Microsoft unveils Maia 300, plans multi-gigawatt rollout by 2027
On October 12, Microsoft CEO Satya Nadella announced the company's biggest bet yet on its in-house AI chips: the Maia chip family will scale to multi-gigawatt deployments, with the third-generation Maia 300 expected to enter production in 2027. Officially, it delivers about 40% more tokens per megawatt (per watt) than mainstream GPU solutions, a clear efficiency advantage.
Confirmed
- Maia 200 is already in production, being deployed across the US, Europe, and Australia, and runs production workloads like Microsoft Copilot while also supporting the MAI team's training work.
- Official figures: Maia 200 achieves up to 1.7x higher tokens/watt efficiency versus GPU systems (in scenarios such as GitHub Copilot).
- Maia 300, the next-generation chip, is expected to enter volume production in 2027 and will extend deployments to multi-gigawatt scale.
- Scott Guthrie, Microsoft's executive vice president of Cloud and AI, confirmed the timeline and performance targets in an interview.
- According to The Verge's Tom Warren, Maia 300 is positioned as the "primary AI inference system" for Microsoft's products and services — beyond cost and efficiency gains, it is optimized for the latest reasoning models and agentic workloads.
Why it matters
- This is Microsoft's largest in-house chip investment to date, marking Azure's infrastructure shift from relying on Nvidia GPUs toward a hybrid "build + buy" model that could substantially cut inference costs.
- Optimizations for agentic workloads signal that Microsoft is paving the way for the next generation of agent-driven AI services; the multi-gigawatt deployment scale also reflects its long-term bet on inference compute demand.
2026-10-12 ~ 2026-10-12 · 10 related posts
Primary sources
- Microsoft unveils Maia 300 chip, to make in-house silicon its primary inference system in multi-gigawatt 2027 scale-up — satyanadella ·
- Microsoft's Scott Guthrie on Maia 300 AI chip, insatiable AI demand, and the model price war — firstadopter ·
- Microsoft's Maia 300 chip to cut Nvidia reliance, entering production in 2027 — tomwarren ·
- [source] Microsoft unveils Maia 300 chip, to make in-house silicon its primary inference system in multi-gigawatt 2027 scale-up — satyanadella · 2026-10-12
- [source] Microsoft's Maia 300 chip to cut Nvidia reliance, entering production in 2027 — tomwarren · 2026-10-12
- Microsoft bets big on custom silicon: Maia 300 claims 40% more tokens per watt, multi-gigawatt deployments — TinfoilTricorn · 2026-10-12
- [source] Microsoft's Scott Guthrie on Maia 300 AI chip, insatiable AI demand, and the model price war — firstadopter · 2026-10-12
- Microsoft puts Maia 200 into production, plans multi-gigawatt Maia rollout from 2027 — aigclink · 2026-10-12
- Microsoft to Scale Azure Maia to Multi-Gigawatt Deployments; Maia 300 Claims 40% More Tokens per Megawatt — Recoil42 · 2026-10-12
4 near-duplicate retellings: mustafasuleyman · thoefler · SuB8u · ai