Physix FrontierPhysix Frontier · News Briefing Card (IT Home · Oct 12, 2026)
Microsoft Unveils Maia 300 AI Chip, Deploys in 2027
KEY FACTS
- Microsoft has officially unveiled the Maia 300 AI chip, with plans to deploy it in 2027.
- The chip is built on TSMC's N2P 2-nanometer process and uses a multi-chiplet design.
- It packs 288GB of HBM4 memory with 18TB/s of memory bandwidth.
- It delivers 13.8 PFLOPS of FP4 compute and 6.9 PFLOPS of FP8 compute.
- The SoC is rated at 950W, and a single rack can hold 96 chips.
KEY DATA
288GBHBM4 Memory Capacity
18TB/sMemory Bandwidth
13.8 PFLOPSFP4 Compute
950WSoC Power
PHYSIX OBSERVATION
With its in-house chip line now iterating to the 300 series and adopting TSMC's most advanced 2-nanometer node, Microsoft is making it plain it wants to cut its dependence on Nvidia. The 950W power draw and 12,000-chip cluster scale show this is aimed at hyperscale inference, not a small side project. For cloud providers, in-house silicon is shifting from a backup option to a primary force, and the compute cost structure could be rewritten.
Source: IT Home report