Blog
Microsoft announces Maia 300 and plans to make its own chips the main engine for its AI services
On 12 October 2026 Microsoft announced Maia 300, its third-generation in-house AI inference chip, with production deployments due in 2027. Microsoft says Maia will become the primary inference system for its products. Its efficiency figures are its own projections and have not been independently tested.
On 12 October 2026 Microsoft announced Maia 300, the third generation of its in-house AI chip. In a blog post, Scott Guthrie, Microsoft's executive vice president for Cloud + AI, said the company will scale its Azure Maia family to "multi-gigawatt deployments starting in 2027", making Maia "the primary inference system for Microsoft products and services". Maia 300 production deployments are due to begin in 2027.
Inference is the work of running trained models to answer requests, as opposed to training them. Microsoft says Maia 300 is designed for long-context reasoning, latency-sensitive chat and sustained, multi-step agent tasks.
What Microsoft claims
These are Microsoft's own projections, not independent measurements:
- 40% more tokens per megawatt than leading GPU-based systems.
- Up to 7x the rack-scale token throughput of Maia 200, in a full 96-accelerator configuration.
- For Maia 200, the current chip, Microsoft says it delivers up to 1.7x more tokens per watt than leading GPU infrastructure for built-in AI features in Teams, Copilot and Cowork.
The hardware
Microsoft's official blog and an architecture deep dive by Rani Borkar and Saurabh Dighe give these details:
- Chiplets: compute, memory-controller and I/O functions are split into separate chiplets. The compute chiplets use TSMC's N2P process.
- Memory: each accelerator has eight HBM4 stacks with 288 GB of capacity and 18 TB/s of peak bandwidth, within a 950-watt TDP. Microsoft says that is 2.5 times the memory bandwidth and 33% more capacity than Maia 200.
- Low-precision formats: support for an enhanced MXFP4 format and sub-3-bit storage formats, which Microsoft says cuts memory use and data movement and allows longer context windows.
- Racks: 96 accelerators in a 167-kilowatt rack, double Maia 200's density, with 27.6 TB of HBM in total. Liquid cooling removes more than 80% of the rack's heat, and racks can work in air-cooled data centres through a heat-exchanger unit.
- Networking: a custom Ethernet-based AI transport links up to 128 racks into a "superpod" of up to 12,288 accelerators that software can treat as one system.
Where it fits
Microsoft says Maia 200 already serves production workloads including Microsoft Copilot and runs OpenAI models, Microsoft's own MAI models and open-source model families. It is being deployed in the United States, Europe and Australia. Microsoft says Maia 200 and 300 together will support GitHub's coding tools, model serving in Microsoft Foundry, the MAI model team and Dragon Copilot for clinicians.
The shift away from outside suppliers will be gradual. GeekWire noted that Maia, unlike Google's and Amazon's custom chips, is not available for outside customers to rent. It also reported that Microsoft still relies on Nvidia as a key data-centre supplier. CNBC reported an Evercore ISI estimate that Maia will make up about 10% of Microsoft's AI chip spending by the first quarter of 2027. A Microsoft general manager declined to give CNBC a timeline for the multi-gigawatt target. CEO Satya Nadella called Maia 300 "our biggest bet yet on our own silicon" in a post on X, according to GeekWire.
Why it matters
Power has become one of the main limits on how fast AI services can grow. If Microsoft's efficiency claims hold up in production, it could serve more Copilot, GitHub and Foundry requests from the same electricity and depend less on GPU makers' pricing and supply. That could eventually affect the cost and capacity of the Microsoft AI tools many businesses use, though customers will not rent Maia directly. The figures are projections for hardware that is not yet deployed. "Nobody has tested them outside of Microsoft yet," analyst Patrick Moorhead told CNBC. Microsoft says it will share more at Ignite 2026 next month in San Francisco.
Sources
- Microsoft: Microsoft to expand Azure Maia system to multi-gigawatt deployments (12 Oct 2026)
- Microsoft Azure Blog: Deep dive into Azure Maia 300 architecture (12 Oct 2026)
- GeekWire: Microsoft makes 'biggest bet yet' on its own silicon, unveils new AI chip (12 Oct 2026)
- CNBC: Microsoft will mainly use its Maia chips for AI inference (12 Oct 2026)