Anthropic Commits $200B to Google for TPUs and Cloud in Five-Year Mega-Deal

Anthropic has reportedly committed to spending $200 billion with Google over the next five years for cloud capacity and chips, in what would rank as one of the largest infrastructure deals in the AI industry's short history. The arrangement, first reported by The Information on May 5, sent Alphabet shares up roughly 2% in extended trading and briefly pushed Google's parent company past Nvidia in market value.

The commitment formalizes and dramatically expands a relationship that began earlier this year, when Google and Broadcom signed a deal to deliver Anthropic multiple gigawatts of TPU capacity starting in 2027. The new $200 billion figure covers both the custom Tensor Processing Unit chips and the Google Cloud infrastructure to run them, with spending set to ramp up next year. Industry projections already pegged Anthropic's 2026 server bill at roughly $20 billion annually.

The scale is hard to overstate: the commitment is believed to account for more than 40% of the revenue backlog Google disclosed to investors in its most recent earnings report, helping explain why cloud divisions at Google, Amazon, Microsoft, and Oracle have collectively booked something like $2 trillion in AI-related future revenue. Anthropic has separately committed more than $100 billion to AWS over a decade, signaling a deliberately multi-cloud strategy aimed at infrastructure redundancy and avoiding single-supplier lock-in.

The deal also tightens the financial loop between Anthropic and Google in ways that have drawn scrutiny. Google is already a major investor in Anthropic — pledging up to $40 billion in cash and compute last month — meaning a meaningful share of the cash Google puts into the AI lab now flows back to Google as cloud revenue. Critics have flagged this category of "circular" arrangement as a structural risk in how AI infrastructure is being financed, even as Wall Street has so far rewarded the participants.

For Anthropic, the spending plan underwrites the compute needs of an Opus and Mythos roadmap that has pushed deeper into agentic coding, cybersecurity, and enterprise workloads — and signals confidence the company can monetize those capabilities at the kind of scale required to justify a $40 billion-a-year run-rate on infrastructure alone. The lab is reportedly weighing a $50 billion fundraise at a $900 billion valuation, which would be its last private round before a potential IPO.