Amazon's in-house chip business has quietly become one of the world's three largest data center semiconductor operations, with $225 billion in contracted Trainium revenue and a $20 billion annual run rate.
Back
Amazon chip business hits $20B run rate with $225B in Trainium commitments
Amazon's custom chip business has crossed a $20 billion annual revenue run rate with triple-digit growth, backed by more than $225 billion in customer commitments for its Trainium AI accelerators.
"Our custom silicon business is now one of the top three data center chip businesses in the world," Andy Jassy, chief executive officer of Amazon, said on the company's first-quarter earnings call. If the unit sold its chips externally like a traditional semiconductor vendor, the equivalent run rate would be about $50 billion, he added.
Trainium2 delivers about 30% better price-performance than comparable graphics processing units and has largely sold out, Jassy said. Trainium3, which began shipping early this year, is nearly fully subscribed, while much of Trainium4 — still more than a year from broad availability — has already been reserved. OpenAI committed to roughly two gigawatts of Trainium capacity starting in 2027, and Anthropic secured up to five gigawatts across current and future chip generations. Meta Platforms has deployed tens of millions of Graviton processor cores, and Uber uses Graviton chips for ride-matching.
The chip momentum sits inside an AWS business that grew 28% year over year to $37.6 billion in the first quarter, the fastest pace in 15 quarters, with a 37.7% operating margin. Jassy said Trainium will save Amazon "tens of billions of dollars of CapEx each year" and provide "several hundred basis points of operating margin advantage." At roughly 30 times earnings, Amazon's stock is up about 10% this year — a modest gain for a company whose biggest profit engine is accelerating while a hidden semiconductor franchise scales beneath it.
The Economics of Owning the Stack
Amazon's approach differs from Nvidia's model. Rather than selling chips on the open market, Amazon keeps its silicon inside AWS, where it charges customers for compute time. The company also continues buying Nvidia GPUs — it announced plans to deploy more than 1 million Nvidia chips starting in 2026 — meaning Amazon profits whether customers run workloads on Trainium or Nvidia hardware inside its cloud.
The financial stakes are large. Amazon expects about $200 billion in capital expenditures across the company in 2026, and its free cash flow for the trailing 12 months fell to $1.2 billion from $25.9 billion a year earlier as AI investments ramped up. The AWS commercial backlog stood at $364 billion entering the first quarter, excluding the recent $100 billion-plus Anthropic expansion — a cushion that reduces the risk that the spending outruns demand.
For investors, the chip business strengthens an already compelling case. Amazon trades at about 30 times earnings, though that multiple is flattered by a $16.8 billion pre-tax gain on its Anthropic investment. Even excluding that gain, the valuation is reasonable for a company whose cloud business is growing at its fastest pace in 15 quarters while a $20 billion chip operation with triple-digit growth and $225 billion in commitments runs inside it. Trainium does not need to beat Nvidia for Amazon shareholders to win — it just needs to keep saving the company money and locking in customers.
This article is for informational purposes only and does not constitute investment advice.
