Anthropic taps Google TPUs in multibillion deal to power Claude

Anthropic has expanded its partnership with Google in a deal that will grant the AI company access to as many as one million of Google’s tensor processing units (TPUs) to train future versions of its Claude models. The arrangement, valued in the tens of billions of dollars, includes more than one gigawatt of computing capacity expected to come online in 2026.

The company said it chose Google’s TPUs for their price-performance and energy efficiency, as well as Anthropic’s existing experience deploying and serving Claude on the chips. The agreement also extends Anthropic’s use of Google Cloud services, deepening a strategic relationship that gives the startup a pipeline of compute at a time when demand for AI hardware remains intense.

The AI industry continues to race for processing power amid supply constraints on Nvidia’s graphics processors. Other tech firms, including OpenAI, are striking massive compute deals to secure long-term capacity, as companies scale models and expand enterprise products.

Anthropic, which emphasizes safety and reliability in its systems, is targeting enterprise use cases and developer tools, and expects rapid revenue growth as adoption accelerates. The company says the additional compute will help it advance next-generation Claude models while keeping a focus on efficient training and responsible deployment.