Microsoft unveils Maia 300, its third-gen in-house AI chip, for 2027 rollout
Microsoft announced Maia 300 on October 12, its third-generation in-house AI accelerator, positioned as the “primary AI inference system” for Microsoft products and services and aimed at long-context reasoning, latency-sensitive conversation and sustained multi-step agentic tasks. Built on TSMC’s N2P process with a chiplet architecture that separates compute, memory-controller and I/O functions, it uses eight stacks of HBM4 with a claimed 18TB/s peak bandwidth inside a 950W thermal design power. Cloud chief Scott Guthrie says Maia 300 should deliver about 40 percent more tokens per megawatt than leading GPU-based systems, and up to 7x the rack-scale token throughput of Maia 200 in a 96-accelerator configuration; Microsoft plans multi-gigawatt Azure Maia deployments starting in 2027. Microsoft says it will not stop buying Nvidia or AMD GPUs entirely, and Maia 200 already runs production workloads such as Microsoft Copilot, GitHub Copilot and Microsoft Foundry. The performance figures come from Microsoft and have not been independently verified.
It marks Microsoft’s push to make homegrown silicon the backbone of its AI inference, potentially cutting reliance on Nvidia GPUs and reshaping cloud AI inference economics.