Microsoft on October 12 unveils Azure Maia 300 in-house AI chip and plans multi-gigawatt deployment from 2027

Microsoft on October 12 announced its third-generation in-house AI chip Azure Maia 300 on its enterprise blog and at the OCP Global Summit, and said it will scale the Maia line to multi-gigawatt deployments from 2027, making Maia the primary inference system for Microsoft products and services. Maia 300 uses a multi-chiplet design on TSMC's N2P node. Each accelerator carries eight HBM4 stacks, providing 288GB capacity and 18TB/s peak bandwidth inside a 950-watt thermal envelope. Executive vice president Scott Guthrie said Maia 300 delivers 40% more tokens per megawatt than leading GPU systems. In a full 96-chip rack, Maia 300 achieves up to 7x the per-rack token throughput of Maia 200 and more than 2.5x performance per watt. The system targets long-context reasoning, latency-sensitive conversational experiences, and sustained multi-step agent workloads. Maia 200 is already serving Microsoft Copilot in production and supporting the MAI model team, with GitHub Copilot and Microsoft Foundry slated to move onto Maia by year end. Microsoft emphasized that no single chip or model will define frontier AI and will continue partnering with Nvidia and AMD for training workloads.