Maia 300: Microsoft Bets on Its Own AI Silicon to Cut Nvidia Dependence

Microsoft has announced Maia 300, the third generation of its in-house AI accelerator. The company intends it to serve as the "primary AI inference system" for its products and services. The chip not only improves the cost and efficiency of running AI workloads in the cloud, but is also more capable of handling the latest reasoning models and agentic tasks.

Maia 300 will begin rolling out to production in 2027, and it was designed with efficiency in mind. "We expect Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems," says Scott Guthrie, executive vice president of Microsoft's cloud group. In a full 96-accelerator configuration, he adds, the chip delivers up to seven times the rack-scale token throughput of Maia 200.

Less reliance on others' GPUs

Those efficiency gains are what should let Microsoft lean mainly on its own accelerators rather than hardware from Nvidia and AMD. "We 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," Guthrie says.

Maia 300 succeeds the 200, which was announced only in January. Its predecessor already handles real-world workloads, including Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft's own AI models. The new chip, Guthrie says, is built for emerging workloads: long-context reasoning, latency-sensitive conversational experiences, and sustained, multi-step agentic tasks.

A chiplet architecture

The efficiency gains come from a new chiplet architecture built on TSMC's N2P process node, which separates compute, memory controller, and I/O functions. According to the Azure hardware and infrastructure team, this increases the silicon area dedicated to compute performance and lets each function use the technology best suited to its role.

The chip also matters for Microsoft's own model ambitions, where the company hopes to catch up with Google DeepMind, OpenAI, and Anthropic. "We have to work across the full stack, all the way down to the silicon," says Microsoft AI CEO Mustafa Suleyman. "At that scale we need a platform that we own, and that we co-design with our models."

Source: The Verge