Microsoft plans multi-gigawatt rollout of Maia AI chips
The company says its next accelerator will support Microsoft services from 2027.
Microsoft says it will begin multi-gigawatt deployments of its in-house Azure Maia AI accelerators in 2027. The company presented Maia 300 as a system designed to handle inference-heavy workloads, as Microsoft seeks to reduce its reliance on off-the-shelf graphics processors.
Microsoft said on Monday that its Azure Maia family would become the primary inference system for its own products and services, with large-scale deployments starting next year. Inference is the stage where a trained model generates answers or performs tasks for users, rather than the earlier process of training the model.
The announcement centres on Maia 300, the third generation of Microsoft’s custom AI accelerator. Microsoft says it is designed for long-context reasoning, low-latency conversational systems and multi-step agentic tasks. The company claims the system will deliver 40% more tokens per megawatt than leading GPU-based systems.
Microsoft also says a full rack containing 96 Maia 300 accelerators can deliver up to seven times the rack-scale token throughput of a Maia 200 rack. Its published specification gives that rack a power envelope of 167 kilowatts and 27.6 terabytes of high-bandwidth memory.
The company’s existing Maia 200 hardware is already handling production workloads, including Microsoft Copilot, according to Microsoft. The company says Maia 200 is being deployed across the United States, Europe and Australia, with GitHub Copilot and Microsoft Foundry among the services expected to use the platform by the end of 2026.
Reuters reported that Maia 300 uses a chiplet design, with key computing functions built using 2-nanometre technology from Taiwan Semiconductor Manufacturing Company. The design places Microsoft among a growing group of cloud and AI companies developing their own processors as they seek more control over cost, performance and supply.
The efficiency figures are based on Microsoft’s own comparisons and have not been independently verified. The rollout is also a plan for 2027, not a completed deployment. The scale of the programme means chip production, advanced packaging, electricity and data-centre capacity will all be important constraints, even if the promised performance gains are achieved.
Fact-check Approved · Priya Shah — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The article accurately attributes technical specifications and performance comparisons to Microsoft. Reuters supports the chiplet design and competitive context, while the article makes clear that the rollout and efficiency claims remain prospective.
- confirmed Microsoft plans multi-gigawatt deployments of Azure Maia accelerators starting in 2027. — Announced by Microsoft and reported by Reuters. source
- confirmed Microsoft says Maia 300 will deliver 40% more tokens per megawatt than leading GPU-based systems. — This is explicitly presented as Microsoft’s own expectation and comparison. source
- confirmed A full Maia 300 rack contains 96 accelerators and has a 167-kilowatt power envelope. — The rack configuration and power figure are given in Microsoft’s announcement and Reuters’ report. source
- confirmed Maia 200 is already handling Microsoft production workloads including Copilot. — Stated by Microsoft in its official announcement. source
- confirmed Maia 300 uses a chiplet design with key computing functions built using 2-nanometre technology from TSMC. — Reported by Reuters from Microsoft’s announcement and interview. source
Editor's note
Microsoft’s deployment plans, specifications and performance claims come from the company. Reuters independently confirms the chiplet design and broader competitive context, but the stated efficiency gains remain vendor claims and the large-scale rollout is still scheduled for 2027.Sources
- Microsoft to expand Azure Maia system to multi-gigawatt deployments — Microsoft
- Microsoft plans huge deployment of its own AI chips, with an eye on saving costs — Reuters via Investing.com
More on this in other media
- The Guardian — „microsoft azure”
- BBC News — „microsoft azure”
- Sky News — „microsoft azure”