Maia 300
Microsoft announced Maia 300, its third-generation in-house AI chip built on TSMC N2P, to be the primary inference system for its products, with production rollout in 2027.
Maia 300 succeeds Maia 200, which was announced in January. It uses a chiplet design that separates compute, memory controller and I/O. Microsoft says it will scale Azure Maia to multi-gigawatt deployments starting in 2027, and a full rack holds 96 accelerators.
- Date
- Monday 12 October 2026
- Lab
- Microsoft
- Kind
- infra
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| Tokens per megawatt versus leading GPU-based systems | 40% more expected, per Scott Guthrie | company |
| Rack-scale token throughput versus Maia 200 | up to 7x full 96-accelerator configuration | company |
| Accelerators per rack | 96 rack power about 167 kW per Reuters via Techmeme headline | company |
| Memory | 288GB HBM4 eight HBM4 stacks, per Andreas Schilling's chiplet breakdown quoted on Techmeme | company |
Chip for Microsoft's own services, not sold externally. Production workloads begin in 2027.
Sources
- theverge.com/news/1009465/microsoft-maia-300-chip-improvements-rollout
- techmeme.com/261012/p29#a261012p29
- theinformation.com/briefings/microsoft-debuts-new-homegrown-ai-chip-plans-deploy-multiple-
This record was checked and corrected against its sources on 12 October 2026. How we check