By 2027, the company plans to shift the majority of its AI services to its own custom processors, with a full rack of the new chips expected to provide a sevenfold performance boost for tasks involving AI agents.
Microsoft has unveiled the third generation of its AI accelerator, the Maia 300. Scott Guthrie, head of Azure, described the chip as the “primary AI inference system for Microsoft products and services,” while CEO Satya Nadella called it the company’s “biggest bet on custom silicon to date.” According to Microsoft, the new chip delivers 40% higher performance for the same power consumption as leading systems based on Nvidia GPUs. A full rack of Maia 300 chips offers up to seven times the performance of the current generation when handling tasks such as running AI agents or processing long documents.
While Microsoft handles the design of the chips in-house, manufacturing is outsourced to Taiwan’s TSMC. Mass deployment of the Maia 300 is scheduled for 2027. Unlike Google or Amazon, however, Microsoft does not lease its accelerators to third-party clients. Analysts at Evercore ISI estimate that by early 2027, the Maia chips will account for only about 10% of Microsoft’s total spending on AI hardware.
Microsoft issues its biggest challenge to Nvidia: the Maia 300 chip delivers 40% more performance at the same power level
Image source: Microsoft
The current generation, Maia 200, is already powering Microsoft 365 Copilot and OpenAI models, with deployments currently underway in the United States, Europe, and Australia. GitHub Copilot and Microsoft Foundry are set to transition to this hardware by the end of the year. The first Maia chip was announced in 2023; a second version arrived in January 2026 following some delays, and, according to The Information, it was operational in only two U.S. data centers by the summer.
More efficient chips reduce power consumption—a critical factor, as energy supply remains one of the primary constraints on the expansion of AI infrastructure. Meanwhile, the partnership with Nvidia continues: last week, Satya Nadella and Jensen Huang jointly unveiled Windows PCs powered by Nvidia chips. Furthermore, Microsoft will be among the first cloud providers to deploy systems based on the Vera Rubin and AMD Helios architectures. The company is set to reveal additional details this November at the Ignite conference in San Francisco.
About the Author
