Amazon chief executive Andy Jassy put a striking figure on the company’s in-house silicon effort during the first-quarter earnings call in April, saying customers have committed more than $225 billion in future revenue for its Trainium artificial-intelligence accelerators. The number signals that Amazon’s custom-chip operation, spanning Graviton processors, Trainium accelerators and Nitro networking chips deployed inside Amazon Web Services, has moved well beyond an internal cost-saving project into a business that Jassy estimates would carry a $50 billion annual revenue run rate if it sold chips on the open market.
As it stands, the unit already generates an annual revenue run rate above $20 billion and grew nearly 40 percent quarter over quarter in the first quarter, expanding at triple-digit percentage rates year over year. Jassy told analysts the custom silicon business now ranks among the top three data-center chip operations globally. Named commitments underscore the breadth of demand: OpenAI has agreed to consume roughly two gigawatts of Trainium capacity starting in 2027, Anthropic will secure up to five gigawatts of current and future Trainium generations, Uber relies on Graviton chips for rider-driver matching and Meta Platforms has signed on to deploy tens of millions of Graviton cores.
Supply remains tight. Jassy said Trainium2, which delivers about 30 percent better price-performance than comparable graphics processors, has largely sold out. Trainium3 began shipping at the start of 2026 and is nearly fully subscribed, while much of Trainium4, still more than a year from broad availability, has already been reserved. The momentum sits inside an accelerating cloud business: AWS revenue reached $37.6 billion in the first quarter, up 28 percent year over year and marking the segment’s fastest growth in 15 quarters. AWS operating income rose 23 percent to $14.2 billion from $11.5 billion a year earlier.
Nvidia remains the dominant force in AI data centers, and Amazon itself plans to deploy more than one million Nvidia GPUs beginning in 2026. Trainium’s pitch is cost per unit of compute, and the chips are available only through AWS. Jassy’s $50 billion figure is hypothetical, it assumes a merchant-silicon model Amazon does not currently pursue, and the company does not break out the unit’s profitability, leaving investors without a clear view of margins.
The spending required to sustain this trajectory is substantial. Amazon expects roughly $200 billion in capital expenditures across the company in 2026, and free cash flow for the trailing twelve months fell to $1.2 billion from $25.9 billion a year earlier as AI investments ramped. If demand for AI computing cools before those outlays pay off, profits and the share price could come under pressure. At about $255 per share, Amazon trades near 30 times earnings, a multiple that does not appear to price in a stand-alone chip business of this scale.