Microsoft has unveiled its third-generation AI chip, Maia 300, with deployment for production workloads scheduled to begin in 2027. The company intends to establish the Azure Maia family as the primary infrastructure for AI inference across its service ecosystem.
Scott Guthrie, executive vice president of Microsoft's cloud division, stated that the new system is designed to improve the efficiency of cloud AI workloads. It is specifically optimized for reasoning models and agent-based tasks.
According to company data, the Maia 300 chip provides 40% more tokens per megawatt compared to leading GPU-based systems from Nvidia and AMD. In a full configuration featuring 96 accelerators, the system delivers seven times the performance of the previous model.
The company is already utilizing the Maia 200 chip for AI workloads, including Microsoft Copilot, GitHub Copilot, and Microsoft Foundry. Mustafa Suleyman, head of Microsoft AI, emphasized that controlling the full technology stack down to the silicon level is critical for the firm's AI strategy.
