Microsoft plans multi-gigawatt Azure Maia deployments from 2027
Microsoft said Maia 200 is in production and it plans to scale its custom AI inference systems to multi-gigawatt deployments starting in 2027.
Event / data period: 2026-10-12
Published: · Updated:
Microsoft said on October 12 that its Maia 200 accelerator is serving production workloads, including Microsoft Copilot. The company plans to expand the Maia family to multi-gigawatt deployments from 2027, with Maia 300 the next generation designed for inference workloads.
Microsoft estimates that Maia 300 will deliver 40% more tokens per megawatt than leading GPU-based systems. It also says a 96-accelerator rack can provide up to seven times the rack-scale token throughput of Maia 200 within a 167-kilowatt power envelope. These are company performance comparisons, and Maia 300 deployments are planned to begin in 2027.
Data
Maia deployment roadmap
Production and planned deployment stages Microsoft reported for Maia accelerators.
- · Occurred
Maia 200 serving production workloads
Microsoft said Maia 200 serves its production workloads, including Microsoft Copilot, and it plans multi-gigawatt Maia deployments from 2027.
Methodology
Dates and performance statements are reported from Microsoft’s October 12, 2026 announcement. The 2027 deployment is a future plan; performance comparisons are company estimates.
Updated:
Maia performance and rack specifications
Microsoft-reported performance estimates and rack specifications.
- Estimated tokens-per-megawatt improvement for Maia 300
- 40 %
- 2026-10-12 · Estimate
- Upper bound: rack throughput as a share of Maia 200
- ≤700 %
- 2026-10-12 · Estimate
- Accelerators in Maia rack
- 96 count
- 2026-10-12 · Target
- Rack power envelope
- 0.167 MW
- 2026-10-12 · Target
Methodology
40% is the estimated tokens-per-megawatt improvement over competing systems. 700% expresses the up-to-7-times upper bound as a share of Maia 200’s baseline; 167 kW is converted to MW by dividing by 1,000. Performance statements are company estimates; rack values are company specifications.
Updated:
Subjects and places
Subject proposals
- Artificial intelligenceSuggested
- Subject suggested from the title or summary: AIblogs.microsoft.com ↗
Geographic references
Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.
- United States · Place role: Coverage · Place type: Country or area · Location precision: RegionalEvidence
- Explicit preserved editorial country/region; no city or national centroid inferredblogs.microsoft.com ↗