Patrick Moorhead on Maia 300: ‘primarily a margin lever, not a benchmark story’
Patrick Moorhead @PatrickMoorheadX
Why it matters
Industry analyst’s assessment with the caveat that the chip is untested and GPUs will move before 2027.
Summary
Posted Oct 12, 2026 (13:20 UTC). Moorhead (Moor Insights & Strategy) says he never expected Microsoft ‘to hit it with Maia V1’, that V2 ‘runs Copilot in production’ and V3 ‘is the commitment’. He repeats Microsoft’s figures (288GB HBM4 at 18TB/s in 950W, Ethernet scale-up to 12,288 accelerators, 40% more tokens per MW than ‘leading GPU-based systems’) and adds: ‘We haven’t tested it, and GPUs won’t stand still before 2027. At this scale, Maia is primarily a margin lever, not a benchmark story.’
Metrics as of 2026-10-13 via api.fxtwitter.com: 13,664 views, 142 likes, 10 reposts. Author: about 69,439 followers.
Archived text
I never expected @Microsoft to hit it with Maia V1. No one hits a V1. V2 got more competitive and runs Copilot in production.
V3 is the commitment and very real: multi-gigawatt scale from 2027, making Maia the “primary inference system” for Microsoft’s own branded products and services. Think primarily SaaS and PaaS. That’s huge.
Maia 300 is a systems play again: 288GB HBM4 at 18TB/s in 950W, Ethernet scale-up to 12,288 accelerators. Microsoft expects 40% more tokens per MW than “leading GPU-based systems.”
We haven’t tested it, and GPUs won’t stand still before 2027. At this scale, Maia is primarily a margin lever, not a benchmark story. More details to come from @MattKimball_MIS and I. $MSFT
