A beat on every line, Azure growing faster, investors pleased. And the only AI revenue figure any of these companies reports was quietly left out of the call.
| Line | Expected | Delivered |
|---|---|---|
| Revenue | $87.6B | $90.0B (+18% on a year earlier) |
| Earnings per share | $4.24 | $4.81 ($4.74 adjusted) |
| Azure growth on a year earlier | 40% (previous quarter) | 43% +43% vs +40% required |
| Azure next quarter, own guidance | 41.4% (analysts) | 45% at constant exchange rates |
| Capital spending in the quarter | — | $41B; ~2/3 on short-lived chips (CPUs/GPUs) |
| Free cash flow | — | $19.6B |
| AI revenue, yearly rate | $37B, +123% (last results) | not updated; the figure was absent from the call |
Capital spending (capex) is money spent on things that last for years: chips, buildings, power. Free cash flow is the cash left over after that spending. Full financial year (Microsoft's year ends in June): revenue $331B (+18%), Microsoft Cloud $214B (+27%), Azure above $100B (+41%). 88 data centres added in the year; computing capacity on track to roughly double in two years. $10.2B paid out to shareholders in the quarter.
The evidence of demand was clear. Azure, Microsoft's cloud business, grew 43% on a year earlier, up from 40%, and now brings in more than $100 billion a year. Microsoft's own guidance is 45% growth next quarter. Growing faster while already that large is the hardest thing to do in software. Copilot, Microsoft's AI assistant, added signals of its own. Microsoft 365 Copilot passed 30 million paid users, and the net number added in the quarter was more than double the quarter before. Copilot revenue grew more than 60% on the previous quarter after Microsoft began charging by use. 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). Azure is above that line, but only just. It is the narrowest margin above the line of any company reporting this quarter.
The test this site set for these results was Microsoft's own AI revenue figure: $37 billion a year at the last count, growing 123% on a year earlier. This site said in advance that growth below roughly 80–90% would be the moment the required growth path started to break. The number did not slow. It vanished. The full transcript of the results call contains no update to it. The only AI revenue figure any of these companies reports was simply not refreshed, one quarter after it became the number everyone watched. There are two ways to read that. The kind one: Azure itself is now the AI number, and Azure grew faster, so the old figure was no longer needed. The sceptical one: companies stop reporting a figure when it stops looking good, and 123% growth on a base of $37 billion is almost impossible to keep up. Either way, a test that was set in advance, and that could have failed, was made impossible to run by the company that holds the data. In a boom priced on trust in the numbers companies promise for the future, that is itself a piece of evidence.
Alone among the big firms reporting this week, Microsoft did not raise its capital spending plan: $41 billion in the quarter, and a full-year plan it called "steady". Free cash flow, the cash left over after capital spending, was $19.6 billion: positive, and large. Alphabet pays for its build-out partly by borrowing. Meta pays for its own with nearly every dollar of cash its advertising business brings in. Microsoft pays for its own out of its own cash, with room left over to pay dividends. One caution belongs on the record. Microsoft is changing how it accounts for future data-centre leases, from one kind of lease (finance) to another (operating). That makes the reported capital spending figure look smaller without changing the cash that goes out. The next "steady" needs reading with that in mind.
One disclosure speaks directly to the front page's central mechanism. Roughly two-thirds of the quarter's capital spending went on short-lived equipment: CPUs and GPUs, the processing chips, which are written off over three to five years. That confirms, from inside the company, what the front page calls the depreciation wall: most of the sector's record spending buys hardware that has to be replaced within five years. That is why the front page's model treats 2027–29 as the years when a second round of spending has to be paid for out of revenue rather than out of investors' faith.
Azure is infrastructure: computing power sold on to others, one step removed from the customers who finally pay for AI. The vanished AI revenue figure was the closest thing to a measure of what those customers pay. Without it, the proof has to come from indirect evidence: numbers of paid users, charges by use, and in the end the jobs figures. The question the front page asks, whether AI is paid for from software budgets or from wages, got no closer to an answer. It lost its best measuring instrument.
Meta reported the same evening (full analysis): the bottom of its spending range rose and free cash flow fell close to zero. Apple and Amazon reported on 30 July: Apple, the control group, the big tech firm with no AI build-out of its own, and Amazon, where demand was proven and the spending bill rose with it. The July US jobs report, the other side of the ledger (do the jobs figures move with the results?), lands on 7 August. Nvidia, which sells the chips and so sees the spending from the supplier's side, reports on 26 August.