Microsoft unveils Maia 300, its new AI chip to reduce reliance on Nvidia and AMD

Microsoft has unveiled the Azure Maia 300, its latest AI chip, as it seeks to reduce its reliance on third-party processors such as those from Nvidia and AMD. According to the article, the chip was announced on Monday at the OCP Global Summit 2026 in San Jose.

The Maia 300 will join the Maia 200 and Maia 100 in what the company describes as multi-gigawatt deployments starting next year. Microsoft says this will make its Maia chips the "primary AI inference system for Microsoft products and services."

On performance, Microsoft claims the Maia 300 delivers 40% more tokens per megawatt than "leading GPU-based systems," a comparison the article reads as a clear reference to Nvidia. In addition, a 96-chip rack delivers seven times the rack-scale token throughput of the Maia 200. Tokens are a basic unit of measurement in AI systems and can represent fragments of words, images and video; the more tokens per megawatt, the more efficient an AI rack is.

The Maia 200, for its part, already powers Microsoft production workloads, including Microsoft 365 Copilot, as well as OpenAI models, Microsoft's MAI models and open-source models. Microsoft says it delivers 1.7 times more tokens per watt for AI features in Teams, Copilot and Cowork compared with other GPU-based systems.

The industry backdrop is a broader trend: Amazon, Google and Meta, as well as the labs OpenAI and Anthropic, already offer their own chips or are working on them. The article notes that this serves as a hedge against overreliance on Nvidia and AMD and allows silicon to be tailored to specific needs.

For Nvidia, the article sees a potential problem: the company typically generates more than 50% of its quarterly revenue from hyperscalers. Moreover, Amazon and Google already rent out their custom chips to outside customers, posing a direct threat to its business, and Amazon is considering selling chips for customers to install in their own data centers.

Even so, the article cautions that this does not necessarily mean hyperscalers will stop buying Nvidia chips. Customers want access to different chips depending on the workload, so Amazon, Microsoft and others will keep stocking the most in-demand processors, which for now are Nvidia's.

Note: the performance figures are Microsoft's own claims, as reported in the article, and have not been independently verified.