Anthropic secures Massive Google Cloud Deal to Power AI Inference with New TPU v7 Chips
SAN FRANCISCO, CA - Anthropic, the AI safety and research company founded by former OpenAI researchers, has entered a multi-year agreement with Google Cloud worth tens of billions of dollars to significantly expand its computing capacity. The deal will provide Anthropic with access to up to 1 million Tensor Processing Units (tpus), Google’s AI-optimized hardware, and will deliver over 1 gigawatt of computing power by 2026.
The expanded partnership builds on existing collaboration and comes as Anthropic focuses on its Claude family of AI models for enterprise customers. The company reports a rapidly growing client base, exceeding 300,000 businesses, with large accounts – those spending over $100,000 annually – increasing sevenfold in the past year.
A key driver of the decision to increase TPU usage is the efficiency and cost-effectiveness of Google’s latest generation of Ironwood accelerators (TPU v7). Unveiled at Google Cloud Next 2025, Ironwood is Google’s first TPU specifically designed for inference – the process of deploying and running AI models, rather than training them. Each Ironwood chip boasts 192 GB of memory and 7.2 TB/s bandwidth, making it well-suited for handling the high volume of requests required for large-scale AI deployment.
Anthropic has historically implemented stricter usage limits on the number of inferences, even for paying customers, highlighting the critical need for increased inference capacity. While expanding its reliance on Google’s TPUs, Anthropic will continue to leverage Amazon Web Services infrastructure, including the Trainium 2 chip, and Nvidia GPUs for training and research purposes.
