Nadella makes Microsoft's own Maia chips its main AI inference system from 2027
The new Maia 300 is built for running models, and Microsoft expects 40% more tokens per megawatt than leading GPU systems.
On October 12, 2026, Microsoft said its own Azure Maia chips will become the primary inference system for its products and services, scaling to multi-gigawatt deployments starting in 2027. Scott Guthrie, who is responsible for Azure, set out the plan on the Official Microsoft Blog, along with a new chip, Maia 300.
Satya Nadella posted the news at 9:03 am Eastern.
Today we are announcing our biggest bet yet on our own silicon. Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services. And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems. We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot. Our goal is to build on the full-stack efficiency gains we’ve already seen, while continuing to bring the latest innovations from across the ecosystem into our fleet.
What Maia 300 is for
Inference is the work a model does after training, every time someone sends Copilot a request. Microsoft calls Maia 300 an inference-first design, tuned for long contexts and memory-heavy jobs like multi-step agent work.
The compute chiplets are built on TSMC's N2P process. Each chip gets eight HBM4 memory stacks, and a full rack packs 96 of them.
- Chip
- Maia 300
- Process
- TSMC's N2P
- Memory
- Eight HBM4 stacks, 18.0 TB/s
- Rack
- 96 accelerators, 167-kilowatt power envelope
- Production deployments
- 2027
Microsoft expects 40% more tokens per megawatt than leading GPU-based systems, and up to 7x the rack-scale throughput of Maia 200. Those are Microsoft's own projections for a chip that isn't deployed yet, and the post doesn't name the GPU systems it's measuring against.
The design bet is on memory. Guthrie's post puts it in one line:
Moving data efficiently, rather than raw compute power, is increasingly the bottleneck in large inference systems.
Where Maia already runs
Maia 200, the current generation, is serving Microsoft Copilot now and supports the MAI team building Microsoft's next models. GitHub Copilot and Microsoft Foundry are due to follow by the end of the year. Microsoft says it runs OpenAI models, its own MAI models and open-source families, at up to 1.7x more tokens per watt than leading GPU infrastructure for AI features in Teams, Copilot and Cowork.
Mustafa Suleyman, whom Nadella named in 2024 to lead Microsoft AI, kept his reaction to one line later that morning.
Our new silicon is 🔥 40% more tokens per watt!
Neither post says Microsoft will stop buying GPUs. Nadella's says it'll keep bringing "the latest innovations from across the ecosystem into our fleet," which reads to me like Nvidia stays, just with less of the work.
Microsoft says it will share more at Microsoft Ignite 2026 in San Francisco next month.