Anthropic to Use One Million Google AI Chips to Train Next Claude Models

AI startup Anthropic is deepening its partnership with Google, securing access to as many as one mln of the tech giant’s AI chips. The deal is valued in the tens of billions of dollars.

The agreement, announced October 23, will provide Anthropic with over one GW of computing power from 2026 to train future versions of its Claude AI model using Google’s tensor processing units (TPUs), previously reserved mainly for internal use.

Anthropic said it chose TPUs for their cost efficiency and performance, as well as its positive track record using them to build and deploy Claude models.

Google, through its Cloud division, will also supply Anthropic with additional computing infrastructure, positioning its TPUs as a key alternative to Nvidia’s heavily sought-after GPUs.

The deal comes as rivals like OpenAI make billion-dollar investments to expand their computing capacity, reportedly securing up to 26 GW of power to meet soaring demand.

Anthropic, known for its focus on AI safety and enterprise applications, has seen rapid revenue growth and is helping drive innovation in startups using its technology, such as coding assistant developer Cursor.

Kursiv also reports that OpenAI has launched ChatGPT Atlas, directly challenging Google Chrome’s dominance in the market.