Microsoft and Anthropic are negotiating a deal that would see the Claude maker lease Azure servers equipped with Microsoft’s custom-designed Maia 200 AI chips. The talks are still in their early stages, with no formal agreement in place.

The discussions center on Anthropic using Microsoft’s Maia 200 chips specifically for inference workloads. Inference is the part of AI where a trained model actually generates responses to user queries, as opposed to the training phase where models learn from massive datasets.

Microsoft unveiled the Maia 200 in January 2026. The chip was fabricated on TSMC’s 3-nanometer process. The goal is straightforward: deliver more efficient inference performance than what’s currently available, which mostly means Nvidia’s H100 and B200 GPUs.

Microsoft poured $5 billion into Anthropic back in November 2025. Anthropic, for its part, has committed $30 billion toward Azure compute infrastructure. The negotiations were first reported on May 21, 2026, by The Information and have since been substantiated by CNBC and Reuters.

Anthropic’s interest in the Maia 200 reflects a broader effort to diversify its hardware suppliers. Google has its TPUs. Amazon has Trainium and Inferentia. Microsoft now has Maia.

The $30 billion Anthropic has earmarked for Azure compute underscores how capital-intensive frontier AI development has become, and how the economics of inference at scale are now a first-order strategic concern for every major AI lab.

There’s also a secondary implication for decentralized GPU networks. If centralized providers like Microsoft can meaningfully reduce inference costs through custom silicon, it pressures the value proposition of decentralized alternatives. Investors in projects like Render, Akash, and io.net should be paying attention to how aggressively the hyperscalers are vertically integrating their chip supply chains.