Alphabet's Google will supply up to 1 million specialized artificial intelligence (AI) chips to Anthropic PBC in a deal valued at tens of billions of dollars, further deepening the partnership between Google and the fast-growing artificial intelligence startup.
The deal, announced on Thursday, will significantly increase Anthropic’s computing power while solidifying Google’s position as a major investor and provider of critical AI infrastructure.
Google said in a statement that these TPU chips designed independently by Google to accelerate machine learning tasks are expected to be deployed in 2026, and will have more than 1 gigawatt of computing power online. The size of the deal, which ranks among the largest in the current AI hardware race, highlights the sharp rise in the cost of training and inference computing power required to develop cutting-edge AI systems.
Anthropic Chief Financial Officer Krishna Rao said in a statement: "Anthropic has a long-term cooperation with Google. This further expanded cooperation will help us continue to improve computing power and promote the development of AI frontiers."
By using Google's TPUs, Anthropic will have access to one of the most advanced chip infrastructures outside of Nvidia products, reducing its reliance on scarce and expensive GPUs. The company's choice to deepen its cooperation with Google, the world's third-largest cloud service provider, also demonstrates its confidence in Google's dedicated AI technology.
Google has invested approximately $3 billion in Anthropic to date, including $2 billion in 2023 and an additional $1 billion earlier this year. At the same time, Amazon has pledged to invest up to US$8 billion and provide self-developed chips to the AI start-up through Amazon Cloud Service (AWS). Anthropic relies on the two companies' cloud services to train and deploy its Claude family of models, which compete with OpenAI's GPT product and Google's Gemini.