Amazon's AWS chip laboratory in Austin is developing its Trainium AI chips. The latest Trainium3 chip reduces costs by up to 50% compared to traditional cloud servers while delivering comparable performance. These chips have now been optimized for AI inference, powering services like Amazon Bedrock and supporting mainstream AI models including Anthropic Claude, which are already running on over 1 million Trainium2 chips. Amazon recently reached an agreement with OpenAI to provide 2 gigawatts of Trainium capacity. The lab team focuses on rapid "launch" and design of chips, aiming to provide a cost-effective alternative to counter Nvidia's dominant GPUs.