Following the Money: Who Pays for Intelligence

On 28 October 2026 Microsoft reported revenue of about $92.3 billion for the quarter to September, up 19%. Azure, its cloud, grew 48%. Its commercial remaining performance obligation, the revenue customers had contracted for but not yet received, stood at about $677 billion, up 73% in a year, and a large share of it came from two model labs: OpenAI's ~$250 billion commitment to buy more Azure computing and Anthropic's ~$30 billion (company and reported). Microsoft's cash additions to property and equipment in the quarter were about $39.4 billion (reported, verify). In the same reporting season Amazon said that AWS revenue had risen 39% to about $46.0 billion, with a backlog of about $578 billion, while about $220 billion of 2026 cash capex had pushed Amazon's trailing free cash flow to about −$42.4 billion (company). Google Cloud's revenue more than doubled to about $30.7 billion, with a backlog of about $561 billion, and Alphabet's quarterly free cash flow turned negative, at about −$9.1 billion, on about $51.4 billion of capex (reported, verify). Meta, which sells no cloud capacity, spent about $34.9 billion in the quarter and reported free cash flow of about −$2.1 billion (company).

Two days before Microsoft's report, on 26 October, Anthropic filed its registration statement for a public listing. Coverage of the filing reported revenue of $4.585 billion in 2025 and $16.286 billion in the first half of 2026, and compute obligations of about $265.6 billion in fixed or minimum commitments plus about $321.4 billion of leases not yet commenced (from coverage of the filing; the widely quoted ~$518 billion total is not in it). Much of that commitment was the other side of the cloud backlogs that Amazon, Google and Microsoft had just reported.

The numbers describe a chain. A lab promises to buy hundreds of billions of dollars of computing; a cloud records the promise as backlog and borrows or spends its cash to build data centres; the data centres are filled with chips from Nvidia and others; the chips are made in Taiwan and run on electricity from regional grids. At the top of the chain are the people and companies who actually pay for answers, and in the middle, filling the gap between what they pay today and what is being spent, are investors. This chapter follows one query through that chain to show who earns what, and why the money settles where it does.