Anthropic has committed to deploying multiple gigawatts of next-generation Google TPU capacity starting 2027, securing dedicated compute for Claude’s growing agent workloads across a second major cloud ecosystem. The deal with Google and Broadcom represents Anthropic’s largest single compute commitment to date, according to EdgeIR.
The Commitment
The TPU deal is part of Anthropic’s $50 billion U.S. computing infrastructure investment announced in November 2025. Krishna Rao, Anthropic’s CFO, called it a direct response to customer growth: “We are making our most significant compute commitment to date to keep pace with our unprecedented growth,” he told EdgeIR.
The numbers behind that growth: Anthropic’s 2026 revenues surpassed $30 billion, with over 1,000 business customers spending more than $1 million annually, per the same report.
Two Cloud Deals in Three Days
The Google TPU commitment arrives just days after Amazon agreed to invest up to $25 billion in Anthropic, with Anthropic committing to spend over $100 billion on AWS technologies over the next decade and securing 5 gigawatts of AWS Trainium capacity, according to Reuters.
Claude now runs on three distinct hardware paths: AWS Trainium chips, Google TPUs (designed with Broadcom), and NVIDIA GPUs. The multi-cloud, multi-silicon strategy insulates Anthropic from single-vendor dependencies while ensuring Claude-based agents have dedicated capacity on whichever cloud platform an enterprise already uses.
The Agent Infrastructure Angle
The shift from query-based API usage to continuous agent workloads fundamentally changes how much compute a foundation model company needs. Managed Agents, Claude’s enterprise agent product, runs Claude instances persistently rather than on-demand. That means sustained compute draw measured in always-on gigawatts, not peak throughput measured in tokens per second. When Anthropic pre-secures multi-gigawatt capacity across both Google Cloud and AWS, it is betting that enterprise customers will run Claude agents continuously at a scale that requires utility-grade infrastructure commitments years in advance. That bet now totals over $50 billion in committed infrastructure spend across two hyperscalers.