Anthropic has deepened its collaboration with Google in a massive new deal that will give the AI startup access to up to one million of Google’s advanced tensor processing units (TPUs) — artificial intelligence chips valued at tens of billions of dollars. The agreement, announced Thursday, provides Anthropic with over one gigawatt of computing capacity, coming online in 2026, to train future versions of its Claude AI models.
Anthropic said it chose Google’s TPUs for their strong price-performance ratio, efficiency, and seamless integration with its existing Claude infrastructure. The expansion underscores the industry’s surging demand for compute power as AI firms race to develop increasingly capable and human-like systems.
Read More: AI Assistants Distort News in Nearly Half Their Responses
Google, through its Alphabet-owned cloud division, will also provide Anthropic with additional cloud services. The move positions Google as a key partner and alternative supplier to Nvidia, whose chips dominate the AI hardware market but remain in limited supply.
Read More: ‘Science, Technology, and Engineering for Development’ initiative inaugurated
The announcement follows reports that rival OpenAI has signed multitrillion-dollar agreements to secure up to 26 gigawatts of computing capacity using Nvidia and AMD hardware. In contrast, Anthropic continues to focus on AI safety and enterprise-grade applications, with its Claude models already driving growth across productivity and coding startups like Cursor.