Microsoft makes Maia its primary inference system from 2027, deploying Maia 200 across three regions

Satya Nadella called the move Microsoft's 'biggest bet yet on our own silicon.' Maia will scale to multi-gigawatt deployments starting next year. Maia 200 already delivers up to 1.7x more tokens per watt in Teams and Copilot, and Maia 300 claims 40% more tokens per watt than leading GPU-based systems.

GitHub Copilot and Microsoft Foundry are expected to move to Maia by the end of 2026. One report says Maia 300 has about 20% less memory bandwidth than NVIDIA's Rubin.

Axios frames this as hyperscalers turning from NVIDIA customers into rivals. Mustafa Suleyman also claimed Microsoft AI matched DeepSeek-V4-Pro-Base's pre-training score on held-out code using 12x less compute.

Watch delivery against the 2027 targets.