Microsoft plans Maia 300 rollout and multi-gigawatt expansion from 2027
Microsoft says Maia 200 is already serving production workloads, while Maia 300 is due in 2027 as it moves toward using its own inference system across its services.
Key points
- 1Microsoft schedules Maia 300 production deployments to begin in 2027.
- 2The wider Azure Maia expansion is planned to reach multi-gigawatt scale, starting in 2027.
- 3Microsoft says Maia 200 already serves Copilot; GitHub Copilot and Microsoft Foundry are due to follow by year-end.
- 4The announced Maia 300 rack has 96 accelerators and a 167-kilowatt power envelope.
- 5The claim of up to seven times Maia 200's rack-level token throughput comes from Microsoft.
Full story
Microsoft plans to start deploying Maia 300 into production in 2027, alongside an expansion of the Azure Maia family to multi-gigawatt scale beginning that year. The company said Maia is intended to become its main inference system for Microsoft products and services. Those plans describe a future rollout, not capacity already delivered.
Maia 200 is already running production workloads for Microsoft Copilot and supporting the company's MAI team as it develops new models, according to Microsoft. The company says GitHub Copilot and Microsoft Foundry will follow by the end of 2026. It also reports that Maia 200 is being deployed in the United States, Europe and Australia.
Microsoft says Maia 300 targets reasoning over long contexts, conversational applications sensitive to delays, and tasks involving multiple steps by AI agents. Its chiplet design separates compute, memory-controller and input/output functions, with compute chiplets built using TSMC's N2P process. The company presents that separation as a way to improve efficiency and scalability.
The announced rack design combines 96 accelerators within a 167-kilowatt power envelope and provides 27.6 TB of HBM capacity, Microsoft says. In that full configuration, the company claims up to seven times Maia 200's rack-level token throughput. The figures are Microsoft's specifications and performance claims for the planned system; its production deployment is still scheduled for 2027.
Why it matters
Analysis: The plan ties Microsoft's inference capacity more closely to its own hardware roadmap. Existing Maia 200 use shows adoption within the company, but it does not establish the gains Maia 300 will deliver. Those depend on the planned system's performance in production.
Timeline
· Published
Topics#Microsoft#Azure#Maia 300#AI infrastructure#Inference
Sources
This story draws on the following sources. Read them for full context.
