AWS CEO Matt Garman defends AI spending with a spread-out customer base

Amazon Web Services is about to spend an enormous amount of money on AI infrastructure. CEO Matt Garman wants everyone to know it is not betting the farm on one or two customers to pay for it.

In an October 8, 2026, appearance on the a16z podcast, Garman said AWS’s largest single customer accounts for only a single-digit percentage of total revenue.

What Garman actually said

The headline number is capital expenditure. Garman projected AWS capex of around $220 billion for 2026.

Garman said AWS is setting aside cloud capacity for emerging startups while continuing to serve large enterprises. Startups make up an estimated 30-40% of AWS revenue.

The growth numbers behind the confidence

Garman pointed to an annual revenue figure for AWS estimated between $169 billion and $170 billion. That reflects a 37% year-over-year increase.

By the first quarter of 2026, the AWS AI revenue run rate had topped $15 billion, with projections calling for more than $25 billion later on.

Amazon Bedrock, AWS’s platform for building applications on top of AI models, has posted 170% quarter-over-quarter revenue growth in certain periods.

Capacity on Trainium, Amazon’s in-house AI training chip, is sold out for the next year. Meanwhile, AWS says its Graviton processors offer about 20% better cost-performance for many tasks.

Core workloads still matter

On the enterprise side, Garman pointed to companies shifting AI projects from experiments into full production. According to survey data he cited, 90% of CIOs say they are on track for positive returns on their AI initiatives.

Background: partners and positioning

AWS has positioned itself as a neutral platform in the AI race. It maintains partnerships with both Anthropic and OpenAI, two of the most prominent AI model developers.

The sold-out Trainium capacity suggests the bottleneck right now is supply, not interest.

What this means

For investors, the most important signal is the link Garman draws between spending and signed demand. Garman’s answer has three parts: customer agreements reportedly underpin the spending, revenue is growing 37% year over year, and the client base is spread wide enough that no single customer can sink the plan.

Run rates and projections, including the forecast for AI revenue to exceed $25 billion, are forward-looking and could miss. The 170% Bedrock growth came in certain periods, so it should not be read as a steady quarterly pace. Sold-out Trainium capacity signals strong demand, but it also means some customers may look elsewhere if AWS cannot build fast enough.