Microsoft unveils Maia 300 AI chip, plans 1M+ deployments in 2027 to cut Nvidia reliance

What happened

Microsoft on Monday unveiled Maia 300, its third-generation custom AI accelerator, and said it plans to deploy potentially more than one million of the chips next year to back its internal AI services, with large-scale rollout beginning in 2027 at gigawatt scale. In a blog post, Azure chief Scott Guthrie said Maia will become 'the primary AI inference system for Microsoft products and services,' and CEO Satya Nadella called it 'our biggest bet yet on our own silicon.' The chip uses a multi-chiplet design — TSMC N2P (2nm) compute chiplets plus cheaper 3nm I/O chiplets — with 288 GB of HBM4 memory and 18 TB/s of bandwidth, delivering up to 13.8 PFLOPS of FP4 compute. Microsoft claims 40% more tokens per megawatt than 'leading GPU-based systems' and 7x the rack-scale token throughput of Maia 200, with 96 accelerators packed into a 167-kW rack. Microsoft stock rose 1.1% to $540.84 Monday while Nvidia fell 1.5% and AMD sank 3.4%.

Who’s affected

• MSFT (watchlist): margin/cost lever on AI services; stock +1.1% to $540.84 Monday
• NVDA (watchlist): competitive overhang from custom-inference shift; -1.5% Monday
• AMD (watchlist): same overhang; -3.4% Monday
• TSM: positive manufacturing read-through — N2P (2nm) compute chiplets, reports of Microsoft in talks for 300,000+ chips for 2027
• Semiconductor sector: chip stocks fell broadly (SOX -2.3%, fourth straight decline) as inference-custom-silicon narrative intensifies
• Hyperscaler custom-silicon cohort: validates Google TPU / Amazon Trainium positioning

Sources

reuters.combarrons.comgeekwire.comnotebookcheck.net