Amazon’s custom silicon division, which designs the Trainium AI accelerator, Graviton server processor, and Nitro networking chips, is now running at an annualized revenue pace of roughly $20 billion, according to disclosures from CEO Andy Jassy. The unit has grown at a triple-digit percentage rate year-over-year, fueled almost entirely by demand for Trainium chips used to train and run AI models.
Jassy said the business already holds more than $225 billion in multi-year revenue commitments, describing the figure as contracted demand rather than a forecast. Anthropic accounts for a major share of that backlog, with a multi-year, multi-gigawatt agreement tied to over $100 billion in AWS spending. OpenAI has separately committed to roughly two gigawatts of Trainium compute capacity.
Amazon’s second-generation Trainium2 chip, which offers about 30% better price-performance than comparable GPUs, is largely sold out. Trainium3, which began shipping in early 2026, delivers a further 30-40% price-performance improvement over its predecessor and is already close to fully subscribed, according to the company. Even Trainium4, not expected for broad availability for another 18 months, has attracted significant advance reservations.
Jassy noted that if Amazon’s chip unit operated as a standalone company selling silicon the way traditional chipmakers do, it could generate around $50 billion annually, placing it among the top three data center chip businesses in the world. The disclosures come as Amazon continues to position custom silicon as a way to reduce dependence on Nvidia GPUs and control costs across its AWS cloud infrastructure.
Sources: The Motley Fool, The Register
