As efforts to lessen reliance on Nvidia accelerate, Microsoft and Anthropic are reportedly considering a partnership that would run Claude on Microsoft’s in-house silicon. Microsoft is negotiating to provide its proprietary artificial intelligence chips to Anthropic, a potential agreement that, if completed, would represent a pivotal development in the escalating competition among cloud providers to move away from Nvidia’s powerful yet costly hardware.
Under the proposed setup, Anthropic would utilize Azure servers running on Microsoft’s custom-built Maia 200 AI chips, which are tailored for inference tasks, the stage where trained AI models produce answers to real-world questions.
The Maia 200 chip, launched in January 2026, uses TSMC’s 3nm process and links four accelerators with direct connections for faster AI inference. It runs existing AI models more efficiently than Nvidia hardware. Microsoft reports over 30% better tokens-per-dollar, with deployment already in Arizona and Iowa data centers.
Anthropic-Microsoft Deal
A deal would represent a meaningful win for Microsoft, which currently lags behind cloud rivals Amazon and Google when it comes to supplying clients with special-purpose AI silicon. Google has its TPUs. Amazon has Trainium and Inferentia. Microsoft has been the outlier, investing heavily in Maia but yet to convert that investment into major external customers.
Landing Anthropic, maker of the Claude family of models and one of the most prominent AI labs in the world, would change that calculus.
Microsoft has been deepening its relationship with Anthropic in recent months, integrating Claude models into products including its Copilot AI assistant, as its long-standing partnership with OpenAI gradually loosens. The company made a $5 billion investment in Anthropic in late 2025, with Anthropic committing $30 billion toward Azure's computational infrastructure in return.
Anthropic is diversifying its hardware strategy to avoid relying on a single provider. It recently signed a $100B+ deal with AWS for Trainium chips and is now in talks with Microsoft, signaling a multi-vendor approach. Maia 200 could improve cost and latency for Claude models as demand grows ahead of a potential IPO.