Maia 200 is Microsoft's second-generation inference chip: 3 nm, about 10 FP4 petaflops at 750 W, 216 GB of HBM3e and 272 MB of SRAM. Microsoft says it delivers three times the FP4 performance of Trainium3 and Ironwood. It runs in Azure's US Central region and serves Microsoft's and OpenAI's models rather than customer workloads.
Key Takeaways
- What it is — Hyperscaler custom silicon. Inference ASIC with large on-chip SRAM, scaled over standard Ethernet.
- Best for — Nobody buys Maia directly. It matters because it lowers the cost of the Azure models you already pay for.
- Watch out — Microsoft's comparison with Trainium3 and Ironwood is its own claim, on FP4, without independent benchmarks.
At a glance
| Company | Microsoft |
|---|---|
| Current chips | Maia 200 (TSMC 3 nm), announced January 2026 |
| Architecture | Inference ASIC with large on-chip SRAM, scaled over standard Ethernet |
| Memory | 216 GB HBM3e plus 272 MB on-chip SRAM per chip |
| Scale-up | Four accelerators per tray, 2.8 Tb/s bidirectional scale-up bandwidth per chip over Ethernet |
| Software | Internal Azure stack serving Microsoft and OpenAI models |
| Where to get it | Not rented directly; you reach it through Azure AI services and Copilot that run on it |
| Pricing | No per-chip price; Microsoft claims 30% better performance per dollar than its prior hardware |
What changed in 2026
- Jan 2026. Microsoft unveils Maia 200, deployed first in US Central near Des Moines, with US West 3 to follow.
Scores
Scored 1 to 5 on the same five axes as every vendor in the AI chip companies analysis.
| Performance | Scale-up | Software | Access | 2026 momentum | Total / 25 |
|---|---|---|---|---|---|
| 4 | 3 | 2 | 1 | 3 | 13 |
Verdict
Not a procurement option, but a signal: Azure inference pricing now has a custom-silicon floor under it, as Google's and AWS's do.
Alternatives
In the same lane (hyperscaler custom silicon):
- Google TPU: large, steady training runs and high-volume serving of standard transformer models, by teams already on Google Cloud.
- AWS Trainium: teams already committed to AWS that want a cheaper path for large PyTorch training and inference.
In other lanes:
- AMD Instinct (merchant gpu)
- Cerebras (inference specialist)
- Groq (inference specialist)
- SambaNova (inference specialist)
- Tenstorrent (open hardware)
Head-to-head comparisons
- Google TPU vs AWS Trainium vs Nvidia GPU: The three ways to train and serve frontier-scale models in 2026, scored on scale-up domain, software, access and price.
- Cerebras vs Groq vs SambaNova: The SRAM-first inference specialists compared on speed, token price, deployment options and company risk.
New to custom AI silicon? Our partner site CTAIO explains how TPUs, pods and GPU clusters differ in TPU vs GPU.
Frequently asked questions
What is Microsoft Maia?
Maia 200 is Microsoft's second-generation inference chip: 3 nm, about 10 FP4 petaflops at 750 W, 216 GB of HBM3e and 272 MB of SRAM. Microsoft says it delivers three times the FP4 performance of Trainium3 and Ironwood. It runs in Azure's US Central region and serves Microsoft's and OpenAI's models rather than customer workloads.
How do you get access to Microsoft Maia?
Not rented directly; you reach it through Azure AI services and Copilot that run on it. Pricing: No per-chip price; Microsoft claims 30% better performance per dollar than its prior hardware.
What software does Microsoft Maia run?
Internal Azure stack serving Microsoft and OpenAI models.
Is Microsoft Maia a good alternative to Nvidia?
Not a procurement option, but a signal: Azure inference pricing now has a custom-silicon floor under it, as Google's and AWS's do. It fits best for: nobody buys Maia directly. It matters because it lowers the cost of the Azure models you already pay for. The main caveat: microsoft's comparison with Trainium3 and Ironwood is its own claim, on FP4, without independent benchmarks.
Sources
Facts verified 2026-10-31. Vendor performance figures are as stated by the vendor.
Ready to Find the Right AI Tools?
Browse our data-driven rankings to find the best AI tools for your team.