Alphabet Gave the Perfect AI Quarter and the Stock Fell. Free Cash Flow Went Negative for the First Time.

On the evening of July 22, Alphabet reported the single best cloud quarter any company has ever printed. Google Cloud revenue grew 82 percent year over year to $24.8 billion. Cloud operating income more than tripled, from $2.83 billion to $8.81 billion, lifting segment margin from 20.7 percent to 35.6 percent. Total revenue rose 24 percent to $119.8 billion. The contracted backlog, the money customers have already signed for but Alphabet has not yet recognized, hit $514 billion, up more than $50 billion in a single quarter.

The stock fell about 5 percent.

Every piece of coverage I read blamed the capex guide, and the capex guide is real: full-year 2026 moved to $195 billion to $205 billion, up from $180 billion to $190 billion a quarter earlier, against a street modeling roughly $188 billion. But raising capex by fifteen billion dollars is not what broke the print. The number that broke it was three lines further down the cash flow statement.

The Line That Actually Moved the Stock

Alphabet spent $44.9 billion on capital expenditure in the quarter, roughly double the year-ago figure. The business generated $39.1 billion of operating cash flow. Subtract one from the other and free cash flow came in at negative $5.9 billion, the first negative free cash flow quarter in Alphabet's history as a public company.

That last row is the one nobody is quoting. Alphabet, a company that has spent two decades as the most cash-generative business in technology, raised almost $70 billion in the quarter across stock, mandatory convertible preferred, and senior notes. Management told analysts free cash flow will remain under pressure because of investments in technical infrastructure. Roughly 60 percent of the spend is servers and 40 percent is data centers and networking, so this is not a one-quarter timing artifact. It is the shape of the next two years.

For a decade the market treated Alphabet as a capital-light advertising monopoly with a cloud option attached. What reported on July 22 was a capital-intensive infrastructure company with an advertising monopoly attached. Those get different multiples. The 5 percent move was not a verdict on the quarter. It was a partial reclassification of the business.

Three More of These Land This Week

Microsoft and Meta report Wednesday, July 29. Amazon reports Thursday, July 30. They walk into a market that just rewrote the scoring rule in public. In 2025 the question was whether the AI demand was real. That question is settled: 82 percent cloud growth and a half-trillion-dollar backlog is not a demand problem. The question for this week is whether the demand converts to cash fast enough to cover what the buildout costs while it is being built.

Combined 2026 AI capital spending across the four is running near $725 billion, up roughly 77 percent from about $410 billion in 2025. AI capex has gone from about 33 percent of the group's operating cash flow in 2023 to something near 93 percent this year. Alphabet is the first to cross the line where the business no longer funds the buildout by itself. It will not be the last.

Look at the last column, because it is the defense every one of these companies will run on the call. Backlog, reported as remaining performance obligations, is the answer to "how do you justify the spend." You justify it by showing revenue that is already signed. Three of the four have that answer ready.

The Hole in the Backlog Defense

Here is what I would ask on Wednesday if I had a line on the call. Microsoft's commercial backlog is around $625 billion, up about 110 percent year over year, which is an extraordinary number until you learn that roughly 45 percent of it sits with a single customer, OpenAI, and that excluding OpenAI the same figure grows about 26 percent. Both numbers are true. They describe very different companies.

Amazon has the same structure in a milder form. AWS remaining performance obligations reached $364 billion, up 93 percent, and that figure excludes a new $100 billion OpenAI commitment layered on top of an existing $38 billion contract. Alphabet's own backlog leans meaningfully on Anthropic, a company Google has invested in directly.

So the defense reads: our spending is justified by contracted demand. And a large slice of that contracted demand comes from AI labs that these same hyperscalers helped capitalize, whose ability to pay depends on funding rounds the hyperscalers participate in. That is not fraud and it is not even unusual in infrastructure buildouts. But "contracted revenue from an independent customer" and "a commitment from a company we funded" are different assets, and the RPO line does not distinguish between them.

I have written before about circular structures in the chip deals. This is the same pattern one layer up, and it is larger. When a supplier finances the customer, the backlog is partly a mirror.

Meta Has the Hardest Job

Meta guided 2026 capex to $125 billion to $145 billion. It has no cloud business, no external customers for its compute, and therefore no backlog line to put on a slide. Every dollar Meta spends has to be justified by internal return: better ranking, better ads, better recommendations, eventually better products. Those returns are real and Meta has a genuine track record of converting compute into revenue per user. They are also unauditable from the outside on a quarterly cadence.

Microsoft can say "look at the $625 billion." Amazon can say "look at the $364 billion." Alphabet said "look at the $514 billion" and still fell 5 percent. Meta has to say "trust the ads," on the same day Microsoft reports, in a market that just demonstrated it is grading on cash conversion. If any of the three gets punished disproportionately this week, my money is on the one with nothing contracted to show.

Our Take

The bubble argument and the demand argument have been talking past each other for a year, and Alphabet's quarter is useful precisely because it refuses to settle either one. Demand is obviously real. Cloud does not accelerate to 82 percent on vapor, and margins do not go from 20.7 to 35.6 percent while a company is stuffing an empty warehouse. The bears do not have a demand story anymore.

What they have instead is a duration story, and it is a better one. The revenue arrives over years. The concrete, the transformers, and the accelerators are paid for now. Depreciation on all of it starts immediately and runs on schedules the companies themselves set. In between sits a financing gap that Alphabet just filled with $70 billion of fresh capital in a single quarter, and Alphabet has the strongest balance sheet in the group.

At TensorFeed we track this from the serving side, which is why the capex cycle matters to us at all. Capacity that gets built is capacity that eventually shows up as headroom, lower prices, and fewer rate-limit errors. Capacity that gets delayed because a CFO blinked shows up as throttling and degraded tiers six months later. The financing conditions of 2026 are the availability conditions of 2027.

Three things I am watching this week.

First, whether Microsoft volunteers a backlog figure excluding OpenAI rather than waiting to be asked. Disclosing the less flattering number is a sign of confidence; leaving analysts to reverse-engineer it is a sign of the opposite.

Second, whether any of the three follows Alphabet into negative free cash flow, and whether the market punishes the second one as hard as the first. The first negative quarter is a shock. The second is a category.

Third, whether Meta raises the top of its capex range without a contracted revenue line to point at. That would be the most confident single act of the week, and the one with the least external evidence behind it.

Alphabet did not miss. Alphabet delivered a quarter almost nobody thought it could deliver, and the market moved anyway, because the thing being measured changed while the companies were busy hitting the old target. Growth was the 2025 exam. Cash conversion is the 2026 exam. Three more sit it this week.