THE CRUNCH
Microsoft has announced Maia 300, its third-generation in-house AI chip, which will start rolling out to production AI workloads in 2027. The company is positioning it as the "primary AI inference system" for Microsoft's products and services, a shift that would let Azure lean mainly on its own silicon rather than GPUs from Nvidia and AMD.
Scott Guthrie, executive vice president of Microsoft's cloud group, said Microsoft expects Maia 300 to deliver 40 percent more tokens per megawatt compared with leading GPU-based systems, and up to 7x the rack-scale token throughput of Maia 200 in a full 96-accelerator configuration. Those are vendor claims, so real-world figures will depend on production results.
The chip is built on TSMC's N2P process node and uses a chiplet architecture, which separates compute, memory controller and I/O functions onto distinct pieces of silicon. Microsoft's Azure hardware team says this increases the silicon area dedicated to compute and lets each function use the technology best suited to its role.
The cadence is striking: Maia 200, the chip Maia 300 succeeds, was only announced in January and is already running production workloads including Microsoft Copilot, GitHub Copilot and Microsoft Foundry. Guthrie called it the best performing AI infrastructure in Microsoft's fleet, measured in real-world metrics.
Maia 300 is also tied to Microsoft's model ambitions. Microsoft AI CEO Mustafa Suleyman said running the company's own AI models on Maia chips is a critical focus, part of an effort to work "across the full stack, all the way down to the silicon" and, Microsoft hopes, catch up with Google DeepMind, OpenAI and Anthropic.