Microsoft to Deploy Self-Designed Maia 300 Chips at Gigawatt Scale in 2027

Microsoft announced on October 12 that its next-generation in-house chip, Maia 300, will be deployed in data centers at gigawatt scale starting in 2027, according to Bloomberg. The chip is designed for AI inference workloads and is expected to process about 40% more tokens per megawatt than mainstream GPU systems, allowing more AI workloads to run at the same power consumption.

Rani Borkar, who leads Microsoft's data center hardware business, said the company built what it believes is the best product for these workloads and would back that confidence with gigawatt-scale Maia deployments from next year. Unlike Google and Amazon, which offer their custom chips to cloud customers, Microsoft does not currently plan to rent out Maia compute capacity to external clients. Instead, it will prioritize supporting its own large-scale AI services, reducing reliance on Nvidia GPUs and lowering inference costs.

Microsoft has been expanding its in-house chip efforts for some time. In early 2026, the previous-generation Maia 200 began installation in Microsoft data centers, helping power AI software such as Copilot and other workloads. It supports OpenAI models, Microsoft's own models, and several open-source model families across data centers in the United States, Europe, and Australia. Maia 300 deployment is expected to begin in 2027.