year, and a $514 billion pile of signed contracts.

So, there's a lot to like here. The supply constraint Pichai described is another way of saying Alphabet has pricing power, and the Anthropic agreements convert that scarcity into contracted revenue years into the future -- something that should help an already thriving cloud business over the long haul.

And shares trade near $339 as of this writing, at about 23 times next year's expected earnings, which is arguably a reasonable price for a company growing total revenue by 24% (with an explosive cloud business underneath).

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet and Broadcom. The Motley Fool has a disclosure policy.

September begins tomorrow, and for shareholders of Advanced Micro Devices (NASDAQ:AMD), the month arrives with a track record worth knowing. Over the past decade, September has been the chipmaker's least friendly month by a wide margin: the stock fell in eight of the past 10 Septembers, with a median decline of about 5%.

A pattern built on 10 data points deserves skepticism. But it also deserves an honest look, because the numbers say the weakness isn't entirely the market's fault.

Here's the full record, and what I'd take from it, with shares around $467 as of this writing.

Image source: Getty Images.

Ten Septembers, eight declines

AMD fell in the Septembers of 2016 (down about 7%), 2017 (down 2%), 2019 (down 8%), 2020 (down 10%), 2021 (down 7%), 2022 (down 25%), 2023 (down 3%) and 2025 (down less than 1%). It rose just twice, jumping 23% in 2018 and 10% in 2024.

Those figures come from month-end closing prices, and AMD pays no dividend, so the price change is the whole return.

For perspective, September is not what a normal AMD month looks like. Across all 120 months of that decade, the stock's median month was a gain of about 1.6%, and 63 of the 120 finished higher. September's median was a 5% decline.

The worst of the bunch, 2022's 25% plunge, landed in a year when the stock lost more than half its value. And the best, 2018's 23% jump, capped a run that had the stock up about 200% for that year by the end of September. In other words, the extremes came in years when everything about the stock was extreme. September just happened to be along for the ride.

The market shares only part of the blame

A fair test asks how the broader market did over the same 10 months. The S&P 500 (SNPINDEX:^GSPC) fell in five of those Septembers and rose in the other five, and its median September move was roughly flat. So the index's September record is a coin flip, while AMD's is eight down months out of 10.

The overlap tells the sharper story. In the five Septembers when the market fell, AMD fell every time, and in four of the five it fell harder -- including that 25% drop in 2022, when the index lost 9%. That part is no mystery. AMD is a volatile growth stock, and it tends to amplify whatever the market is doing.

The part the market can't explain is the other half. In the five Septembers when the S&P 500 rose, AMD still declined three times. A stock that usually exaggerates the market's moves should have had more good Septembers than that.

The pattern has no mechanism behind it

So does September carry something specific for AMD? The record leans that way, but a cause is harder to find. AMD's calendar puts no earnings report in September, and no major recurring product launch lands there either. After a decade covering tech stocks, I've seen narratives built on thinner evidence than this, and most of them probably dissolve the moment you ask what would cause the pattern to repeat.

And 10 observations are just that -- 10. Strip out the 2022 plunge, and the typical September decline shrinks to less than 3%. The two up years, 2018 and 2024, show the stock is perfectly capable of a strong September when the business gives it a reason.

Notably, the month's extremes are nearly symmetrical, too. The best September (up 23%) was almost as large as the worst (down 25%). That looks like volatility, not a curse.

Ultimately, I'd read the September record as context for expectations, never as a trading signal. It says a volatile chip stock has often had a rough early fall, especially when the market wobbled. It says nothing about what the company's data center ramp or the demand for artificial intelligence (AI) computing will do over the next 30 days, and those are what move this stock.

So history gives no reason to sell a stock you own for September, and no reason to delay buying one you want. What the record does earn is patience: with a stock that has fallen in eight of the past 10 Septembers, a weak month ahead shouldn't surprise anyone -- and it wouldn't say anything about the business.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices. The Motley Fool has a disclosure policy.

Microsoft (NASDAQ:MSFT) will send its shareholders their next dividend payment ($0.91 per share, declared on June 10) on Sept. 10. For most dividend stocks, that would be the least interesting data point of the month, because the payment arrives right at the time of year when the board has historically announced its annual increase. The latter is likely what dividend investors will care about more.

The software giant has raised its dividend every year for more than a decade -- 16 consecutive annual increases, all announced in September. The most recent came on Sept. 15, 2025, when the board brought the quarterly payment from $0.83 to $0.91 -- an increase of just under 10%.

That track record is steady enough to support a real forecast. So let's make one.

Image source: The Motley Fool.

A raise in the same band year after year

Microsoft's dividend growth has been strikingly stable.

The last six annual increases were 9.8%, 10.7%, 9.7%, 10.3%, 10.7%, and 9.6% -- all within a band between 9.6% and 10.7%. And a longer window barely changes the story. The quarterly payment has grown from $0.36 at the end of 2015 to $0.91 at the end of 2025, which equates to about a 9.7% annual compound rate over that decade.

Apply that band to the current payment of $0.91, and the next quarterly dividend lands between about $1.00 and $1.01.

So the answer from history is specific. Expect about $1.00 per quarter, or $4.00 a year, which would be an increase of about 10%, likely announced in September. Of course, the calendar is a pattern, not a promise. Microsoft has not scheduled or confirmed anything, and a board can always go off script.

Can the spending surge bend the pattern?

The reasonable concern is Microsoft's capital expenditures. The company allocated $115.9 billion to property and equipment in fiscal 2026 (the year ended June 30) -- an 80% jump from last year as it builds artificial intelligence (AI) data center capacity.

All that construction eats into the cash that would otherwise accumulate. Despite a 34% rise to $182.9 billion in operating cash flow, only about $67 billion in free cash flow remained after capital expenditures -- compared with about $72 billion a year earlier.

The earnings underlying the payment, however, are growing much faster than the payment itself. Fiscal 2026 revenue grew 18% to $331.8 billion, and Azure revenue crossed $100 billion for the year while rising 41%. The company's net income of $133.7 billion, meanwhile, came in 31% above the prior year.

A dividend that grows 10% a year while earnings grow at rates like those becomes safer each year, not riskier.

Now consider what the dividend actually costs. At $0.91 per quarter across about 7.4 billion shares, Microsoft pays out about $27 billion a year. That's about 15% of operating cash flow, and about 20% of the $17.95 per share the company earned in fiscal 2026. And a 10% increase adds something like $2.7 billion a year to the tab -- manageable but still meaningful.

Still, the AI spending surge is squeezing Microsoft's free cash flow, and even the squeezed figure still covers the dividend more than twice over.

The only unknown is the size

If anything bends this September's figure, I would expect it to bend toward the lower end of the band and not below it. With data center construction of that scale still underway, boards tend to protect flexibility. An increase near 9% or 10% preserves the streak and is easy to fund.

Could the board surprise with something larger? Yes, it has room. But nothing in its behavior for a decade suggests it wants to grab headlines with the dividend, and I don't expect it to start now.

The dividend yield will remain small either way. At about $505 per share, Microsoft yields about 0.7%, and an extra dime per quarter doesn't change that.

The increase, assuming one occurs, matters for what it signals, which is a payment that grows through every cycle -- AI construction included.

So what will this year's raise amount to? I expect a move to about $1.00 per quarter, announced in September, in the same band as the last six. For investors who own Microsoft, the most important thing to watch is free cash flow. The dividend is easily affordable today. Whether it remains so depends on a data center bill that's still rising.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Microsoft. The Motley Fool has a disclosure policy.

Meta Platforms (NASDAQ:META) stock peaked more than a year ago. The record close of $790.00 came on Aug. 12, 2025 (shares briefly traded as high as $796.25 three days later), and the stock, trading for about $579 as of this writing, sits about 27% below that mark.

In between came an expensive year: capital spending plans that kept climbing, a quarter of falling earnings, and a landmark legal settlement.

Here's my prediction anyway. Meta stock closes above that record before 2029.

That call doesn't require the stock's price-to-earnings ratio to rise. Getting back to the record by the end of 2028 requires about 36% appreciation, which works out to about 14% a year. And Meta's core business is already growing considerably faster than that.

Image source: Getty Images.

The ads business is still compounding fast

Whatever the stock has done, the advertising business is still compounding fast. Meta's second-quarter revenue rose 28% year over year to $60.8 billion, or 27% on a constant-currency basis. That's a step down from the first quarter's 33% growth, but volume and pricing are both still climbing. Ad impressions increased 14% year over year, while the average price per ad rose 12%. The company's apps now reach 3.60 billion daily active people, up 3%.

Why is the stock down, then? Because the bottom line hasn't kept up.

Costs and expenses in the second quarter jumped 55% year over year to $42.0 billion, dragging its operating margin down to 31% from 43% a year earlier. Diluted earnings per share fell 13% to $6.18.

To be fair, the quarter absorbed $2.4 billion of charges tied to legal proceedings and $1.18 billion of severance from a May headcount reduction -- about $3.6 billion of items that shouldn't repeat. The more durable weight is the build-out itself. Meta expects 2026 capital expenditures of $130 billion to $145 billion for artificial intelligence (AI) and its core business, a range whose floor it raised in July.

Then, last week, Meta agreed to pay up to $16.7 billion to settle claims from a coalition of state attorneys general that it misled the public about its apps' harms to teenagers, plus a separate $1 billion agreement with Texas. It has been a long time since this company gave the market an uncomplicated quarter.

The math, at today's multiple

Still, the prediction doesn't need an uncomplicated quarter. It needs math. Shares trade at a forward price-to-earnings ratio of about 17, based on expected 2027 earnings. For the stock to sit at $790.00 at that same valuation multiple, the earnings the market is pricing in would need to be about 36% higher than today's -- mid-teens annual earnings growth between now and the end of 2028. No multiple expansion required.

For a business growing revenue 28%, that could prove a modest ask.

One more charge comes first, though: Meta says it expects to book about $10 billion of legal expense in the third quarter to cover the settlement, a cost it hadn't built into its prior outlook. But charges like that end. The settlement converts an open-ended legal risk into a mostly known number -- about $12.7 billion of a roughly $18 billion package going to the states over 10 years, with the remaining $5.3 billion contingent on rival platforms accepting similar terms. And the severance reflects a company cutting headcount while revenue compounds -- the reported 75,472 still counts about 8,000 people cut in May, most of them gone by the end of this quarter.

The spending has to pay off

The honest risk to this forecast is the same thing that knocked the stock down in the first place. Capital spending of $130 billion to $145 billion this year becomes depreciation for years afterward, and depreciation lands directly on the earnings line.

If total expenses keep growing anywhere near 55% while revenue grows 28%, earnings won't compound in the mid-teens. They'll keep shrinking, and the math above falls apart. Second-quarter free cash flow of $784 million, down from $8.5 billion a year earlier, shows how much of the profit the build-out is consuming.

But CEO Mark Zuckerberg's claim that "AI is accelerating our core business today" is showing up in the numbers, at least on the revenue line. Ad prices rising 12% while impressions grow 14% is what an effective AI advertising system looks like. The spending has a return attached, and the question is timing.

Will Meta see its record again before 2029?

I believe it will. The required return is about 14% a year, the advertising business is compounding at twice that rate, and the stock's forward price-to-earnings ratio of about 17 is a modest price for this kind of growth.

If 2027 arrives with expenses still growing twice as fast as revenue, I'd rethink the call. Until then, I'd rather own the stock.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Meta Platforms. The Motley Fool has a disclosure policy.

Electricity demand is doing something it hasn't done in decades in the United States: growing fast. Vistra (NYSE:VST), one of the country's largest competitive power producers, told investors in its latest quarterly filing that data centers, the electrification of oil field operations, and electric vehicles are contributing to projected "fast-paced load growth" in the markets it serves.

You wouldn't know it from the stock. Shares have dropped about 37% from a 52-week high of $219.82, to about $139 as of this writing. And Vistra has company, as the whole independent power group has sold off this year. Nuclear operator Constellation Energy, for instance, is down about 32% from its own high.

With demand for Vistra's product climbing while its share price falls, is this a buying opportunity?

Image source: Getty Images.

A strong year, mostly locked in

Vistra's latest results, reported earlier this month, showed a business moving in the opposite direction from its share price. Adjusted earnings before interest, taxes, depreciation, and amortization (EBITDA) from ongoing operations rose about 31% year over year in the second quarter, to $1.77 billion from $1.35 billion a year earlier, helped by higher realized power and capacity prices and contributions from recently acquired plants.

Management also reaffirmed its 2026 adjusted EBITDA guidance of $6.8 billion to $7.6 billion. Even more, it said it expects to land at or above the midpoint of that range.

The cash generation behind those earnings is substantial. The company guides to adjusted free cash flow before growth investments of about $3.9 billion to $4.7 billion this year. Against a market capitalization of about $47 billion, the midpoint works out to a roughly 9% free-cash-flow yield.

And unusually for a business tied to commodity power prices, this year's results are largely spoken for. Management says about 100% of its expected 2026 generation volumes are hedged. Topping it all off, the company has been shrinking its share count aggressively, repurchasing about $6.5 billion of stock since late 2021 and reducing shares outstanding by about 30%.

Amazon and Meta signed on for 20 years

The development I find more important for the long run, though, is who is signing up to buy Vistra's power -- and for how long.

In September 2025, the company struck a 20-year power purchase agreement with Amazon Web Services, the cloud computing arm of Amazon (NASDAQ:AMZN), to supply 1,200 megawatts of carbon-free power from its Comanche Peak nuclear plant in Texas. Deliveries are expected to begin in late 2027.

In January, Vistra followed with 20-year agreements with Meta Platforms (NASDAQ:META) covering 2,609 megawatts of nuclear power and capacity from its Perry, Davis-Besse, and Beaver Valley plants, including new capacity from planned upgrades to all three. Deliveries under the Meta deals start late this year.

Notably, those Meta agreements aren't even in the company's 2027 outlook yet. Management points to an adjusted EBITDA "midpoint opportunity" of $7.4 billion to $7.8 billion for 2027 excluding them (and excluding a pending acquisition of gas plants). Vistra has also committed up to $1.0 billion to Helix, a new data center infrastructure venture where it will serve as the preferred power partner.

In short, nearly 4,000 megawatts of the company's nuclear output is now contracted to two of the world's largest technology companies for two decades each. That's revenue visibility competitive power producers rarely get.

Should you buy it?

Adjusted EBITDA is up 31%, guidance is intact, and decades-long contracts keep stacking up. Yet the stock trades at a forward price-to-earnings ratio of about 13. The drawdown looks less like a verdict on Vistra and more like the market cooling on the AI-power trade that got crowded in 2025.

Sure, there are risks. Vistra sells into competitive markets, so beyond its hedges and contracts, its results ride on power prices no one controls. A slowdown in data center construction could test the demand thesis. And second-quarter net income was just $305 million, weighed down by unrealized losses on hedging positions -- lumpy accounting that comes with this business model.

But at a forward price-to-earnings ratio of about 13, with this much of the future under contract, I think the stock is attractive. And I'd be a buyer at today's price. If power prices roll over or the data center deals stop coming, that would change my thinking. For now, I'd simply size the position with the volatility in mind.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Amazon, Constellation Energy, Meta Platforms, and Vistra. The Motley Fool has a disclosure policy.

Boeing (NYSE:BA) is finally building 737s at a pace it has not achieved in years. The program began transitioning to a production rate of 47 aircraft per month in the second quarter, according to the company's July earnings release, and Boeing initiated initial production on a new 737 line in July. For an aerospace giant that spent early 2024 limited to 38 per month by regulators, this production ramp is notable.

But now there's something that could get in the way. On Aug. 21, the two SPEEA units representing Boeing's approximately 17,000 engineers and technical workers rejected the company's contract offers and authorized a strike by overwhelming margins. The current contracts expire at midnight on Oct. 6.

Image source: Boeing.

A faster 737 line, at last

Boeing delivered 171 commercial aircraft in the second quarter -- a 14% increase from the 150 a year earlier. And revenue rose 8% year over year to $24.6 billion.

Free cash flow (non-GAAP), meanwhile, swung to a positive $631 million from an outflow of $200 million in the same quarter a year earlier.

Of course, Boeing still does not generate positive net income. Its non-GAAP (adjusted) core loss of $0.76 per share narrowed from a loss of $1.24 a year earlier.

The balance sheet also still holds $45.9 billion in consolidated debt -- more than double the $20 billion in cash and marketable securities on hand.

Further, Boeing's order backlog hit a record $715 billion in the quarter, including over 6,200 commercial aircraft. Demand, therefore, is not the constraint. Building and delivering fast enough is. The Federal Aviation Administration limited 737 production to 38 per month in January 2024 following the door plug accident on a nearly new MAX 9. It raised the limit to 42 last October, approved the move to 47 this past spring, and Boeing began the ramp-up to that pace in the second quarter.

So what would a strike actually halt?

SPEEA members do not assemble aircraft (Boeing's factory workforce belongs to a different union). The people who just voted are engineers and technical workers, and their vote was overwhelming.

The professional unit rejected the offer with around 64% voting against it, and the technical unit with around 72%. The two units authorized a strike with approximately 88% and 90% support, according to results published by SPEEA.

But their work underpins everything the production increase needs. Engineering supports production, deliveries, and the certification work Boeing expects to finish this year. On that last front, the FAA certified the smallest MAX variant, the 737-7, on Aug. 3 -- and Boeing says the larger 737-10 is next.

And Boeing is taking the risk seriously.

"We are now implementing our strike contingency plan and diverting the dollars we had wanted to invest in our SPEEA-represented team to prepare for a potential strike," said Ben Nimmergut, Boeing's vice president and functional chief engineer for production engineering, following the vote.

However, there is a new reason for optimism. Leeham News reported on Thursday that SPEEA and Boeing will meet on Monday to restart talks, after the union spent the week surveying its members on what a better offer needs. Still, not all signs point in that direction: Boeing has posted job openings for replacement engineers and technicians, according to the same outlet.

A deadline, not a strike

No strike can occur while the current contracts remain in effect, and they expire at midnight on Oct. 6. That leaves more than five weeks, and both sides say they want a deal.

But the recent precedent is uncomfortable. In 2024, more than 32,000 Boeing machinists went on strike in September after rejecting a tentative agreement, and the strike lasted more than seven weeks before a much richer contract ended it.

And SPEEA's rejection followed a similar path. The union's negotiating teams recommended the contracts, but their bargaining unit councils had already declined to endorse them -- and members then voted against the agreements by wide margins, citing deep distrust of Boeing's leadership.

Notably, an engineers' strike would not likely directly halt the assembly lines as the machinists' strike did. But I'd say investors shouldn't find much comfort in that. A walkout would stall the engineering support the production ramp-up depends on, and likely the 737-10 certification work still outstanding, exactly when Boeing is trying to prove it can sustain a pace of 47 per month.

As for the stock, it trades at about $210 as of this writing -- around 17% below its 52-week high of $254.35 -- and has trended lower over the two weeks surrounding the vote. Even after the drop, the shares trade at about 1.74 times sales.

At that valuation, the recovery arguably has to stay on schedule. And the next five weeks at the negotiating table will decide whether it does. Until Oct. 6, the delivery increase and contract talks are the same story.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Boeing. The Motley Fool has a disclosure policy.

Gap Inc. (NYSE:GAP) shares jumped about 13% Friday, the day after the apparel retailer reported fiscal second-quarter results and nudged its full-year profit outlook higher. The market liked the margins, the raised guidance, and news of a new leader for the company's biggest brand.

But the quarter was more lopsided than a pop like that suggests. Gap Inc. runs four brands (Old Navy, Gap, Banana Republic, and Athleta), and in the fiscal second quarter, exactly one of them was growing in any meaningful way. Total company net sales fell 2% year over year to $3.7 billion.

The brand that was growing happens to share its name with the stock. Here's a closer look at the quarter, brand by brand.

Image source: Getty Images.

Only one brand is really growing

The namesake Gap brand grew net sales 9% year over year to $844 million, with comparable sales up 10% -- what CEO Richard Dickson called "another quarter of double-digit comparable sales." That's the second quarter in a row of double-digit comparable growth, and it's momentum most mall retailers would love to have.

The rest of the portfolio went the other way. Old Navy's net sales fell 4% to $2.1 billion, with comparable sales also down 4% -- a reversal from growth a year ago, which management attributed partly to a weak women's seasonal assortment and slowing traffic.

Banana Republic inched up 1% to $478 million, with comparable sales up 3%. And Athleta, the activewear chain, saw net sales sink 12% to $264 million. Its comparable sales fell just as much, on top of a 9% decline a year earlier.

Old Navy is more than half the company

Why does one brand's stumble outweigh another's surge? Scale. Old Navy's $2.1 billion in quarterly net sales is about 57% of companywide net sales. The growing Gap brand, at $844 million, is well under half Old Navy's size. Growth of 9% at the smaller brand cannot offset a 4% decline at the bigger one. In dollars, Old Navy's slip erased roughly $85 million of quarterly sales while the Gap brand added about $70 million, which is how a company with a hot brand still shrank overall.

Management is acting on it. Gap Inc. named retail veteran Michael Francis as Old Navy's next president and CEO, succeeding Haio Barbeito.

And the company's updated outlook quietly acknowledges the problem. It now assumes Old Navy comparable sales of flat to down 1% for the year, cut from flat to up 1%, while its assumption for the Gap brand moved up to high-single-digit to low-double-digit comparable growth. The full-year plan got better, in other words, and the only brand assumption that moved up was the smaller one.

Check the adjusted numbers, not the reported ones

The reported results look spectacular. Gross margin came in at 52.8%, and earnings reached $1.38 per share.

But most of that is an accounting event, not retailing. The quarter included a $417 million net benefit to cost of goods sold from refunds of U.S. tariffs the company had previously paid, with the remaining refund cash expected in the third quarter.

Set the refund aside, and the underlying quarter was solid rather than stunning: an adjusted gross margin of 41.4%, up 20 basis points, and adjusted earnings of $0.52 per share. For the full year, the company raised its adjusted outlook to between $2.35 and $2.45 in earnings per share, up from a range of $2.30 to $2.40, on an adjusted operating margin of about 7.4% to 7.6%. Notably, the net sales outlook came down a touch at the top end (up 1% to 1.5%, versus up 1% to 2% before).

The case rests on one brand

With the stock near $24 as of this writing, the valuation works out to about 10 times the midpoint of this year's adjusted earnings-per-share outlook.

And the dividend yields about 3%. That's an inexpensive price for a company whose profit outlook just improved.

But the cheapness has a reason. The company's largest brand is shrinking, its fourth brand is shrinking faster, and the raised guidance leans on the one brand that's working (plus margin discipline) to cover for Old Navy until a new leader can get its sales growing again.

To management's credit, the Gap brand's turnaround has now run long enough to take seriously, and I think it has earned the benefit of the doubt. Just know what you'd be buying. Until Old Navy grows again, one brand is carrying the company.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.

Broadcom (NASDAQ:AVGO) CEO Hock Tan put a number on the company's future months ago, and he has not walked it back. On the chip giant's June 3 earnings call, Tan said the company expects about $56 billion of artificial intelligence (AI) semiconductor revenue this fiscal year, up approximately 180% from fiscal 2025.

And then he went further.

"We reiterate our AI semiconductor revenue guidance to be in excess of $100 billion" for fiscal 2027, he told analysts.

The business behind the forecast delivered on the same call. AI semiconductor revenue reached $10.8 billion in the fiscal second quarter of 2026 (the period ended May 3), a 143% jump from a year earlier, with guidance calling for $16.0 billion in the third quarter.

The growth stock has traveled in the opposite direction. It trades near $370 as of this writing. The 52-week high is $495, which puts today's price about 25% below the peak.

So heading into Broadcom's fiscal third-quarter report on Wednesday, Sept. 2, the big question isn't the size of the forecast. It's what the market is discounting by paying far less for the same $100 billion promise.

Image source: Getty Images.

The forecast is built on contracts

Tan's fiscal 2027 target is not a hope. On the June call, he walked through the commitments underneath it: a long-term agreement with Google parent Alphabet (NASDAQ:GOOG)(NASDAQ:GOOGL) covering multiple generations of its TPU chips, an arrangement giving Anthropic access to another 5 gigawatts of TPU-based compute beginning in 2027, and a contractual commitment to deploy 1.3 gigawatts for OpenAI next year. A partnership with Meta Platforms adds 3 gigawatts of custom chips through the end of 2028. And purchase orders totaling $6 billion had arrived from two additional customers as of the call, too.

Customers, if anything, were booking further out than they used to.

"During the quarter, bookings for AI semiconductors were over $30 billion against the $10.8 billion we shipped," Tan said.

Asked later why the backlog had grown so fast, he added, "Our visibility runs all the way to 2028 right now."

The price cut, measured

Because Tan's forecast hasn't moved, the drawdown has landed squarely on the price of the company's future profits. At its 52-week high, Broadcom traded at about 25 times the earnings analysts expect for fiscal 2027, on an adjusted basis -- the year the $100 billion forecast covers. Today it trades at about 19 times those same expected earnings.

That leaves the same forecast selling for about 25% less than it commanded at the peak. So what, specifically, is the market discounting?

Timing, concentration, or the number itself?

The candidates come down to three.

Timing is the most concrete. Tan said Broadcom plans to ship about 10 gigawatts of AI compute in fiscal 2027, weighted toward the back half of the year. A back-loaded ramp means the revenue that justifies today's price shows up late, and any slip pushes it into fiscal 2028. It also means a quarter or two of results could look ordinary while the forecast stays intact.

Concentration is the familiar one, and it has fresh evidence. On Aug. 19, Broadcom shares fell about 5% after Marvell Technology disclosed an expanded custom-chip agreement with Google, whose TPU chips Broadcom has long designed. Six core customers carry the AI number, so a shift at even one matters.

Still, nothing says Google is leaving. Broadcom announced its own long-term agreement in April covering multiple generations of TPUs, and these customers sign multiyear contracts, not one-off orders.

And the forecast itself is the hardest one to doubt. After all, doubting it means doubting signed agreements Tan has described in detail, plus bookings running at nearly three times shipments.

To me, timing is the likeliest answer, with the Google question the newest one to watch. A back-half-loaded ramp facing a market that wants proof now is enough, on its own, to explain a drawdown like this -- no broken thesis required.

That is why the Sept. 2 report matters more than a typical quarter. The results will show how the $16 billion AI quarter came in, and the fiscal fourth-quarter guidance will show the state of the $56 billion full-year number, the base the fiscal 2027 ramp builds on. Those two numbers are the first hard checkpoints between June's promises and next year's $100 billion.

Ultimately, I view the stock as a hold here. The forecast, I think, is credible. And the drawdown has made it much cheaper to own. But a back-loaded ramp can test investors' patience for quarters at a time, and the valuation -- 19 times earnings that still have to be delivered -- is a discount only if the delivery happens. If the Sept. 2 report holds both numbers and the stock stays near today's level, I'd get more interested.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet, Broadcom, Marvell Technology, and Meta Platforms. The Motley Fool has a disclosure policy.

Amazon (NASDAQ:AMZN) is running the largest capital-spending program in its history. The company expects about $220 billion in capital expenditures this year, an estimate CEO Andy Jassy raised from $200 billion in July.

And the spending runs well past this year. On Aug. 26, Amazon Web Services (AWS) and Nvidia announced plans to put 2 million more Nvidia graphics processing units (GPUs) into AWS's infrastructure in 2027 and 2028, adding to plans to put more than 1 million GPUs in place starting in 2026, announced earlier this year.

So what has spending on this scale historically meant for the stock? Amazon has been here before, and the record is specific. In the past 15 years, the stock's two worst years were also years its bottom line went negative in the middle of a heavy investment stretch, and both were followed by enormous rebounds.

But the record holds exactly two instances. And the spending, on its own, was never what did the damage.

Image source: Getty Images.

The two bad years

In 2014, Amazon's capital expenditures reached $4.9 billion, up 42% year over year and about five times what the company spent in 2010. Sales still grew 20% to $89 billion. But operating income shrank to $178 million, and the company posted a net loss of $241 million. The stock fell 22% that year.

Then came 2015. Operating income rebounded more than tenfold to $2.2 billion, the company swung back to a profit, and the stock rose 118% -- its best year of the past 15.

The 2022 episode was bigger in every direction. Capital expenditures hit a then-record $58.3 billion, and even with revenue up 9% year over year, Amazon reported a $2.7 billion annual net loss. Operating income halved to $12.2 billion that year, and a $12.7 billion pre-tax valuation loss on the company's investment in Rivian Automotive dragged the bottom line into the red. The stock lost about half its value.

A year later, in 2023, net income came in at $30.4 billion, and the shares rebounded 81%.

Spending alone was never the signal

Amazon's other heavy spending years (2021, 2024, and 2025) saw capital expenditures of $55.4 billion, $77.7 billion, and $128.3 billion. The stock's returns in those years: up 2%, up 44%, and up 5%. Uninspiring in two cases, but nothing like 2014 or 2022.

Notably, even a loss year wasn't automatically fatal. In 2012, Amazon reported a small net loss of $39 million while investing heavily, and the stock rose 45% anyway.

What set 2014 and 2022 apart is that the income statement stopped keeping up. Operating profit nearly disappeared in 2014 as the spending rose. In 2022, operating income halved while the Rivian write-down pushed the bottom line negative. When investors could still see earnings growing through a build-out, they kept paying for the build-out.

Which setup is 2026?

On the cash-flow statement, today looks like the bad years. Amazon's trailing-12-month purchases of property and equipment, net of proceeds, have reached $169 billion -- up $66.1 billion from a year earlier, an increase the company attributes primarily to artificial intelligence (AI).

Free cash flow has flipped negative: an outflow of $7.6 billion over the trailing 12 months, against an inflow of $18.2 billion the year before. That capital spending now runs at about 22% of trailing revenue, arguably a heavier weight than the company carried through 2014 or 2022.

On the income statement, however, today looks nothing like them. Operating income rose 43% year over year to $27.5 billion in the second quarter of 2026. AWS revenue grew 37% year over year last quarter, its fastest pace since 2021, after accelerating through the first half of the year. The profit erosion that marked both bad years is, so far, absent. Of course, that could change -- depreciation from the build-out may weigh on margins in the quarters ahead.

So, does the market pay for a build-out while it's happening, or only after it stops? Amazon's history answers both ways. It has paid right through the biggest spending years, whenever profits kept growing underneath them. It punished the two years profits vanished, then handed the stock two of its best years once they returned. So far, the market is paying right through this one: shares trade near $266 as of this writing, up about 15% in 2026.

In short, the number to watch from here isn't the size of the capital budget. It's whether operating income keeps climbing while the budget runs. I'd start worrying if that growth stalls. But two instances of history say the spending alone isn't a reason to sell, and I think they have it right.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Amazon and Nvidia. The Motley Fool has a disclosure policy.

Peter Thiel's hedge fund disappeared from the stock market for six months. Thiel Macro, the firm that manages the billionaire's money, reported no U.S. long stock holdings at all for two consecutive quarters (the periods ended December 2025 and March 2026). Then, earlier this month, it filed a portfolio of eight names worth $418.7 million as of June 30.

One caveat belongs up front. A 13F filing covers only a manager's long positions in certain U.S.-listed securities. Short bets, futures, currencies, private stakes, and cash are all invisible to it. Those two empty quarters, then, don't mean Thiel's fund held nothing. They mean it held nothing the form counts.

Still, an empty stretch says something. And so does what the fund bought on the way back in.

Because the new portfolio has a theme. About 72% of it sits in companies that generate electricity, deliver it, or supply the fuel behind it.

Image source: Getty Images.

Six months of nothing

In the third quarter of 2025, the fund sold out of an Nvidia (NASDAQ:NVDA) position it had valued at $85 million three months earlier -- a sale that got attention at the time, landing amid a loud debate about an artificial intelligence (AI) bubble. That left its reported holdings at just $74 million across three stocks. The next two filings showed nothing at all.

Notably, the fund has gone years without filing at all, including a long stretch from late 2020 into early 2025. This was different. It filed the form both quarters and reported nothing on it.

Whatever the fund was doing during those six months, it wasn't holding U.S. stocks the form counts. The second-quarter filing is the first evidence of where Thiel wanted to be next.

Almost three-quarters of it is energy

The largest position is the one exception to the theme. It's a stake in e-commerce giant Amazon (NASDAQ:AMZN) worth about $118 million, or 28% of the portfolio -- the fund's only technology holding, and a name it has owned before.

Everything else is tied to energy or power, in one form or another. Vista Energy (NYSE:VIST), an oil and gas producer developing Argentina's giant Vaca Muerta shale field, is the second-largest position at about $76 million, or 18% of the portfolio. And merchant power producer Vistra (NYSE:VST), another returning name that the fund owned briefly in 2025, comes in at about $59 million, or 14%.

Then come four regulated utilities (steady, slow-growing dividend stocks), sized almost identically. American Electric Power (NASDAQ:AEP) is about $42 million, while DTE Energy (NYSE:DTE), FirstEnergy (NYSE:FE), and CMS Energy (NYSE:CMS) sit at about $40 million each (roughly 10% of the portfolio apiece). A small $3.7 million stake in X-Energy (NASDAQ:XE), a nuclear reactor developer that completed its initial public offering (IPO) in April, rounds out the eight.

Add it up, and the energy and power names come to about $301 million of the $419 million total -- about 72%.

Power, not chips

Consider the sequence. The last big move this fund disclosed before its empty stretch was selling Nvidia, whose graphics processing units (GPUs) sit at the center of the AI trade. The first move it disclosed on the way back was buying the electricity complex that AI's data centers depend on. Taken together, the portfolio amounts to a view that the bottleneck in AI may no longer be the chips -- it's the power to run them.

And the way the bet is spread says something, too. This isn't a concentrated swing at one hot generator. It's the whole supply chain: the fuel, the plants, and (the part I find most interesting) the wires. Owning four regulated utilities, nearly equal-weighted, is arguably a bet that growing electricity demand can lift the entire grid, including its most boring corners.

Of course, investors should be careful about how much to take from any of this. A 13F is a snapshot that is already about 45 days old by the time it becomes public, and it says nothing about positions on the other side. After all, $419 million is likely a modest slice of Thiel's total wealth, and he could be positioned differently by now.

Still, the shape of the portfolio is hard to miss. A technology billionaire's fund came back from six months of nothing, and seven of its eight buys were energy. Whether or not anyone should copy the positions, the view behind them is clear: the next phase of the AI build-out may belong to the companies that supply the electricity.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Amazon, Nvidia, and Vistra. The Motley Fool has a disclosure policy.

On Sept. 1, Apple (NASDAQ:AAPL) gets its first new CEO in 15 years. John Ternus, the company's head of hardware engineering and a 25-year Apple veteran, takes over from Tim Cook, who becomes executive chairman of the board.

Apple announced the plan in April, and the board approved it unanimously. It is a planned handoff to someone from inside the company, just as large tech firms typically do.

The stock, meanwhile, enters the change near record territory. Apple's market cap approaches $4.6 trillion as of this writing, over 30% above where it was a year ago, and shares sit about 9% off the all-time high they set this summer.

So what does a moment like this typically do to a stock? Since 2011, four planned CEO handoffs with internal promotions at U.S. megacap tech companies have a completed first year to judge. A fifth is too recent to qualify: Oracle promoted two in-house executives to co-CEO last September, so its first year is not complete. The four on record have almost nothing in common.

Image source: Apple.

Four handoffs, four different years

The most famous is Apple's own. Steve Jobs stepped down as CEO on Aug. 24, 2011, handing the post to Tim Cook, the company's head of operations. Over the following 12 months, Apple stock rose about 76%.

Microsoft named Satya Nadella, an in-house veteran, CEO on Feb. 4, 2014. The stock rose about 15% over the following 12 months (a good year, though hardly a preview of the cloud-driven streak that followed).

Alphabet promoted Sundar Pichai, who already ran Google, to CEO of the parent company on Dec. 3, 2019. Twelve months later, the stock had risen about 41%, even with the 2020 pandemic crash falling in the middle of that window.

And then there is the outlier case. Amazon founder Jeff Bezos handed the CEO post to veteran cloud chief Andy Jassy on July 5, 2021. Over the following 12 months, Amazon shares fell about 38%.

Average those four first years and you get a gain of about 24%. But the average hides the point. Such scattered results -- a deep loss, a modest year, and two large gains -- suggest that the handoffs themselves did not drive them.

The handoff was never the variable

Look closely at the four cases, and what really decided each first year was the starting point the new CEO inherited, not the person.

Cook took over a stock trading near 15 times earnings just as the iPhone was entering its most pronounced growth years. Nadella inherited a similar price, about 14 times earnings, with much of Microsoft's transformation to the cloud still ahead. And Pichai took Alphabet at about 26 times earnings with digital advertising still compounding.

Jassy, by contrast, took Amazon just days from what was then its all-time high, at a price near 70 times earnings, right as pandemic-era e-commerce growth was stalling. The stock's terrible first year under Jassy seems to have had little to do with him. It was the price and the cycle, unwinding at the same time.

In each case, the market spent the first year repricing the business the new CEO received. None of the four stocks seems to have moved much because of the succession itself -- they were planned transitions to insiders the market already knew.

Ternus inherits a good hand at a high price

Apple's starting point today sits somewhere between the comfortable ones and Amazon's. At about 36 times earnings, the stock is more than twice as expensive as the one Cook inherited. And the market value rise of more than 30% over the past 12 months already covers both the succession announcement and a streak of good results. Easy gains, put another way, may already be behind the stock.

However, 36 times earnings is also nowhere near Amazon's multiple of about 70 times earnings during its handoff. And Apple enters the change with momentum instead of a stall, with a new generation of iPhone likely this fall and a services business that continues to compound.

After a decade covering large tech stocks, I can't think of a planned megacap succession the market actually feared, and the four first years above say it was right not to. History says the handoff itself will probably be a non-event. It also says that the first-year return will be decided by what Ternus inherited -- the iPhone cycle, artificial intelligence (AI) features, and a demanding price.

For investors who already own the stock, the transition is not a reason to sell. I'd keep holding Apple through it. I just wouldn't expect Cook's version of the first year at today's valuation.

Daniel Sparks and his clients have positions in Apple. The Motley Fool has positions in and recommends Alphabet, Amazon, Apple, Microsoft, and Oracle. The Motley Fool has a disclosure policy.

Flash memory specialist Sandisk (NASDAQ:SNDK) and its long-term manufacturing partner Kioxia said Thursday that they plan to invest more than $31 billion in Japan through 2032. The money is earmarked for infrastructure at the Yokkaichi and Kitakami plants (the factories where the two companies produce their NAND flash memory), along with related technology development.

The plan is contingent on Japanese government support.

Over the alliance's more than 25 years, the two companies have invested more than $50 billion in Japan, according to the announcement. The new plan would spend about 60% of that sum again in about six years.

Both figures are floors ("more than"), so the proportion is approximate. The plan's scale is not. And the announcement looks odd next to what Sandisk management itself told investors three weeks earlier: that the company is increasing supply through technology improvements rather than large capacity expansions, with capital expenditures falling as a percentage of revenue.

So which one is it?

Image source: Getty Images.

Who pays what

The plan is joint, not a $31 billion check from Sandisk alone. The two companies manufacture through a joint venture structure called Flash Ventures, which operates at eight facilities in Japan (six in Yokkaichi and two in Kitakami). In January, they extended that framework through December 2034.

Sandisk holds a 49.9% stake in the Flash Ventures entities, and Kioxia owns the facilities themselves. Each side gets roughly half of the production. And Sandisk's annual report says the company is obligated to finance between 49.9% and 50% of the capital expenditures that the joint ventures decide to make, to the extent that the joint ventures' own cash flow cannot cover them.

Neither company has detailed its share, and Sandisk's obligation covers only the joint ventures' own investments. But if about half of the plan flows through Flash Ventures, something close to $1.3 billion a year falls on Sandisk, before what the Japanese government contributes.

Doesn't that break the capital-light story?

"We grow supply primarily through nodal transitions rather than wafer additions, delivering mid- to high teens bit growth," CEO David Goeckeler said on the company's earnings call on Aug. 5. And chief financial officer Luis Visoso supplied the figure, guiding capital expenditures to about 6% of revenue for fiscal 2027 even as the company accelerates its newest manufacturing technologies.

At first glance, a $31 billion build program appears to contradict all that. But if you follow how the money flows, I would say the capital-light story holds up for the most part.

For one thing, Sandisk's funding obligation is a backstop, not a blank check. The company covers its share of the joint ventures' investments only when Flash Ventures' own operating cash flow cannot.

That said, the 6% guidance and the $31 billion plan are the same money. What Visoso guided is gross capital expenditures, which already includes Sandisk's share of what Flash Ventures builds. The company's own property purchases totaled just $177 million in fiscal 2026, far short of 6% of revenue, and it also put a net $275 million into the joint ventures. So the plan's bill has to fit within that guidance, not sit beside it.

And then there is Sandisk's explosive revenue base. The company's revenue in fiscal 2026 rose 175% year over year to $20.25 billion, and guidance for the fiscal first quarter of 2027 alone projects revenue of $10.3 billion to $10.8 billion. Against a business of that size, that bill fits within Sandisk's 6% guidance.

Demand still has to last

Of course, the hardest issue for shareholders is durability. The plan runs through 2032, and memory has long been a wildly cyclical business.

However, Sandisk has more visibility on that than in past cycles. Long-term agreements with eight customers already cover about half of the company's expected bit shipments for fiscal 2027, and Sandisk values those agreements at $93.9 billion over their lives, based on the minimum prices they guarantee. The demand secured in writing may be what makes a six-year build plan defensible.

Sure, the uncontracted half of the business still floats on market prices, and no contract protects years beyond its term. But this announcement amounts to two of the industry's biggest players betting that the storage boom for artificial intelligence (AI) will last longer than this quarter's debates about it.

Still, the growth stock closed Thursday near $1,485, 37% below its June peak.

At that price, the shares cost about 7 times forward earnings for the next fiscal year. In other words, the market still doubts how long the boom's earnings can last. The $31 billion headline sounds like a strategy shift. The structure beneath it -- jointly funded, contingent on government support, and sized to a revenue base that nearly tripled last year -- looks more like the plan management described, operating at the scale the boom now demands.

Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has no position in any of the stocks mentioned. The Motley Fool has a disclosure policy.