Microsoft unveils Maia 300 AI chip to reduce Nvidia reliance

Microsoft said Maia 300 is designed for inference-first workloads and is expected to deliver 40 percent more tokens per megawatt than leading GPU-based systems. In a full 96-accelerator rack configuration, the company said it can deliver up to seven times the rack-scale token throughput of Maia 200. Scott Guthrie, executive vice president of Microsoft’s cloud group, said the company will “scale our Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary AI inference system for Microsoft products and services.”

The chip is the successor to Maia 200, which Microsoft announced in January and which is already running production AI workloads across the company, including Microsoft Copilot, GitHub Copilot, Microsoft Foundry and Microsoft’s own AI models. Microsoft introduced its first AI chip, Maia 100, in 2023. CEO Satya Nadella told analysts in July that Maia 200 “delivers 30% better performance per dollar than the latest generation hardware in our fleet” and said it was handling OpenAI models. Finance chief Amy Hood said at a Goldman Sachs conference last month that use of Microsoft’s own chips has helped with “margin improvement over the course of a decade,” according to CNBC.

Maia 300 uses a new chiplet architecture built on TSMC’s N2P process node that separates compute, memory-controller and input/output functions. Microsoft said this increases the silicon area dedicated to compute performance and allows each function to use the technology best suited to its role. The company says the accelerator uses eight stacks of HBM4 memory, with a claimed peak bandwidth of 18TB/s within a 950W thermal design power. A Maia 300 rack will contain 96 accelerators within a 167kW power envelope, with 27.6TB of HBM capacity and 1.7 petabytes per second of aggregate HBM bandwidth. Microsoft says support for low-precision formats, including an enhanced MXFP4, will reduce memory and bandwidth demands. Networking is designed as part of the system’s dataflow rather than as a separate layer, using high-bandwidth connections between accelerators and a custom AI transport based on Ethernet, Microsoft said.

Microsoft says the chip is designed for long-context reasoning, latency-sensitive conversational experiences and sustained, multi-step agentic tasks. It is expected to support agentic coding in GitHub, large-scale model serving through Microsoft Foundry, and improved inference and data generation for Microsoft AI products. Jim Kleewein, a technical fellow at Microsoft, said parts of Microsoft 365 Copilot should speed up with Maia 300. Microsoft also expects improvements in Copilot, Cowork and Teams, according to TechRadar. Microsoft AI CEO Mustafa Suleyman said in The Verge that Microsoft has to work across the full stack down to the silicon. “At that scale we need a platform that we own, and that we co-design with our models,” Suleyman said. “That’s why running our Microsoft AI models on top of the Maia chip is a critical focus for our team.”

Microsoft still expects Nvidia to remain a major part of its AI infrastructure. Alistair Speirs, a general manager, told CNBC that Microsoft partners with Nvidia and AMD when it comes to training models. Nvidia CEO Jensen Huang spoke at an event last week where Microsoft announced a Surface laptop containing an Nvidia processor. Microsoft’s accelerators differ from Google’s tensor processing units and Amazon’s Inferentia and Trainium because they cannot be rented by the hour to other companies. “Nobody has tested them outside of Microsoft yet,” said Patrick Moorhead, CEO of advisory firm Moor Insights and Strategy, according to CNBC.

In a note distributed to clients, Evercore ISI analysts estimated that Maia will represent about 10 percent of Microsoft’s spending on AI chips by the first quarter of 2027. The analysts recommend buying Microsoft shares. Microsoft said production of Maia 300 starts in 2027, though Speirs declined to provide a timeline for multi-gigawatt deployments. “We reserve a little bit of optionality there, because this is an industry that’s changing very fast, and we want to deploy the right silicon for the right models that we’re using at the right time as well,” he told CNBC.

Editor's Summary

Microsoft’s Maia 300 marks a larger push to make in-house silicon the primary inference engine for its AI products and services, with production set to begin in 2027. The company says the chip will improve efficiency and reduce reliance on Nvidia GPUs, while Microsoft will continue to partner with Nvidia and AMD for training and other workloads. Analysts expect Maia to account for a small but growing share of Microsoft’s AI chip spending by early 2027.