AWS Eyes Third-Party Trainium Sales in Challenge to Nvidia
Amazon Web Services is in talks to sell its Trainium AI chips to outside companies for use in their own data centers. That's a notable expansion for a chip line Amazon built to serve its own infrastructure.
Jassy's $50 Billion Figure
In his April 2026 shareholder letter, CEO Andy Jassy put a number on the potential: roughly $50 billion annual run rate if the Trainium business served both AWS and third-party customers. That's the hypothetical, not the current state. The talks are ongoing.
Demand signals are real, though. Current Trainium capacity sold out almost immediately after availability. Trainium4 has already sold out before it ships, with availability still more than a year out.
The Nvidia Gap
Nvidia is running at $326 billion annual revenue. Jassy's $50 billion figure represents the upside scenario for a business that doesn't exist yet at commercial scale. The gap is large.
Nvidia is also expanding. CEO Jensen Huang recently identified a new $200 billion opportunity in CPUs for AI workloads, beyond the GPU business that built the company. Both companies are moving outward at the same time.
Supply Chain
TSMC manufactures Trainium chips for AWS. AWS also added OpenAI to the models served on its cloud platform, putting OpenAI's models alongside Amazon's chip infrastructure.
The Open Question
Selling chips to yourself and selling chips to competitors who run competing infrastructure are different businesses. Third-party customers have options. AWS captured its own demand by building Trainium into its own stack. Convincing external data center operators to adopt Trainium over established alternatives is a different sales motion entirely.
The sell-out demand is a useful signal. It is not a contract.
Source: Techcrunch