State of AI World
Open in the map →

Microsoft Maia

AI silicon · US

Microsoft Maia is a division of Microsoft, so it is not valued separately; its value sits inside Microsoft.

Microsoft's in-house AI accelerator line and its NVIDIA hedge - the fourth hyperscaler silicon programme and, until now, the one this map did not carry alongside Google TPU and AWS Trainium.

Part of
Microsoft
Layer
AI silicon
Website
azure.microsoft.com

Relationships (3)

Microsoft Maia powers

Microsoft Maia fabricated by

Microsoft Maia part of

Notes

Microsoft's in-house AI accelerator line and its NVIDIA hedge - the fourth hyperscaler silicon programme and, until now, the one this map did not carry alongside Google TPU and AWS Trainium. Maia 300, announced 2026-10-12, is a multi-chiplet inference part: compute chiplets on TSMC's N2P process with separate memory-controller and I/O dies, eight HBM4 stacks giving 288 GB at 18 TB/s, a 950 W TDP, and 13.8 PFLOPS at FP4. A rack holds 96 accelerators in about 167 kW - against 72 GPUs at 142 kW for NVIDIA's GB300 - and Microsoft claims up to 7x the rack-level token throughput of Maia 200 and 40% more tokens per megawatt than leading GPU systems, without naming the baseline. Scott Guthrie says Azure Maia moves to multi-gigawatt deployments from 2027 and becomes the primary inference system for Microsoft's own products; The Information reported in August 2026 that Microsoft was in talks with TSMC for more than 300,000 of the chips for 2027, which Microsoft has not confirmed. The TSMC leg is `reported`, not verified: no Microsoft page naming the process surfaced and it reaches this map through trade coverage of the announcement.

Sources

See something wrong or out of date? Suggest a correction (GitHub) or email it. Please include a source.

Last updated 12 October 2026. Follow Microsoft Maia in any feed reader to hear when its valuation moves.

State of AI World maps every significant AI company, what it is worth and how it is connected. Every number carries a source and an as-of date; see the methodology. Data is free to reuse with credit: open data. · All companies · Daily digest · Corrections and contact