Every line beat analysts' forecasts, and the shares fell anyway. The quarter made a bust less likely, made the bill the AI spending must one day earn back bigger, and moved a clock that had not moved before: investors' patience.
| Line | Expected | Delivered |
|---|---|---|
| Revenue | $116.9B | $119.8B (+24% on a year earlier) |
| Earnings per share | $2.89 | $9.11, incl. ~$98B of paper gains on shares it holds |
| Google Cloud | $22.2B (+64%) | $24.8B (+82%) +82% vs +40% required |
| Cloud orders signed, not yet delivered | $462B (Q1) | $514B |
| Operating margin | 35.3% | 34% |
| 2026 capital spending, own guidance | $180–190B | $195–205B |
| Free cash flow | — | −$5.9B |
Capital spending (capex) is money spent on things that last for years: chips, buildings, power. It was $44.9B in the quarter, more than double a year earlier. Free cash flow is the cash left over after that spending; it went negative. The company has raised roughly $100B this year by borrowing and selling shares; it holds $242.5B in cash and investments it can sell quickly.
The test this site set for these results was simple: was Google Cloud growing faster again? Faster growth there is the best evidence that real customers want what the spending builds. The answer was yes, by a wide margin. Cloud revenue grew 82% on a year earlier, against the 64% analysts expected, and at the current quarterly rate brings in about $100 billion a year. Orders signed but not yet delivered reached $514 billion, roughly five years of cloud revenue already under contract. Usage points the same way: Google's own AI models handled 22 billion tokens a minute (a token is a fragment of a word, the unit AI models work in), up from 16 billion a quarter earlier. The front page works out that the sector needs roughly 40% growth a year, compounded, to reach the revenue floor by 2030. The revenue floor is the revenue the AI spending must earn every year to pay for itself (the front page's §2). At 82%, this quarter sits well above that line. A bust in the near term, brought on by disappointing demand, became less likely on 22 July.
The first warning sign of a bust would be a company cutting its spending plans. Alphabet did the opposite. Its own guidance for 2026 capital spending rose to $195–205 billion, and it confirmed that 2027 will be significantly higher again. That cuts both ways. It pushes a bust further off, but it also raises the bill. By the front page's arithmetic, every extra trillion dollars of capital spending raises the revenue the sector must earn each year to break even by roughly $450 billion. That revenue has to come from somewhere: in the end, from someone's software budget or someone's wages. At the middle of the new range, Alphabet's capital spending is about 42% of its revenue, 42 cents of every dollar. That is the ratio of an electricity utility, not a software company. Free cash flow has turned negative even at the sector's most cash-rich firm, and the shortfall is being covered by borrowing and selling shares. The bill the sector must one day pay grew this quarter, at the sector's own request.
The signal that matters most came from investors. Alphabet beat almost every measure of its business, and its shares fell 5% in after-hours trading because of the higher spending guidance. It delivered, and was punished for spending. For two years, a bigger spending plan was read as confidence. This one was read as risk. That switch, when spending on infrastructure stops being good news and becomes bad news, is typical of the late stage of an investment boom. It is the first time in this cycle that the patience clock has moved. One quarter does not settle how investors will react from now on, but the reaction has plainly changed.
Firms in the AI boom invest in, buy from and sell to each other, so one firm's spending is another's revenue. This site calls that the loop. Two things in these results belong to it. The headline earnings of $9.11 a share rested on roughly $98 billion of paper gains: rises in the recorded value of shares Alphabet holds, not cash. Reports say much of that came from the rise in the recorded value of its stake in Anthropic ahead of Anthropic's planned listing on the stock market. So an AI lab's rising valuation shows up as profit at a company that funds the lab and sells it computing power. And the quarter's new lines of revenue are infrastructure sold to the AI economy itself. Alphabet began selling its TPU chip systems (most of that revenue will be booked in 2027). At the same time it is renting computing capacity from others to meet demand, including a reported $920 million a month from SpaceX. The boom is showing up more and more inside the very results that are meant to prove it is real.
Cloud revenue is still infrastructure: computing power sold to companies and AI firms. Alphabet is one step further along the chain than the chipmakers, but it is still selling the tools, not the finished work. The $514 billion of signed orders proves that buyers have committed to pay for computing power. It does not yet prove those buyers can earn the money back from their own customers, whether from software budgets or from wages that AI has replaced. So the question the front page asks, whether AI is paid for as a tool or as a substitute for workers, is not answered by this quarter. It has moved one step down the chain. What to watch has not changed: whether the business software firms start reporting revenue priced by the number of workers replaced, and whether the jobs figures ever start to move together with the quarterly results.
These results are now written up: Microsoft and Meta reported on 29 July; Apple and Amazon followed on 30 July. Before its results, Microsoft was expected to report $87.6 billion of revenue and diluted earnings per share of $4.24 (the diluted figure also counts shares that could still be issued). Its AI revenue, which it last put at $37 billion a year and growing 123% on a year earlier, is the closest thing the sector has to a measure of what end customers pay for AI. Growth at that size slows every year, so to stay on course the number has to beat the long-run average by a lot in the early years. A slowdown to 60–70% growth, fast as that still sounds, would be the first sign of the required growth path breaking. Meta reported the same day; the question there was whether it would match the others' higher spending (its own guidance was $125–145B, raised because parts cost more). The Federal Reserve was due to decide on interest rates the same week.