Microsoft confirmed during Ignite 2026, on November 17, that Maia 200, its second-generation AI accelerator, is now in production and runs the company's largest models in its data centers. The chip, designed by Microsoft itself for inference, underpins the company's strategy to run services such as Microsoft Copilot with greater energy efficiency — and paves the way for Maia 300, due in 2027.
According to Microsoft, Maia 200 delivers up to 1.7x more tokens per watt than leading GPU infrastructure for AI features built into Microsoft Teams, Copilot and Cowork. The figures are the manufacturer's own claims, measured in its own tests, and have not been independently verified.
What Maia 200 brings
Originally announced on January 26, 2026, Maia 200 is manufactured by TSMC on a 3-nanometer process and packs more than 140 billion transistors. The specifications disclosed by Microsoft include:
- Over 10 petaFLOPS in FP4 and over 5 petaFLOPS in FP8;
- 216 GB of HBM3e memory with 7 TB/s of bandwidth;
- 272 MB of on-chip SRAM;
- 750 W TDP;
- Clusters of up to 6,144 accelerators connected over Ethernet.
The chip is already deployed in data centers in the United States, Europe and Australia, running OpenAI models, Microsoft's own MAI models and open-source models. According to Microsoft, GitHub Copilot and Microsoft Foundry will move to Maia 200 by the end of the year.
Maia 200 and the new MAI models
The confirmation came alongside the announcement of seven new MAI models at Ignite 2026, including MAI Code 2 and MAI Cyber 2 Flash. Microsoft positions Maia 200 as the infrastructure behind these models — and as the step before Maia 300, which promises 40% more tokens per megawatt and arrives in 2027, as detailed in our Maia 300 coverage.
The company says it will scale the Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for its products and services.
Sources
- The Official Microsoft Blog — Azure Maia at multi-gigawatt scale
- Microsoft Tech Community — Deep dive into the Maia 200 architecture
- Tom's Hardware — Microsoft introduces Maia 200
- Neowin — Microsoft takes on OpenAI and Anthropic with seven new AI models at Ignite 2026
Image: Microsoft (official blog).
Transparency: this article was produced with the assistance of artificial intelligence, based on public sources on the internet and public posts on X, and reviewed before publication. The performance figures cited are Microsoft's own claims, not independently tested results. Please consult the original sources listed above to confirm the details.
By Muse