AWS Capacity Sold Out Through 2027 as Jassy Calls AI Demand Insatiable

Two hundred and twenty billion dollars. That is roughly what Amazon said earlier this year it would spend on capital projects in 2026, and its chief executive now says it still is not enough. AWS capacity sold out is the phrase investors heard on Friday, when Amazon CEO Andy Jassy said the cloud unit has already committed most of its computing power through 2027, with a good chunk of 2028 spoken for as well.

Jassy made the comments on a video podcast with Andreessen Horowitz co-founder Ben Horowitz and general partner Erik Torenberg, according to Reuters. He described demand from AI labs and large companies as "insatiable." The remarks land less than a week before Amazon reports its third quarter results.

The figures Jassy and Amazon put on the table

• $220 billion: Amazon's planned capital spending for 2026, as stated earlier this year and cited by Reuters.
• 2026: fully sold out, according to The Information's account of the podcast.
• 2027: "largely sold out," in Jassy's words, meaning the big blocks of capacity are booked.
• 2028: "a good chunk" already sold, though no percentage was given.
• October 29: the date Amazon reports third quarter earnings.

Reading between the lines on 2027 and 2028

"Largely sold out" does not mean a customer today gets turned away. Jassy said Amazon always keeps some capacity in reserve for on demand use and for two of its core services: EC2, which rents out cloud servers, and Bedrock, which gives customers access to AI models for building their own apps.

What it does mean is that the huge, multi year deals, the kind AI labs sign to train and run frontier models, have mostly been claimed. A company hoping to lock in a large slab of AWS compute for 2027 is now competing for leftovers.

That fits a pattern across the industry. Cloud providers have been racing to add data centers, chips and power, and the money required keeps growing. We saw the same pressure in SpaceX's $40 billion debt raise for Nvidia chips and in the way OpenAI spread its models across rival clouds, including Google Cloud's Model Garden.

Why $220 billion might grow

The most telling line from the podcast was not about capacity. It was about spending. Jassy said Amazon "could invest a lot more and it wouldn't be a bad use of capital," according to Reuters.

When a CEO says that a week before earnings, analysts listen for one thing on the call: a higher capex number. TipRanks noted that investors will be watching both how fast AWS is growing and whether Amazon raises its spending budget.

There is a logic to it. If every server Amazon can build in the next two years is already rented, each new data center is close to guaranteed revenue. The risk sits further out. Big AI customers sign long contracts based on their own growth forecasts, and those can change. Investors got a reminder of how quickly sentiment can swing when CoreWeave's stock plunged as bond yields climbed.

What a sold out cloud means outside Amazon

For most people, AWS is invisible. It runs behind streaming apps, banking sites and thousands of business tools. A tight supply of new capacity is unlikely to change how those services work tomorrow, since existing customers already have their contracts.

The bigger effects are slower and harder to see:

• Startups building AI products may find it harder or pricier to secure large amounts of compute, pushing some toward smaller providers.
• Rivals such as Microsoft and Google gain an opening if Amazon cannot serve every big deal.
• Communities near new data center sites will keep seeing construction, a trend that has already drawn political attention, including the White House push for data center dividends for local residents.

The number to watch on Thursday

Jassy gave plenty of color but few hard figures beyond the $220 billion plan. Thursday's report should fill in the gaps: AWS revenue growth, the size of its contract backlog and any updated spending forecast. If Amazon raises its capex target, it would confirm what Jassy hinted at on the podcast, that the company sees more risk in building too little than too much.

For more on the companies shaping the cloud and AI buildout, see our tech coverage.