Microsoft Turns Maia 300 Into a Multi-Gigawatt Inference Rail

Microsoft announced Maia 300 on October 12, 2026, saying Azure Maia will scale to multi-gigawatt deployments starting in 2027 and become the primary inference system for Microsoft products and services. The company says the system is aimed at long-context, latency-sensitive and agentic workloads, with 40% more tokens per megawatt than leading GPU-based systems. Reuters reported Microsoft may deploy more than one million custom AI chips next year as it tries to lower costs and reduce reliance on outside accelerators.

AI Nexus