News · 2026-10-24

Andy Jassy says AWS’s big 2027 capacity blocks are largely booked

Amazon chief executive Andy Jassy said Amazon Web Services’ large capacity blocks for 2027 are largely committed and that a good portion of 2028 is already booked. He made the statement in an October 23 a16z interview, while explicitly saying AWS preserves capacity for on-demand use. The claim signals pressure on large advance infrastructure purchases, rather than the disappearance of ordinary cloud access.

Key facts

• Jassy said 2027 was largely booked for big capacity blocks, with part of 2028 also sold.
• He explicitly preserved an exception for on-demand use and services including Bedrock and EC2.
• The statement came October 23; Amazon’s Q3 results call is scheduled for October 29.
• Primary source: the a16z Show interview with Jassy.

“2027 is largely sold out,” Jassy said, before explaining the capacity retained for other uses. The qualifiers are central to the announcement’s meaning. He discussed the large chunks that customers reserve ahead of time, not a published exhaustion date for every service, chip, or region. The public interview supplies no percentage, total contract value, or buyer list behind the claim.

A hotel analogy makes the distinction concrete. A hotel can have its big conference spaces booked far into next year while holding rooms for individual travelers. The conference schedule tells a corporate event planner something important without telling a person seeking one room that nothing is available. Large infrastructure reservations and on-demand cloud usage are similarly different purchasing decisions, even though they draw on a shared physical buildout.

The background is a timing mismatch between spending and revenue. In his 2025 shareholder letter, Jassy describes paying for land, power, buildings, servers, chips, and networking before the company can bill customers for the resulting capacity. Depending on the component, the lead can be six to 24 months. Customer commitments can make some of that spending more predictable, but a reservation is not the same as infrastructure already delivered and earning revenue.

That earlier letter says AWS added 3.9 gigawatts of power capacity during 2025 and expected to double total power capacity by the end of 2027. It also describes unserved demand and advance commitments for several generations of Amazon’s own chips. These are Amazon’s descriptions of its supply pipeline, rather than an independent audit or a detailed allocation schedule for the new interview statement. They explain why a booking several years out can coexist with a company still building aggressively.

The AI connection is direct but needs a boundary. Jassy linked the demand pressure to AI labs and enterprise customers, while the quoted capacity claim concerns AWS more broadly. The material reviewed does not divide the reservations into model training, model serving, or non-AI computing. Nor does it tell a buyer which particular machine type will be available in a particular region on a particular date.

For model developers, long-lead reservations can affect when a large training run becomes practical. For enterprises, the question can be whether a planned deployment has a dependable capacity path. Smaller customers may instead care about service availability and response time, which this statement does not measure. Our lesson on queueing and tail latency explains why an available service can still behave differently under load; neither an advance booking claim nor a growth plan predicts those delays by itself.

The strongest bullish reading is that AWS sees enough committed demand to support an expensive, multiyear buildout. The strongest counterargument is that qualitative reservations reveal little about eventual utilization, contract flexibility, customer concentration, or returns on capital. Booked capacity can be a useful business signal without settling whether every dollar of expansion will earn an attractive return. The interview does not disclose the contractual details needed to answer that question.

The dossier’s social intake shows a Reddit headline circulating the stronger formulation that AWS AI capacity is sold out through 2027. Its metadata does not establish commenter sentiment or verify that broader wording. No independent customer account or infrastructure expert assessment of this specific statement was verified. Jassy remains the primary on-record source, and Amazon’s own documents provide the supporting context.

The next dated checkpoint is Amazon’s October 29 earnings call, which had not occurred as of this briefing. It may provide updated spending and demand commentary; it cannot retroactively be the source of the October 23 interview quote. The practical implication today is to distinguish large-block procurement pressure from service-wide scarcity, then seek commitments for the actual workload, location, and delivery date a project requires.

Primary source, verified: read the paper →

Ground Truth. (2026, October 24). Andy Jassy says AWS’s big 2027 capacity blocks are largely booked. Ground Truth. https://groundtruth.day/news/aws-jassy-2027-large-capacity-blocks-reserved.html

@misc{groundtruth:aws-jassy-2027-large-capacity-blocks-reserved,
  title  = {Andy Jassy says AWS’s big 2027 capacity blocks are largely booked},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/aws-jassy-2027-large-capacity-blocks-reserved.html}
}

Topics: infrastructure · aws · compute · capacity · industry