The AI Spending Wave Is Splitting in Two

The AI Spending Wave Is Splitting in Two

Markets have a habit of treating a category as a single trade. Every stock in the sector gets the same label, the same multiple compression, the same selloff when sentiment turns. That is usually where the work starts for investors willing to look one level deeper.

This week’s hyperscaler earnings gave us one of the clearest examples of that dynamic in years. The four largest AI spenders, Alphabet, Amazon, Meta, and Microsoft, are collectively on pace to spend roughly $725 billion on capital expenditures in 2026, up approximately 77% from last year’s already record $410 billion. The market has spent months treating that aggregate number as the story. It is not the story. The split inside it is.

Some of this capital is already converting into margin-expanding, backlog-building, cash-generative cloud revenue. Some of it is consuming operating cash flow with no near-term feedback loop for when the math turns. Same sector. Same headline spending cycle. Completely different return profiles. That gap is where the discipline of value investing does its most useful work.

Why Did the Market Discount These Companies?

Alphabet reported Q2 results on July 22. Google Cloud revenue surged 82% year over year to $24.8 billion, accelerating from 63% growth the prior quarter. Operating income at the Cloud division hit $8.8 billion on a 35.6% operating margin. The backlog jumped $50 billion in a single quarter to $514 billion. Search revenue rose 17%. By almost any operating measure, Alphabet delivered one of the strongest quarters in its history.

The stock fell more than 7%.

The reason sits in one line of the cash flow statement. Capital expenditures reached $44.9 billion in the quarter. Operating cash flow was $39.1 billion. The arithmetic produced free cash flow of negative $5.85 billion, the first negative free cash flow quarter in Alphabet’s 22 years as a public company. Management raised full-year 2026 capex guidance to $195 billion to $205 billion, its third upward revision this year. The CFO said 2027 capex will increase again and will continue to put pressure on profits. Alphabet bought back zero shares for the second straight quarter and has a $40 billion share offering pending.

Investors who had held Alphabet as a cash-generating machine for a decade watched the company transform into a capital-intensive infrastructure builder almost overnight. The reaction was emotional. That does not make it wrong. But it does make it worth examining carefully.

Separating the Temporary from the Structural

The central question for any investor looking at Alphabet today is not whether the free cash flow turned negative. It clearly did. The question is why, and whether the reason is temporary or structural.

Consider the supply context. Alphabet is renting Nvidia chips from SpaceX for roughly $920 million a month and leaning on third-party cloud providers including CoreWeave and Nebius as a bridge while it builds its own data centers, because demand is outrunning its existing capacity. Operating cash flow is actually growing, up 41% year over year to $39.1 billion. The core business is generating more cash than ever. The negative free cash flow is the product of $44.9 billion in capital expenditures landing in a single quarter, not a deterioration in the underlying business.

That distinction matters. A business whose operations are generating less cash is a different problem than a business that is temporarily spending more cash than its operations produce. The first is a quality concern. The second is a capital allocation question. Those require entirely different frameworks.

The evidence that the spending is producing real returns is concrete. Google Cloud’s operating margin expanded from 20.7% in Q2 2025 to 35.6% in Q2 2026. A backlog that stood at $106 billion a year ago now sits at $514 billion. That is not the signature of a business burning capital recklessly. It is the signature of a business investing in infrastructure to support demand that already exists on paper.

The Companies That Have Already Answered the Question

Amazon and Microsoft reported after Alphabet, and both gave investors something Alphabet had not yet shown: confirmation that the infrastructure spending cycle produces cloud economics on the other side.

AWS grew 36.7% in Q2 2026 to $42.23 billion in revenue, its fastest growth rate in 18 quarters, accelerating sharply from 28% in Q1. AWS operating income reached $16.62 billion against a Street consensus of $13.62 billion. The operating margin was 36.8%. CEO Andy Jassy said AWS’s AI business and its chips each crossed an annual revenue run rate above $25 billion, both more than doubling from a year earlier. Jassy described the potential for AWS to become a $1 trillion annual revenue business, from a current $169 billion annualized run rate. Amazon raised full-year 2026 capex guidance to $220 billion.

Microsoft’s Azure and other cloud services grew 43% in its most recent quarter. The Intelligent Cloud segment generated revenue of $33.7 billion. Commercial remaining performance obligations now stand at $627 billion. Microsoft spent approximately $97.2 billion on capital expenditures over the last twelve months, and analysts expect that to reach $118.9 billion over the next twelve months, a 22% increase, before accelerating to $178.9 billion the following year.

Both companies are past the most important question in this cycle. Revenue growth is outpacing incremental capex growth at the margin. Backlogs are expanding. Margins are holding or improving. These are not companies asking investors to wait. They are companies showing the math in real time.

The Company Still Writing the Question

Meta told a harder story on the same evening as Microsoft. Revenue of $60.8 billion came in 28% above the prior year, the fastest growth since late 2021, beating consensus by roughly $600 million. Then the cash flow statement arrived. Free cash flow collapsed 91% to $784 million, from $8.55 billion in Q2 2025. Capital expenditures of $31.1 billion consumed nearly all of the $31.9 billion in operating cash flow the business generated. Shares fell roughly 10% in after-hours trading.

The mechanics are similar to Alphabet. But the investment case is structurally different, and that is the distinction patient investors need to make.

Alphabet’s negative free cash flow is the output of a cloud business growing at 82% with a $514 billion backlog and expanding margins. The spending has a direct, measurable, accelerating revenue tie-back. Meta’s negative free cash flow is the output of a company building infrastructure it does not yet have a commercial architecture to monetize at scale. Zuckerberg offered a sketch of a potential enterprise cloud business on the earnings call. There was no product description, no timeline, and no named competitive strategy against AWS, Google Cloud, and Azure, each of which holds years of enterprise contracts and data center relationships that Meta does not.

Meta is also exploring a potential $10 billion compute rental arrangement with Anthropic as a way to alleviate cash flow pressure. That may be rational capital allocation. It also signals the company is aware that its infrastructure build is running ahead of its monetization plan. Meta has guided full-year 2026 capex to $130 billion to $145 billion, narrowing the floor by raising it from $125 billion, at a midpoint that leaves minimal room for meaningful free cash flow over the full year given the operating cash run rate in the first half. The company has addressed this by issuing new long-term debt.

The Cheap Investor Scorecard

Cheap, Fairly Valued, or a Value Trap?

The market is currently pricing Alphabet as if the negative free cash flow quarter is a sign of deterioration. The evidence suggests something else. Operating income grew 30% to $40.8 billion. The operating margin expanded to 34%. The cloud business more than tripled its margin in a single year. A backlog of $514 billion does not accumulate by accident; it represents signed customer commitments. At approximately 22 times 2027 earnings, Alphabet is not priced at a reckless premium for the quality and growth rate on display.

The risk is real. Management has committed to rising capex in 2027, and the CFO has said it will continue to pressure profits. Alphabet has taken on roughly $100 billion in debt this year and completed its first share sale in more than two decades, a significant reversal for a company that spent years buying back its own stock. The question investors must answer honestly: is the company’s competitive position in AI strong enough to justify this transformation in capital structure? The 82% cloud growth rate and the $514 billion backlog suggest the demand is there. Whether the infrastructure can be built fast enough, and profitably enough, is a different question.

Amazon is the clearest case in the group. AWS is generating $16.62 billion in quarterly operating income at a 36.8% margin, growing at its fastest rate since 2021, with Jassy articulating a credible long-term case for a $1 trillion revenue business from a $169 billion annualized run rate. The capex is heavy. The backlog justifies it.

Meta is the most honest test of investor discipline. The ad business is excellent. Revenue of $60.8 billion growing 28% is not a broken company. But the AI infrastructure spending, $130 billion to $145 billion for the full year, is funding a strategic ambition rather than a proven commercial platform. The company has no cloud business to validate the spending with backlog data. It has an open-source model strategy through Llama that is widely distributed but not directly monetized in the conventional sense. It may be building something transformative. It may also be building a very expensive lesson in competitive moats. At this point the evidence does not clearly support either conclusion, and that ambiguity deserves a commensurate discount.

The Investment Case and the Risks

For Alphabet: the market believes the negative free cash flow signals structural deterioration. The evidence suggests it signals a deliberate, temporary bet on infrastructure capacity to support demand that already exists in its backlog. The disconnect is a matter of time horizon. Investors with a two-to-three-year view, willing to hold through the capex intensity of 2026 and 2027, are being offered a cloud business growing at 82% with expanding margins at a multiple that does not reflect that growth rate. The risk is that capex continues to rise faster than revenue, that Gemini falls behind in the model race, or that enterprise customers consolidate workloads on fewer platforms. All three are plausible. None has shown up in the Q2 numbers yet.

For Amazon: the investment case is the most self-evidently supported by the data. AWS’s re-acceleration to 37% growth, combined with a 36.8% operating margin, makes the capex discussion relatively straightforward. The risk is execution at scale, and whether power and supply chain constraints limit the pace of capacity expansion. Jassy was specific on the Q2 call: 2028 capacity is already being contracted. That is either confidence or overcorrection, and it is impossible to know which from the outside.

For Meta: the downside if the thesis is wrong is not trivial. Between 2021 and 2022, Meta shares fell 77% as investors soured on metaverse spending that lacked a commercial architecture. The AI spending cycle is different in important ways. AI already has proven applications. Advertising revenue benefiting from AI recommendations is measurable. But the enterprise cloud ambition Zuckerberg sketched on the earnings call faces competitors that have spent a decade building the sales relationships and compliance certifications that enterprise buyers require. That is not a gap that capital alone closes quickly.

What to Watch

Three things will determine whether the next phase of this cycle rewards patience or punishes it.

• Whether Google Cloud’s margin expansion continues at pace. A move from 20.7% to 35.6% in a single year is exceptional. If margins plateau or compress as capacity costs rise, the investment case changes materially. The Q3 report will be the first real test of whether Q2’s margin was structural or a one-quarter artifact.
• Whether Meta surfaces a direct monetization pathway for its infrastructure. Revenue from an enterprise cloud offering, or a credible commercial Llama licensing model, would change the risk profile of the spending significantly. Until then, investors are being asked to fund a buildout without a business plan attached to it.
• Whether the power grid keeps pace with the commitments already on paper. AI-optimized server racks now require 50 to 100 kilowatts of power compared to 5 to 10 kilowatts for traditional racks. Morgan Stanley has forecast a shortfall of roughly 49 gigawatts in available U.S. power access by 2028, against projected data center demand of 74 gigawatts. The capital is committed. The electrons are not guaranteed to arrive on schedule. Hyperscalers that solve this constraint first will have a durable infrastructure advantage over those still waiting on grid interconnection approvals that can run for years.

The Verdict

The aggregate $725 billion spending figure is the wrong frame. The right frame is the quality of what each dollar is building and the clarity of the pathway back to investors.

Amazon has answered the most important question in the cycle: the infrastructure spending is producing returns, and those returns are accelerating. Alphabet is answering it in real time, with margin expansion at Google Cloud that was not visible a year ago and a backlog that grew by $50 billion in a single quarter. The negative free cash flow quarter is real, and the capex intensity will persist through at least 2027. But the operating business is not deteriorating. It is investing, and the evidence of that investment producing returns is mounting.

Meta is a different conversation. The ad business is excellent. The AI spending may prove rational over a long enough horizon. But at this point, the company is asking investors to fund a strategic ambition without the commercial architecture that would justify the capital intensity in the near term. That is not necessarily a value trap. It is an unresolved question. And unresolved questions, in a market that has already bid up the AI trade substantially, deserve a margin of safety that is not currently visible in Meta’s numbers.

The highest-conviction ideas in this cycle are not the ones with the most exciting spending announcements. They are the ones where the evidence of return on capital is compounding faster than the market’s skepticism.

The Cheap Investor