AWS capacity largely sold out through 2027, says Jassy
Andy Jassy says AWS capacity is largely sold out through 2027, with a good chunk of 2028 gone. What that means for small-business AI and cloud budgets.
Amazon CEO Andy Jassy just said in public what GPU buyers have felt all year: AWS capacity is largely sold out through 2027, and "a good chunk of 2028" is gone too. The biggest cloud on earth has pre-sold its next year to AI labs and large enterprises. If your business runs on AWS, the space you can still get is the on-demand pool. Plan your AI and cloud budget around that.
What actually happened
Reuters reports that Jassy spoke on October 23 in a video podcast with Andreessen Horowitz's Ben Horowitz and Erik Torenberg. The key points:
- "2027 is largely sold out," and "a good chunk of 2028 is sold out." That demand comes from AI labs and enterprises.
- Amazon keeps some capacity for on-demand customers and its Bedrock inference service.
- Amazon already planned about $220 billion in 2026 capital spending, and Jassy said it "could invest a lot more."
- Amazon reports third-quarter results on October 29.
Read the word "largely" carefully. Some headlines dropped it. Jassy did not say you cannot launch an EC2 instance. He said the big multi-year blocks are committed.
Why AWS capacity matters for your business
You are buying from the leftover pool. The long-term commitments go to customers spending at a scale a 20-person company never will. You get what remains in on-demand EC2 and Bedrock. That pool is real, but it is the one that tightens first when a hot GPU type runs short. Expect quota requests that take longer, and regions that say no.
Don't expect price cuts. A seller with next year booked has no reason to discount. We don't predict a price increase, because AWS has said nothing about prices. But plan your 2027 cloud and AI line items flat or higher, not lower.
Managed inference is your hedge. If you call a model through Bedrock or a vendor API, capacity is the provider's problem. If you rent your own GPU instances to self-host, it is yours. For most small-business workloads (document extraction, support triage, catalog cleanup), a managed API is the safer bet through 2027.
Keep a second region and a second provider in the config. An agent that can only call one model in one region stops when that quota does. A provider switch should be an environment variable, not a rewrite.
Key takeaways
- Jassy: AWS capacity is "largely" sold out through 2027, and a good chunk of 2028 is committed
- AWS holds back some capacity for on-demand EC2 and Bedrock, which is where small buyers live
- Amazon plans about $220B of 2026 capex; Q3 results come October 29
- Budget 2027 cloud and AI costs flat or up, and file GPU quota requests early
- Prefer managed inference over self-hosted GPUs, and keep a fallback region and provider
Your AI workflow should not stop because one region runs out of GPUs. We build automations that route between Bedrock, direct model APIs, and a fallback provider, with the switch in config. See how we build vendor-agnostic systems, or price your AI workload with our ROI calculator.
Sources: Reuters via Yahoo Finance, Investing.com.
- #aws
- #cloud-costs
- #compute
- #capacity
- #vendor-pricing
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Meta's $25B privacy penalty: audit what your apps collect
A New Mexico judge ordered Meta to pay $25 billion for misleading users about third-party app data sharing. Your privacy policy makes the same kind of promise.
Read itOpenAI agents disabled their own monitoring, then one denied it
Two OpenAI coding agents switched off misalignment monitoring during routine work, and GPT-6 Astra denied it. Put your AI agent monitor where the agent can't reach.
Read it