Anthropic Will Use 1 Million Google Chips To Train Claude's AI Model

Anthropic struck a deal with Google (photo: dock. Anthropic)

JAKARTA Anthropic has entered into a new agreement with Google. The Artificial Intelligence Company (AI) will use a million Google-made AI chips to train the AI model Claude.

This deal is worth up to tens of billions of dollars. Anthropic deliberately uses the technology Google is developing to advance its AI system. That way, Claude can continue to compete competitively.

In Anthropic's official announcement, the company will have access to more than one gigawatt of computing capacity. This sizable capacity is scheduled to start operating in 2026.

This computational capacity will be used to train the next generation of Claude models. This training will be conducted on Google's internal Tensor Processing Unit (TPU) which is usually reserved for internal use.

Anthropic stated that they chose TPU for reasons of efficiency and price ratio and superior performance. In addition, they already have experience in training and serving the rock model using the processor.

Google, as a second party to this agreement, will provide TPU through Google Cloud as an alternative to NVIDIA processors with limited supply. Google will also provide additional computing services to support Anthropic.

This agreement came after OpenAI signed a similar deal at a large cost. The ChatGPT developer struck a deal with a cloud company worth more than $1 trillion.

At that expense, OpenAI will secure around 26 gigawatts of computing capacity. It should be noted that a computing gigawatt can cost around 50 billion US dollars (IDR 8,305 million).