Hyperscaler spending now trails backlog and outruns cash flow

Four hyperscalers spent about $130 billion in one quarter. Contracted backlog above $1 trillion backs it, while free cash flow thins fast.
Alphabet, Amazon, Meta and Microsoft put about $130 billion into property and equipment in the first three months of 2026. Their results on 29 April showed Alphabet's spending up 107% to $35.7 billion, while Meta raised its full-year range to $125 billion to $145 billion and cited higher component prices.
What changed is that the demand behind the spending is now on paper. Microsoft's commercial remaining performance obligation, the contracted revenue it has yet to book, rose 99% to $627 billion, and Google Cloud's backlog nearly doubled in a single quarter to more than $460 billion. The build-out is no longer a bet on future demand. It is a race to deliver capacity that customers have already signed for, and that shifts the risk for capital to a narrower question: who carries the cost until contracts turn into revenue.
Amazon shows who carries it today. Its trailing twelve-month free cash flow fell to $1.2 billion from $25.9 billion a year earlier, as spending on AI absorbed almost all the cash its operations produced. Meta's remark on component prices shows where pricing power sits for now, with the suppliers of parts and power that cannot expand as fast as orders arrive, the pattern that first paid the chip supplier.
Compute will stay tight for the rest of 2026, which turns efficiency into a competitive advantage rather than a cleanup task. A product that does the same work with fewer tokens, smaller models or better caching keeps its margin when capacity is scarce and prices stop falling. Software that raises utilization inside data centers, from scheduling to power management to inference serving, sells straight into a capital budget of hundreds of billions of dollars. The backlogs also show how AI capacity is now bought, on multi-year commitments, and a founder whose product becomes a dependable workload should price it the same way, on contracts rather than monthly seats.
A backlog is only as strong as the customer's ability to fund it. How much of this one rests on a handful of model labs is the open question, and concentration is the weak point of the whole case. Cash is the other: if free cash flow at the four spenders turns negative and capital spending moves onto debt, credit markets enter the AI cycle. Component prices in next quarter's guidance will be the clearest sign of whether suppliers keep capturing the margin.
The AI build-out has moved from forecast to contract, and the risk has moved with it, from demand to concentration and financing. Fanaura's attention goes to the layers every one of these contracts needs, where scarcity is real and ownership thin, and to the applications that turn this capacity into revenue. The orders are signed. What remains open is who pays until they are delivered.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building software that gets more useful work out of every GPU and every watt? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.