Amazon Trainium over $20B run rate

Amazon’s AI chip story just got a lot more concrete. On April 29, Amazon said its chips business — Graviton, Trainium, and Nitro together — is now running above $20 billion a year. That matters because Trainium is no longer a side project inside AWS. It is turning into the hardware layer Amazon wants to use to power AI growth without handing all the economics to Nvidia. (ir.aboutamazon.com) ### What actually crossed $20 billion? Not Trainium by itself. Amazon’s reported figure covers its broader custom silicon business, which includes Graviton CPUs, Trainium AI accelerators, and Nitro networking and security chips. Andy Jassy said that business grew nearly 40% quarter over quarter in Q1 and is still growing triple digits year over year. (aboutamazon.com) ### So why is everyone talking about Trainium? Because Trainium is the part tied most directly to the AI spending boom. Amazon has been pitching Trainium as the cheaper, more controllable alternative to renting giant clusters of Nvidia GPUs. In the same earnings win(aboutamazon.com)b experiments into serious capacity planning. (convergedigest.com) ### Where does Bedrock fit in? Bedrock is the software and services layer that makes the chip push matter. It lets companies use foundation models through AWS without building everything from scratch, and Amazon said Bedrock customer spend grew 170% quarter over quarter. Basic(convergedigest.com)on its own silicon underneath. (convergedigest.com) ### What did Jassy mean by a $50 billion business? This is the eye-catching part — and also the part that needs translation. Jassy said that if Amazon’s chips business were treated like a standalone vendor and sold this year’s chip output to AWS and outside customers the way m(convergedigest.com)t-if framing that includes Amazon’s own internal consumption. (aboutamazon.com) ### Why does internal consumption matter so much? Because AWS is both the factory floor and the customer. Amazon designs chips, deploys them in AWS regions, and then sells AI compute and services on top. That means a lot of the value shows up as cloud revenue, not a(aboutamazon.com)even while it may be strategically bigger inside Amazon’s stack. That last point is an inference from how Amazon bundles infrastructure and cloud services. (ir.aboutamazon.com) ### Is this already moving AWS numbers? Yes. AWS revenue rose 28% year over year to $37.6 billion in Q1 2026 — its fastest growth in 15 quarters — and management tied the acceleration to AI demand. Amazon also said AWS is now running at a roughly $150 billion annualized revenue pace. Trainium is not the whole reason for that jump, but it is clearly part of the engine. (ir.aboutamazon.com) ### What is Amazon really trying to do here? Reduce dependence on Nvidia without giving up AI scale. Nvidia still matters a lot — Amazon said it plans to deploy more than 1 million Nvidia GPUs starting in 2026 — but custom silicon gives AWS more control over cost, supply, and system design. In cloud infrastructure, that control is the margin story. (convergedigest.com) ### Bottom line? The news is not that Trainium alone is suddenly a clean $20 billion business. The news is that Amazon’s custom silicon operation has gotten big enough to matter financially, and Trainium is now central to how AWS plans to win the next phase of AI infrastructur(convergedigest.com) economics. (ir.aboutamazon.com)