Google is adopting Nvidia's own financing strategy to break into the AI chip market, backing a $3.2 billion data center project in western New York that will run on its custom TPU processors. The Lake Mariner campus near Niagara Falls will host computing systems built around Google's Tensor Processing Units, with operator TeraWulf supplying capacity to cloud broker Fluidstack. From there, computing power flows to Anthropic for training its Claude AI models.

The structure mirrors how Nvidia locked in demand for its GPUs. Nvidia has long helped customers secure leases and funding for large GPU clusters, effectively financing the infrastructure that runs its own hardware. Google is now playing the same game with its chips.

Google's TPUs were originally built for internal services like Search and YouTube. The company is now positioning them as a direct alternative to Nvidia's GPUs, which dominate the AI training market. By guaranteeing the financial backing for Lake Mariner, Google ensures the facility will deploy thousands of its TPUs rather than Nvidia's H100 or B200 chips.

The deal also deepens Google's relationship with Anthropic, the AI lab behind Claude. Google has invested billions in the startup, and the Lake Mariner arrangement ties Anthropic's compute infrastructure directly to Google's hardware stack.

Nvidia's grip on the AI chip market remains formidable. The company became the world's most valuable publicly traded company on the back of surging data center spending, and its CEO Jensen Huang was mobbed like a rock star at Taiwan's Computex show this month. At the SK Hynix booth, Huang scrawled "Please make more :)" on a memory wafer—a reference to the supply constraints squeezing the entire AI supply chain.

But the boom is lifting Google's competitors too. Amazon, Microsoft, and Meta are all investing in proprietary AI chips.

Google's move into infrastructure financing signals the company is willing to use its balance sheet the same way Nvidia has to lock customers into its ecosystem. The Lake Mariner project is a test of whether Google's TPUs can scale beyond internal use and cloud rental into the kind of dedicated, finance-backed deployments that have made Nvidia indispensable to the largest AI training clusters.