Microsoft unveiled Maia 300, the third generation of its custom AI accelerator, on Monday and said it will scale the Maia family to multi-gigawatt deployments starting in 2027, making it “the primary inference system for Microsoft products and services,” cloud chief Scott Guthrie wrote in a company blog post. Microsoft plans to deploy potentially more than a million of the chips next year to back its internal AI services, Reuters reported.
The shift is meant to reduce Microsoft’s reliance on Nvidia, whose GPUs fill its existing data centers, CNBC reported. The current Maia 200 already serves Microsoft Copilot and runs OpenAI models and Microsoft’s own MAI models, with GitHub Copilot and Microsoft Foundry to follow by the end of the year, according to the blog. Nvidia remains a partner for training models alongside AMD, general manager Alistair Speirs told CNBC, and he declined to give a timeline for the gigawatt build-out.
Microsoft says Maia 300 will deliver 40% more tokens per megawatt than leading GPU-based systems, though Reuters noted the company did not name the systems it compared against. A full rack holds 96 accelerators, double Maia 200, within a 167-kilowatt power envelope and delivers up to seven times Maia 200’s rack-scale token throughput. Reuters compared that with the 142 kilowatts used by Nvidia’s GB300 racks with 72 GPUs.
Each chip pairs eight HBM4 memory stacks with 18 terabytes per second of peak bandwidth in a 950-watt design, according to Microsoft, and its compute chiplets are built on TSMC’s N2P process. “It’s basically delivering more intelligence for every watt, every dollar, and every byte of memory,” Rani Borkar, president of Azure hardware systems and infrastructure, told Reuters. Production deployments begin in 2027.
The bet comes as Microsoft expects to spend more than $175 billion in capital expenditures this calendar year, Reuters reported. Evercore ISI analysts estimate Maia will account for about 10% of Microsoft’s AI chip spending by the first quarter of 2027, according to CNBC. Unlike Google’s TPUs or Amazon’s Trainium, Maia chips can’t be rented by the hour to other companies, and Nvidia is still a close partner: Microsoft announced a Surface laptop built on Nvidia silicon last week.
Sources
- Microsoft (Scott Guthrie): Microsoft to expand Azure Maia system to multi-gigawatt deployments (October 12, 2026)
- CNBC: Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia (October 12, 2026)
- Reuters (Stephen Nellis, Max A. Cherney) via Financial Newsletter: Microsoft plans huge deployment of its own AI chips, with an eye on saving costs (October 12, 2026)