October 20, 2026
Microsoft’s Maia 300 Chip Promises 40% More AI Power per Watt — and It Ships in 2027
Microsoft has unveiled Maia 300, the third generation of its in-house AI chip — and it plans to make it the primary inference engine behind Copilot, GitHub, and its other AI products starting in 2027, at multi-gigawatt scale. Each chip packs 288 GB of HBM4 memory at 18 TB/s, and a full rack holds 96 accelerators.
Microsoft’s headline claim is efficiency: 40% more tokens per megawatt than leading GPU-based systems, and up to seven times the rack-level throughput of the previous Maia 200. If the numbers hold, it’s a direct shot at Nvidia’s grip on AI data centers — though critics note the 40% figure rests on an unnamed baseline, and Maia chips so far run almost entirely inside Microsoft’s own services.