The AI safety company has signed over a dozen preliminary data center leases exceeding 1 gigawatt of capacity as it moves away from cloud rental agreements
Anthropic has signed more than a dozen preliminary agreements to lease US data centers with a combined capacity exceeding 1 gigawatt, marking a decisive pivot from the cloud-first strategy that has defined most AI startups.
The company has been in discussions with Google about a financial guarantee to backstop those lease obligations, a move that would essentially give landlords the comfort of knowing a $2 trillion company is standing behind the rent checks.
From tenant to operator #
For most of its existence, Anthropic has relied on cloud providers like Amazon Web Services, Google Cloud, and Microsoft to supply the computing muscle behind its AI models. The shift to managing its own servers is fundamentally about economics: instead of paying a markup to rent GPUs and TPUs from cloud giants, Anthropic wants to cut out the middleman and control the hardware directly. This latest infrastructure push follows Anthropic’s November 2025 announcement of a $50 billion investment in US data centers. The preliminary lease agreements now suggest those plans are materializing into concrete deals.
Google’s quiet but critical role #
The relationship between Anthropic and Google is one of the more interesting dynamics in the AI industry. Google is simultaneously an investor in Anthropic, a cloud provider to Anthropic, and now potentially a financial guarantor for Anthropic’s real estate obligations.
Google has previously provided financial support for other Anthropic-related projects, including backstops at various sites. Anthropic has also been expanding its utilization of Google’s custom TPU chips, with plans to scale up to as many as 1 million units.
The infrastructure land grab #
Anthropic isn’t making this move in a vacuum. OpenAI, xAI, and other major AI developers are all facing the same fundamental challenge: training frontier AI models requires staggering amounts of compute, and compute requires data centers, and data centers require electricity.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our