Amazon announced that it will invest up to an additional $25 billion in Anthropic to improve its artificial intelligence infrastructure layout. Amazon has invested $8 billion in the artificial intelligence startup in the past few years. In an announcement on Monday, Anthropic said it would invest more than $100 billion in Amazon cloud technology products over the next decade, including Amazon's self-developed artificial intelligence chips.Trainiumof existing and new generation products. At the same time, Anthropic has locked up to 5 gigawatts of computing power resources for training and deployment of its Claude large model.
Amazon CEO Andy Jassy
Amazon CEO Andy Jassy said in a statement: "Anthropic's commitment to run its large language model on Amazon Cloud Technology's Trainium chips over the next ten years confirms the results of the cooperation between the two parties in the field of custom chips. We will continue to provide customers with the technology and infrastructure needed for generative artificial intelligence research and development."
According to the official announcement, this investment includes an immediate capital injection of US$5 billion, and the remaining up to US$20 billion will be subsequently realized based on "specific commercialization milestone targets." The initial investment valuation is based on Anthropic’s latest corporate valuation of $380 billion.
Anthropic revealed that it will complete the online deployment of nearly 1 gigawatt of Trainium2 and Trainium3 chip computing resources before the end of this year.
Currently, major cloud giants around the world are competing to expand their artificial intelligence computing power reserves. Amazon disclosed in February this year that it expected full-year capital expenditures of approximately US$200 billion, with funds mainly invested in artificial intelligence infrastructure.
Just two months before this investment, the e-commerce giant announced an investment of up to $50 billion in Anthropic’s top competitor, OpenAI. Two artificial intelligence companies are scrambling to prove their capabilities to investors and prepare for initial public offerings (IPOs) that could launch as early as this year. In recent months, OpenAI executives have repeatedly accused Anthropic of insufficient computing power reserves and a strategic mistake.
Anthropic said on Monday that demand for Claude models from enterprise customers and developers continues to rise, and usage by ordinary users has also surged, putting pressure on its own infrastructure and affecting model service stability and operating performance. This new cooperation with Amazon will quickly expand its available computing power.
Dario Amodai, CEO of Anthropic, said in a statement: "Users have reported that the Claude model has become an indispensable tool in daily work, and we must build supporting infrastructure to keep up with the rapidly growing demand. Cooperation with Amazon allows us to continue to advance cutting-edge research in artificial intelligence while providing Claude services to a wide range of users, including more than 100,000 companies that rely on Amazon cloud technology for development."
Anthropic was established in 2021. The founding team is composed of researchers and executives who have resigned from OpenAI. The company is famous for its Claude series of large artificial intelligence models, and its commercial layout for enterprises has achieved initial results, with annual revenue exceeding US$30 billion.
In 2023, Anthropic selected Amazon Cloud Technology as its preferred cloud service provider; in 2024, it designated Amazon Cloud Technology as its core model training partner. At the same time, the company has also reached cooperation agreements with competing cloud vendors such as Microsoft and Google.
In November last year, Microsoft announced an investment of up to $5 billion in Anthropic, and Anthropic committed to purchasing $30 billion worth of Microsoft Azure cloud computing power. Earlier this month, Anthropic expanded its cooperation with Google and Broadcom, locking in several gigawatts of computing resources.