Anthropic is currently in early discussions about renting server capacity powered by Microsoft’s Maia chips, as reported by both The Information and CNBC. This comes on the heels of Anthropic’s recent agreement to utilize Google’s cloud infrastructure and Tensor Processing Units in a large multi-year deal. Additionally, Anthropic has a separate agreement with Microsoft, where Microsoft is investing $5 billion in Anthropic, and Anthropic is committing to spending $30 billion on Azure.
As part of the Microsoft agreement, Anthropic has also committed to procuring additional compute capacity of up to 1GW. Anthropic already uses cloud services from Amazon Web Services and Google Cloud. The talks with Microsoft regarding Maia chips are still in the early stages, and no deal has been finalized yet.
Microsoft’s Maia chip is an in-house AI accelerator, with the second-generation Maia chip already announced, although not yet available through Azure. The Maia 200 chip is specifically designed for AI inference and boasts 30% better performance per dollar compared to the latest-generation hardware in Microsoft’s fleet.
Anthropic is best known for its AI models under the name Claude, with the company also gaining wider use of its AI-assisted programming tool, Claude Code. Anthropic CEO Dario Amodei recently mentioned that the company has faced challenges with compute power.
In addition to the potential Microsoft agreement, Anthropic has also made significant deals with Google and AWS for compute capacity. These agreements involve substantial investments and access to custom chips such as Google’s TPUs and AWS’s Trainium chips.
Overall, Anthropic has established relationships with major cloud providers like Amazon, Google, and Microsoft, each offering custom AI chips for cloud infrastructure. A potential deal with Microsoft for Maia chips would further diversify Anthropic’s cloud relationships and computing capabilities.
(Original article content remains unchanged. Photo by BoliviaInteligente)
