Microsoft Unveils Maia 300, Its Third Custom AI Chip, With Deployments Possibly Topping One Million Next Year
Microsoft unveiled the Maia 300 custom AI chip on October 12, the third generation of the family. Reuters reported that Microsoft plans to deploy the chip next year, potentially in quantities above one million, to back its internal AI services.

What happened
Microsoft introduced the Azure Maia 300 on October 12 at the OCP Global Summit in San Jose. In an official blog post, cloud and AI chief Scott Guthrie wrote that Microsoft will scale the Azure Maia family to multi-gigawatt deployments starting in 2027, making it the primary AI inference system for Microsoft products and services. CEO Satya Nadella called it the company's biggest bet yet on its own silicon in a post on X.
Reuters reported that Microsoft plans to deploy Maia 300 next year, potentially in quantities above one million chips. The previous Maia 200, introduced in January, already runs Microsoft's in-house models as well as OpenAI models delivered on Azure. Microsoft said Maia 200 is deploying across the US, Europe and Australia, with GitHub Copilot and Microsoft Foundry set to move onto it by the end of the year.
Technically, Reuters reported that Maia 300's central processor is built from chiplets, with key computing functions on TSMC's 2-nanometer process and other chiplets on cheaper 3-nanometer technology handling tasks such as data movement. Microsoft said the chip is designed for long-context reasoning, latency-sensitive conversational experiences and multi-step agentic tasks.
Key facts
- Deployment scale: Reuters reported that Microsoft plans to deploy more than one million Maia 300 chips next year; Microsoft said the Maia family reaches multi-gigawatt deployments from 2027.
- Efficiency claims: Microsoft said Maia 300 delivers about 40% more useful computing per megawatt than leading GPU-based systems, which it did not name; a 96-chip rack delivers up to seven times the token throughput of a Maia 200 rack.
- Rack power: A Maia 300 rack holds 96 chips and draws about 167 kilowatts, versus about 142 kilowatts for Nvidia's latest GB300 racks with 72 GPUs, according to Reuters.
- Manufacturing: The chips are designed by Microsoft and made by TSMC; unlike Google and Amazon, Microsoft does not rent Maia chips to outside customers, according to GeekWire.
Why it matters
Microsoft plans multi-gigawatt Maia deployments from 2027 as the primary inference system for its own products, with possibly more than a million Maia 300 chips deployed next year — a step up from Maia 200's limited rollout.
Comments
Related
- Oct 17Microsoft's MAI-Base-2 Completes Pre-Training: 2T-Parameter All-MoE Architecture, Trained 38 Days on 16,000 GB200s
- Oct 13Drex 1.5 vs Decision-1: The Models That Score, Not Write
- Oct 10Microsoft Launches Decision-1, a Decision Model That Outputs Calibrated Probabilities, Not Text
- Oct 2Microsoft Ships Its Own Voice Stack: 60-Language Streaming Transcription + Seconds-Level Voice Cloning
- Oct 18Microsoft Copilot Studio Launches Agent Accounts: Own Credentials, Shared Business Memory, Auto Triggers
Today
Oct 19 Monday- Brief
- GitHubGitHub Weekly Vol.1: 8 Projects Worth Your Time
Tested or read — weeklyAll →
- Quiz
- Call
Latest news
All →- Oct 18Anthropic to Commit 5% of Shares to New Public Benefit Fund
- Oct 19Stably AI's Orca Tops GitHub Daily Trending: One Desktop App to Orchestrate 30+ Coding Agents
- Oct 19Jailbroken PS5 Becomes a Local AI Server: Qwen3.8-27B Hits ~20 tok/s, Small Models Top 100
- Oct 19Anthropic Internal Study: Junior Share of Technical Staff Falls to 14%, Meeting Time Up 20%
- Oct 18Claude Cracks 25-Year Cryptography Mystery in Half an Hour: NIST Curve Seeds Came From a 1997 Clock
- Oct 18OpenCode Hit by "Distillation Attack": API Keys of About 1.4% of Users Compromised
™