Amazon in talks to sell Trainium AI chips externally; custom silicon business hits $20B revenue run rate
Amazon is in talks with third-party customers to sell its custom Trainium AI accelerators for deployment in their own data centers, marking a strategic shift from AWS-exclusive infrastructure to direct semiconductor sales. Peter DeSantis, Amazon's AI chief, confirmed the discussions but declined to name potential customers. The move would position Trainium chips as a lower-cost alternative to Nvidia GPUs in the hyperscaler market.
Amazon's broader custom-silicon business—spanning Trainium, Graviton, and Nitro chips—crossed a $20 billion annual revenue run rate in Q1 2026 and is growing at a triple-digit pace. Major AI customers including OpenAI (2 gigawatts of Trainium capacity) and Anthropic (up to 5 gigawatts) are already committed. Trainium3, the latest-generation chip, is nearly sold out, underscoring demand strength.
For infrastructure teams, this signals that Nvidia's pricing power in AI training is under structural pressure from credible hyperscaler alternatives. Amazon's willingness to sell chips externally—backed by proven customer relationships and high-volume manufacturing—indicates the market is widening fast enough to support both incumbents and challengers. The calculus for new capacity procurement now includes a second-source calculus around cost and supply terms.
Sources
- Primary source
- finance.yahoo.com
“Amazon is in talks with third-party customers to sell its custom Trainium AI accelerators for deployment in their own data centers”
- intellectia.ai
“Amazon's Trainium business surpassed a $20 billion annual revenue run rate in Q1 2026, growing at a triple-digit pace, with major clients like OpenAI and Anthropic committed to using its chips”
- fool.com
“OpenAI has agreed to about two gigawatts of Trainium capacity through AWS, and Anthropic has signed on for up to five gigawatts of current and future Trainium chips”