Microsoft Unveils Maia 300 AI Chip, Targeting 40% Efficiency Boost Over GPUs by 2027

Microsoft has announced the Maia 300, its third-generation custom AI inference chip, slated to begin running production workloads in 2027. The hardware is built on TSMC’s N2P process node and features a new chiplet architecture that separates compute, memory controller, and I/O functions to maximize the dedicated compute silicon area. Microsoft claims the Maia 300 will deliver a 40% increase in tokens per megawatt compared to leading GPU-based systems, generating up to 7x the rack-scale token throughput of its predecessor in a 96-accelerator configuration. The company plans to scale the Azure Maia family to multi-gigawatt deployments, establishing the in-house silicon as the primary AI inference system for Microsoft products and explicitly reducing its internal reliance on third-party hardware from Nvidia and AMD. "We expect Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems." — Scott Guthrie