What "sold out" covers, and what it does not

Start with the sentence itself, because the headline version drops half of it. Jassy did not say AWS has no capacity. He said the big chunks are largely sold out, and in the same breath named the reserve: a certain amount is always left for on-demand work and for the inference services, Bedrock and EC2, that customers use without a forward commitment.

That distinction decides who should care. For a lab or enterprise seeking a large, guaranteed block of AI capacity in 2027, Jassy's remarks are a warning to start capacity discussions early. He said the large forward blocks were largely sold out, but gave no figures showing what remained available to individual customers, regions or services. Customers using on-demand services are not necessarily excluded: Jassy said AWS deliberately reserves some capacity for on-demand use and for Bedrock and EC2. He did not disclose the size of that reserve or guarantee availability for any particular workload. Conflating the two produces either panic or dismissal, and the quote supports neither.

What the report does not contain is a denominator. There is no figure for how much capacity exists, how much was booked, or what a "big chunk" weighs in any unit a reader could check. Sold out is a statement about a reservation book Amazon does not publish. The dates and Amazon's previously stated capex plan are checkable. Amazon's 29 October earnings report is the next opportunity to look for additional evidence about AWS growth, capital spending and capacity constraints, although it may not quantify the forward bookings Jassy described.

Why a seller would say this five days before earnings

The timing is part of the information. Jassy chose a long, friendly podcast format, not a filing, and he chose the week before results. A statement like this does three jobs at once. It tells large customers to commit earlier, which is convenient if you are selling forward capacity. It tells investors that the $220 billion spending plan has demand behind it, which is the question that plan keeps raising. And it sets the frame for the earnings call: if AWS growth disappoints, the explanation on offer is supply, not demand.

None of that makes the statement false. It makes it a claim with an author and an interest, which is how the Trainium figures in the fuller account should be read as well. Thirty to forty percent better price-performance for Amazon's own chips, against comparators Jassy did not name, in a podcast, is marketing until a customer or a benchmark publishes the same comparison. BigGo Finance reports that Jassy cited more than $225 billion in future Trainium revenue commitments, including agreements with Anthropic and OpenAI. That is a figure attributed to Jassy through a secondary account of the podcast, not an independently verified total of recognized revenue or delivered capacity.

There is also a plainer reading, and the piece should not lose it. If inference really is, as Jassy put it in remarks reported by BigGo Finance, an order of magnitude too expensive for the broad middle of enterprise software, then the capacity that is sold out is mostly spoken for by the two ends of his own described market: the labs and the fast-growing AI applications.

Sources

  1. [1] The a16z Show, "Amazon CEO Andy Jassy on the $220 Billion AI Buildout" — the video podcast itself, with Ben Horowitz and Erik Torenberg, published 23 October 2026 (the remarks are quoted in this article as reported by Reuters)Read source
  2. [2] Reuters, "Amazon cloud capacity sold out through 2027, CEO Jassy says", 23 October 2026 — primary reporting for the sold-out statement, the on-demand/Bedrock/EC2 qualifier, the about-$220 billion 2026 capex figure, the "could invest a lot more" line and the 29 October results date; read via syndicationRead source
  3. [3] BigGo Finance, "Amazon's AWS Cloud Capacity Sold Out Through 2027 as AI Demand Outpaces $220 Billion Buildout", published 24 October 2026, updated 24 October 2026 — fuller account of the same podcast: the inference-cost remark, Trainium/Graviton price-performance claims, the more-than-$225 billion Trainium commitments figure, and the AWS revenue figures, all attributed there to Jassy or to Amazon's second-quarter reportRead source
  4. [4] Vector Wire, "AWS AI Cloud Capacity Sold Out Through 2027, Jassy Says", 24 October 2026 — independent restatement of the Reuters core: 2026 effectively sold out, 2027 largely, 2028 commitments, on-demand/Bedrock reserveRead source