Claude 5 runs on TPUs and Trainium

Anthropic's multi-chip model strategy serves as a blueprint for surviving the hardware compute wars, proving that frontier models no longer need to be bound to NVIDIA's CUDA ecosystem to deliver state-of-the-art performance. By co-designing their software stack to compile efficiently across TPU, Trainium, and GPU architectures, Anthropic has built a major engineering moat. Deploying Claude 5 on AWS Trainium2 and Google TPUs allows the company to scale inference dynamically across cloud partners, hedging against GPU shortages and gaining significant cost leverage.