Anthropic Locks In Up to One Million Google TPU

Anthropic has significantly expanded its utilization of Google Cloud through a deal worth tens of billions of dollars, encompassing up to one million Tensor Processing Units (TPUs) and more than a gigawatt of computing capacity. This deal, announced on October 22, makes this the largest chip commitment from any AI lab to a single cloud provider, as detailed in Anthropic's draft S-1 filing.

The agreement includes delivery of these resources beginning in 2026. Google Cloud CEO Thomas Kurian highlighted the strong performance and efficiency of TPUs, referencing Google's seventh-generation Ironwood chip. Anthropic CFO Krishna Rao emphasized that this expanded capacity is necessary to meet rapid demand while maintaining a competitive edge with their models.

The commercial success of Anthropic is impressive, with over 300,000 business customers and a nearly sevenfold increase in large accounts (each contributing over $100,000 in recurring revenue) over the past year. This enterprise growth necessitates a reserved supply of computing resources, years ahead of their deployment. The deal aligns with Anthropic's IPO preparations, as listed in the company's S-1 draft, which reported tens of billions of dollars in future cloud commitments, including roughly $450 billion in mostly non-cancelable contracts with Google, AWS, Microsoft, and neoclouds, and up to $150 billion more still in negotiations.

Anthropic maintains a multi-platform compute strategy, including Google TPUs, Amazon Trainium, and NVIDIA GPUs, with Amazon remaining its primary training partner through Project Rainier, a cluster spanning multiple US data centers. This decision underscores Anthropic's commitment to diversification rather than reliance on a single silicon vendor.

The significance of this deal lies in the shift from planning to actual commitments, demonstrating that the largest components of their infrastructure stack are moving from conceptual to signed agreements. Gigawatt-scale compute deals are increasingly common in the AI race, as training runs for leading models demand unprecedented levels of power and silicon, and the companies with the fastest revenue growth are securing capacity ahead of competitors.

For Anthropic, this commitment serves as both a response to growing demand and a proactive risk mitigation strategy, illustrating the company's preparedness to invest heavily in infrastructure despite ongoing debates about the sustainability of current AI capital expenditures.