According toCNBC and Reuters reported on June 11 that Anthropic and Microsoft are in early stages of negotiations to discuss migrating Claude's reasoning workload to Microsoft's self-developed Maia 200 AI accelerator chip. If the deal is reached, this will be the first time that the Maia 200 has received external cutting-edge model verification, and it will be Anthropic's fourth largest computing power option after NVIDIA GPUs, AWS Trainium, and Google TPUs.
Technical specifications and cost advantages of Maia 200
MicrosoftMaia 200 was released in January 2026. It adopts TSMC's 3nm process, is equipped with 216GB HBM3e memory, and has a computing power of over 10 petaflops at FP4 accuracy. Microsoft claims that its reasoning performance per dollar is more than 30% better than competing products, and currently runs OpenAI's GPT-5.2 internally.
YesFor Anthropic, the core attractions of the Maia 200 are:
Cost reduction: The token cost of large-scale reasoning has dropped significantly
Supply chain diversification: Avoid over-reliance on Nvidia
IPO narrative: Demonstrating "controllable unit economics" to public market investors
For Microsoft,Anthropic verification means:
Chip commercialization: upgrading from "internal tool" to "saleable product"
Azure Differentiation: Providing a unique business story of "Running Claude on Azure"
Ecological lock-in: Binding Anthropic deeper to the Microsoft Cloud
Negotiation risk: FP8 quantification may reduce output quality
The potential risks are:Maia 200's FP8 quantification may reduce the output quality of some tasks by 0.5-1.5%. For companies like Anthropic with "safety + quality" as their core brand, any quality damage needs to be carefully evaluated. Negotiations have been going on for three weeks and no agreement has yet been reached.
Anthropic's negotiations with Microsoft's Maia are the epitome of the wave of "de-Nvidia" in the AI computing power market. When Nvidia monopolized more than 90% of the AI training market with its H100/B200, cloud vendors (Google TPU, Amazon Trainium, Microsoft Maia) and AI laboratories (Anthropic, OpenAI) were jointly building alternative supply chains. Anthropic's "four-option" strategy (NVIDIA, AWS, Google, Microsoft) gives it unprecedented bargaining power in computing power procurement negotiations-something OpenAI (deeply bundled with Microsoft Azure) does not have. But the deeper signal is that if Maia 200 is verified by Anthropic, it will mark the upgrade of Microsoft's self-developed chips from an "internal cost optimization tool" to an "independent commercial asset," which may change the power landscape of the cloud computing market-from "NVIDIA selling chips to cloud vendors" to "cloud vendors selling self-developed chips to AI companies."