‘Largely sold out all the big chunks’
Amazon has already sold most of the cloud capacity it will have in 2027. “2027 is largely sold out,” chief executive Andy Jassy said on Friday. “We always leave a certain amount for on demand and our inference service bedrock, or EC2, but largely sold out all the big chunks. A good chunk of 2028 is sold out.”
Jassy made the comments in a video podcast with Andreessen Horowitz co-founder Ben Horowitz and general partner Erik Torenberg, reported by Reuters. Capacity for 2026, he said, is already gone.
EC2 is the Amazon Web Services product that rents cloud servers for running applications. Bedrock is the service through which customers reach AI models. Between them they are the two things Jassy named as the exceptions AWS deliberately holds back for customers who want capacity on demand rather than committing years ahead.
$220bn, and the demand is still ahead of it
The figure that gives the statement its weight is Amazon’s own capital spending. The company said earlier this year it would spend about $220 billion on capital expenditure in 2026. Jassy’s position is that this is not too much: the company “could invest a lot more and it wouldn’t be a bad use of capital,” he said, given that 2026 capacity is already committed.
That is a chief executive saying, with the largest capital programme in the company’s history already running, that the constraint is supply rather than demand. It is also a claim a customer cannot check. AWS does not publish capacity, so “largely sold out” is a characterisation rather than a disclosed number, and the Reuters account carries no figure for how much business AWS is turning away or what it plans to spend in 2027.
Forward commitments are the new disclosure
What Jassy is describing is a market that has moved from buying compute to reserving it. Capacity sold two and three years out is revenue booked against data centres that are not finished, chips that are not delivered and, in many cases, grid connections that are not energised.
For AWS that visibility is the argument for spending. For everyone else it narrows the market: if the large blocks for 2027 are committed and a good chunk of 2028 with them, a company that has not already signed is negotiating for what is left, or for 2029.
What to watch
Amazon reports third-quarter results on 29 October. That is the first place the claim meets numbers — AWS growth, backlog and, above all, whether the 2026 capital figure holds or rises for 2027.
Jassy’s framing will be tested there. A company that is genuinely supply-constrained and says more capital “wouldn’t be a bad use” tends to commit more capital, and the guidance is the test of whether the sold-out language is a constraint or a sales position.
