Published: October 14, 2026
What Microsoft announced
On October 12, Microsoft introduced Azure Maia 300 and said Maia 200 is already serving production workloads, including Microsoft Copilot. The next generation is planned for production deployments in 2027, with multi-gigawatt scaling of the Maia family also starting from 2027.
What the announcement describes
Microsoft presents Maia 300 as an inference-first system for long-context, memory-intensive and multi-step agentic workloads. Its official post claims up to 40% more tokens per megawatt than leading GPU systems and up to seven times the rack-scale token throughput of Maia 200 in a 96-accelerator configuration.
From chip to system
This is not only a processor announcement. Microsoft describes eight HBM4 stacks with 18.0 TB/s of peak bandwidth, a multi-chiplet architecture, scale-up networking and a rack with 96 accelerators. These figures describe the announced system; they do not mean that this configuration is already deployed everywhere.
Why it matters
Custom accelerators let a cloud provider co-design silicon, memory, networking and software around real workloads. The practical test will be production delivery, software stability and operating cost. Maia 300 remains a planned next step rather than a generally available platform.