Alphabet spent $51 billion on AI infrastructure in one quarter — and it still isn't enough
Alphabet closed out the third quarter of 2026 by writing the single biggest capital-spending check in its history: $51.4 billion in one quarter, almost entirely on servers, chips, and data centers, more than double what it spent in the same quarter last year. Google Cloud revenue, at $30.7 billion, grew 103% year over year, the third straight quarter of accelerating growth. On paper, that is the AI infrastructure bet paying off exactly as promised. Underneath it, Alphabet also disclosed that it will lean on rented, third-party compute capacity in the fourth quarter because its own buildout can't keep pace with demand.
The spending is working, which is the problem
Cloud's operating margin jumped to 36.2% from 23.7% a year ago, and the unit's backlog — contracted future revenue Alphabet hasn't recognized yet — climbed to roughly $561 billion, up from $514 billion just one quarter earlier. That is a startling figure for a single company's order book: more than the GDP of Austria, growing by tens of billions of dollars every three months. Alphabet says it expects just over half of that backlog to convert to revenue within 24 months, which implies the company has effectively pre-sold compute capacity it has not yet finished building.
That gap between bookings and physical capacity is the real story. Alphabet's finance chief told analysts that Cloud segment margins specifically -- not the company's margins broadly -- will face "near-term pressure" in the fourth quarter because Google is beginning to use third-party compute capacity to serve customer demand — an unusual admission from a company that has spent three years building out its own data centers and custom TPU chips precisely so it wouldn't have to rent someone else's infrastructure. Leasing capacity is typically a lower-margin stopgap than owning it outright, which is presumably why executives flagged it as a drag rather than a flex.
Unchanged guidance, but a warning for 2027
Alphabet left its full-year 2026 capex guidance at $195 billion to $205 billion, which, with $132 billion already spent through nine months, implies a fourth-quarter outlay of $63 billion to $73 billion — another new high. But the more consequential line came when the CFO said capex is expected to "increase significantly" again in 2027, with specifics reserved for the next earnings call. Coming from a company that just quadrupled its quarterly spend compared with two years ago, "significant increase" is not a modest hedge. It's a signal that the arms race among Google, Microsoft, Amazon, and Meta for data center land, power contracts, and chip supply is nowhere near slowing down, regardless of what it does to free cash flow.
And it is doing something to free cash flow: Alphabet posted negative $9.1 billion in free cash flow for the quarter, even with $42.3 billion in operating cash flow, because the infrastructure spend outran everything the business generated. Depreciation, the bill for infrastructure built in prior quarters, rose 49% to $8.4 billion, and total assets crossed $1 trillion for the first time. Alphabet also issued $28.7 billion in new long-term debt during the quarter, pushing long-term debt to $125.1 billion — a reminder that even the most cash-generative company in tech is increasingly financing its AI buildout with borrowed money rather than cash on hand.
The read-through
None of this is distress — Cloud margins are improving, revenue is accelerating, and the backlog suggests demand isn't the constraint. The constraint is physical: power, chips, and construction timelines that can't be hurried the way a software release can. Renting third-party capacity is the clearest evidence yet that even Alphabet's own, famously vertically integrated infrastructure (its own chips, its own fiber, its own data centers) isn't being built fast enough to match what customers are already paying for. For an industry that keeps promising the compute crunch will ease next year, this quarter's numbers point the other way: whatever 2026 cost, 2027 is set up to cost more, on a bigger base, with less room for backing down.