Anthropic explores Microsoft’s Maia AI chips to train and deploy models

Anthropic is exploring whether Microsoft’s in-house Maia AI chips could become another source of compute for Claude, adding Microsoft to a growing list of cloud and chip partners the startup is using to meet demand. The talks, first reported on May 21 by The Information and matched by Reuters and Bloomberg follow-ups, center on Anthropic renting Azure servers powered by Microsoft-designed accelerators rather than buying chips directly. Microsoft has not publicly announced a deal with Anthropic for Maia capacity, and as of May 25 there was no binding agreement in place, according to the reports. The discussions were described as early-stage and subject to change. ### What exactly is Anthropic discussing with Microsoft? The May 21 reporting said Anthropic was in talks to rent servers powered by Microsoft-designed AI chips as it seeks more computing power for its services. (theinformation.com) That points to an infrastructure arrangement inside Azure, not a broad strategic merger of the two companies’ model roadmaps. People familiar with the matter, as described in secondary reports citing The Information, said the focus was Microsoft’s Maia line, especially Maia 200. The likely use case is running Anthropic models on Maia-backed Azure capacity in Microsoft data centers, potentially in North America where Maia 200 has already begun deployment. (theinformation.com) ### What is Maia 200, and where is Microsoft using it now? Microsoft introduced Maia 200 on January 26 as an inference accelerator built on TSMC’s 3-nanometer process. The company said the chip has 216GB of HBM3e memory, more than 140 billion transistors, and was designed for low-precision AI workloads. Microsoft said Maia 200 delivers “30% better performance per dollar” than the latest-generation hardware already in its fleet. (cnbctv18.com) The company also said the chip is deployed in its US Central datacenter region near Des Moines, Iowa, with US West 3 near Phoenix, Arizona, coming next. Microsoft has framed Maia 200 primarily as an inference chip, meaning it is aimed at serving models and generating tokens efficiently at scale. (blogs.microsoft.com) That matters because reports about the Anthropic talks have referred both to running Claude models and to broader compute needs, so any training use would likely depend on how Microsoft exposes Maia capacity and software tools through Azure. That last point is an inference based on Microsoft’s public Maia description and the reporting on the talks. ### How does this fit with Anthropic’s existing cloud relationships? Anthropic said on November 18, 2025 that it would scale Claude on Microsoft Azure and committed to purchase $30 billion of Azure compute capacity, while also saying Amazon remained its primary cloud provider and training partner. Microsoft separately committed up to $5 billion to Anthropic as part of that broader partnership. (blogs.microsoft.com) That means Maia talks would build on an existing commercial relationship rather than create a new one from scratch. Anthropic already distributes Claude through Microsoft’s ecosystem, while keeping major infrastructure ties with Amazon and Nvidia as well. ### Why is Anthropic looking for more chip options now? Anthropic’s expanding Claude usage has put pressure on compute supply, according to multiple reports on the Maia discussions. (anthropic.com) Reuters’ May 21 item said the company was seeking more capacity to meet rising demand for AI services. Recent reporting also places the Microsoft talks alongside Anthropic’s other chip arrangements. (anthropic.com) Yahoo Finance’s summary of the May 21 reporting said Anthropic agreed in April to use Amazon Web Services’ Trainium chips in a 10-year arrangement valued at more than $100 billion, and had previously announced plans to use Google TPUs. (wixx.com) ### What would Microsoft get if the talks become a deal? Microsoft has been building Maia as part of a broader effort to reduce dependence on Nvidia and prove it can supply competitive AI infrastructure from its own silicon. An Anthropic deployment would give Maia one of its highest-profile outside validation tests to date, according to reports describing the talks. (finance.yahoo.com) The next concrete milestone is whether Microsoft or Anthropic discloses a formal Azure capacity agreement involving Maia. As of Monday, May 25, neither company had announced one publicly, and the most specific public record remained the May 21 reporting that the talks were underway. (theinformation.com) (bloomberg.com)