The Story

Anthropic has announced a major expansion of its infrastructure partnership with Google and Broadcom, signing a new agreement for multiple gigawatts of next-generation TPU capacity expected to begin coming online in 2027.

According to the official announcement from Anthropic, the deal is intended to support extraordinary growth in customer demand for Claude and represents the company’s largest compute commitment to date.

What the deal includes

• Multi-gigawatt TPU capacity: Anthropic says the agreement covers multiple gigawatts of next-generation TPU infrastructure starting in 2027.
• Google and Broadcom partnership: The deal deepens Anthropic’s existing work with Google Cloud and Broadcom.
• Mostly U.S.-based deployment: The company says the vast majority of this new compute will be sited in the United States.

Why Anthropic says it needs more compute

• Revenue growth: Anthropic says its run-rate revenue has now surpassed $30 billion, up from about $9 billion at the end of 2025.
• Enterprise expansion: The number of business customers spending more than $1 million annually has grown from over 500 in February to more than 1,000.
• Model scaling: The new infrastructure will help power future frontier Claude models while serving rising global demand.

Anthropic’s broader infrastructure strategy

Anthropic says it continues to run Claude across a mix of AI hardware platforms including AWS Trainium, Google TPUs, and NVIDIA GPUs. The company says this hardware diversity helps match workloads to the most suitable chips while improving resilience and performance for customers.

Anthropic also notes that Amazon remains its primary cloud provider and training partner, with continued work underway on Project Rainier.

Why it matters

This deal reinforces a central reality of the AI race: access to compute is becoming as strategic as access to models, customers, or talent. Multi-gigawatt capacity commitments suggest that frontier labs are now planning infrastructure on a scale that looks closer to national industrial buildout than traditional cloud purchasing.

For the broader market, the announcement shows how enterprise demand for frontier AI is translating directly into massive long-term bets on chips, cloud platforms, and domestic data center capacity.