Microsoft announces Maia 300, its next in-house AI chip, for 2027

On October 12, 2026, Microsoft announced Azure Maia 300, the next chip in its own family of AI processors. Yahoo Finance reported that the chip was unveiled at the OCP Global Summit 2026 in San Jose.

In a post on the Official Microsoft Blog, Scott Guthrie, Microsoft’s executive vice president for Cloud + AI, wrote that Maia 300 will begin production deployments in 2027. Microsoft plans to grow the Maia family to “multi-gigawatt deployments” starting that year and to make Maia the primary inference system for its products and services.

Microsoft’s stated figures for the chip:

• Compute pieces made on TSMC’s N2P manufacturing process.
• Eight stacks of HBM4 memory with 18.0 TB/s of peak bandwidth, within a 950-watt power rating.
• A full rack of 96 chips in a 167-kilowatt power envelope.
• 40% more tokens per megawatt than leading GPU-based systems.
• Up to 7 times the rack-level token throughput of Maia 200.

Maia 200, the current generation, is already in production and serving Microsoft workloads, including Copilot, according to the post. The Maia 300 numbers are Microsoft’s own projections for a chip that is not yet in production.

Why it matters

Inference is the work of running a trained model to answer a request, and it happens every time someone asks an AI a question. “Tokens per megawatt” measures how much text a system can produce for a given amount of electricity, so it links chip design directly to running costs. Microsoft builds its own chips while also buying from suppliers, and Yahoo Finance described Maia 300 as part of an effort to reduce its dependence on Nvidia and AMD. The What AI Costs course explains how chips, power and datacenters add up to the price of an answer.