Expanding our use of Google Cloud TPUs and Services (www.anthropic.com)

Anthropic announced a major expansion of its Google Cloud usage that could include up to one million TPUs — a multi‑billion‑dollar commitment expected to bring well over a gigawatt of capacity online by 2026. The deal underscores Anthropic’s rapid commercial growth (now serving 300,000 business customers, with large accounts up ~7× year-over-year) and is intended to scale training, testing, alignment research and responsible deployment for its Claude models. Google framed the move around TPU price‑performance and efficiency gains, citing its seventh‑generation “Ironwood” TPU family as part of the accelerator roadmap. Technically, Anthropic will keep a diversified, multi‑vendor compute strategy: expanding TPU use while continuing to leverage AWS Trainium and NVIDIA GPUs, and maintaining its collaboration with Amazon on Project Rainier — a separate massive cluster spanning hundreds of thousands of AI chips across U.S. data centers. That multi‑platform approach preserves flexibility, hedges supply and cost risk, and lets Anthropic match workloads to the most efficient hardware for training, fine‑tuning and inference at scale. The scale of the investment signals faster iteration on larger models and more compute‑intensive alignment workflows, while also raising the industry’s stakes around power, infrastructure and cross‑cloud partnerships for frontier AI development.