An exclusive tour of Amazon’s Trainium lab, the chip that’s won over Anthropic, OpenAI, even Apple

Amazon's Annapurna Labs chip team in Austin gave TechCrunch a rare private tour of the facility where AWS Trainium chips are developed and tested. With 1.4 million Trainium chips deployed across three generations, Anthropic's Claude running on over 1 million Trainium2 chips, and a new $50 billion deal to supply OpenAI with 2 gigawatts of Trainium compute, the chip is becoming central to AWS's AI strategy. Trainium3, a 3nm chip manufactured by TSMC, now supports PyTorch with near-zero migration effort (a one-line code change), directly challenging Nvidia's ecosystem lock-in. The chips cost up to 50% less than comparable cloud servers and now handle the majority of inference traffic on Amazon Bedrock. The piece also covers the intense 'silicon bring-up' process, custom server sled design, liquid cooling engineering, and a new partnership with Cerebras for low-latency inference.