AI Chips · Product page

Microsoft Maia 100: The Complete Guide

The chip that ended Microsoft’s total reliance on other companies’ AI silicon — not for sale or rent, built purely to make Microsoft’s own AI cheaper to run. Simple to expert, every number links to its source.

Stuck at any point? Ask an AI about this page →
Ask an AI about this chip:CAGPX
Microsoft Maia 100 10 sections · 2 levels 6 linked sources Checked September 2026
How to read this page. Each section starts Simple, then goes to a Deep dive: stop wherever you have what you need. The small numbers are sources: click one to open the original document. Where the maker has not published a figure -- a price, a die size, a factory address -- this page says so rather than estimate it.

1.At a glance

SimpleStart here

Maia 100 is Microsoft’s first AI chip — announced 15 November 2023 at Microsoft Ignite and rolled out to Microsoft’s own Azure data centres from early 2024 1. It was designed to run Microsoft 365 Copilot and Azure OpenAI Service more cheaply than renting GPUs alone 1.

Deep diveThe technical detail

OpenAI’s Sam Altman endorsed the chip at launch: “Azure’s end-to-end AI architecture, now optimized down to the silicon with Maia, paves the way for training more capable models and making those models cheaper for our customers” 1. Maia 100 is not for sale or rent — like every Maia generation, Microsoft uses it only for its own services (see the Maia 200 page and the Microsoft brand guide on this site).

2.Launch and history

SimpleStart here

Before Maia, Microsoft ran AI entirely on chips bought from NVIDIA and AMD. Maia 100 was Microsoft’s first attempt at its own silicon for large language models, announced alongside Cobalt, its Arm-based server processor 1.

Deep diveThe technical detail

Microsoft’s Scott Guthrie, then head of its Cloud + AI Group, framed the reasoning as operational: “At the scale we operate, it’s important for us to optimize and integrate every layer of the infrastructure stack to maximize performance, diversify our supply chain and give customers infrastructure choice” 1. Brian Harry, a Technical Fellow on the Azure Maia team, said the chip “was specifically designed for AI and for achieving the absolute maximum utilization of the hardware” 1.

3.What’s inside it

SimpleStart here

Sixteen clusters of four tiles, each tile carrying its own tensor unit for matrix maths and a vector processor for other work 2.

Deep diveThe technical detail

Maia 100 was the first chip to use the Microscaling (MX) number formats — open 4-, 6- and 8-bit formats Microsoft developed jointly with AMD, Arm, Intel, Meta, NVIDIA and Qualcomm through the Open Compute Project 3 4. Smaller number formats let the chip do more calculations per second when the extra precision of a larger format isn’t needed.

4.Full spec table

SimpleStart here
SpecMicrosoft Maia 100
AI compute3 POPS (6-bit); 0.8 POPS BF16 (dense) 2
Memory64 GB HBM2e 2
Memory bandwidth1.8 TB/s 2
Power500 W (designed for up to 700 W) 2
Process nodeTSMC 5 nm 2
Die sizeAbout 820 mm² 2
Networking12 × 400 Gb Ethernet ports, 600 GB/s bandwidth 2
Deep diveThe technical detail

Microsoft’s own description called Maia 100 “one of the largest processors made on 5nm node using advanced packaging technology from TSMC” 3. That packaging is TSMC’s CoWoS-S, which joins the compute die and its memory on one base 2.

5.Where it’s made

SimpleStart here

Designed by Microsoft’s own silicon teams, tested in a chip lab on the Redmond campus, and manufactured by TSMC on its 5 nm process 1 5.

Deep diveThe technical detail

Microsoft does not name which TSMC factory makes Maia 100, or its memory supplier. Its successor, Maia 200, moved to TSMC’s more advanced 3 nm process (see that page on this site) 5.

6.Which systems use it

SimpleStart here

Deployed only inside Microsoft’s own Azure data centres, starting in early 2024, to run Copilot and Azure OpenAI Service workloads 1. There is no Maia 100 virtual machine a customer can rent.

Deep diveThe technical detail

Maia 100 racks shipped with a “sidekick” cabinet that pumps cold liquid to plates mounted on the chips — Microsoft’s own comparison: “a bit like a radiator in a car” 1. Microsoft has not published which specific Azure regions ran Maia 100, unlike Maia 200, where it later named Iowa and Arizona explicitly (see that page on this site) 5.

7.Official pricing

SimpleStart here

Not for sale or rent. Microsoft uses Maia 100 only for its own services — neither its announcement nor its later Azure blog post mentions a chip price or an Azure virtual machine built on it 1 3.

Deep diveThe technical detail

This matches every Maia generation checked on this site: Microsoft has never published a Maia chip price or offered a Maia rental instance (see the Maia 200 page on this site for the same result).

8.Real-world performance

SimpleStart here

Microsoft’s own launch materials do not give a head-to-head performance claim against a named competing GPU for Maia 100, unlike the direct comparisons it later published for Maia 200 (see that page on this site) 1 3.

Deep diveThe technical detail

The clearest performance signal for this generation is indirect: press reporting on its successor said Microsoft had dropped plans for a separate Maia training chip in early 2024, suggesting Maia 100 itself did not fully satisfy Microsoft’s training ambitions, and that the delayed Maia 200 was expected to fall well short of the performance of NVIDIA’s flagship Blackwell chip — a press claim, not a Microsoft one 6.

9.What came before, what came next

SimpleStart here
Came before: Project Brainwave, Microsoft’s FPGA-based AI system, not a custom chip (see the Microsoft brand guide on this site). Came after: Maia 200, covered on its own page on this site.
Deep diveThe technical detail

Maia 200 arrived more than two years after Maia 100’s November 2023 announcement — later than Microsoft originally planned. Press reporting from June 2025 said the chip that became Maia 200, code-named Braga, had slipped at least six months into 2026, citing design changes including features OpenAI requested and staff turnover 6.

10.Hidden in plain sight

SimpleStart here

Maia 100 runs below its own design limit: Microsoft built it for up to 700 W but operates it at 500 W 2.

Deep diveThe technical detail

Maia 100 was the first chip anywhere to use the Microscaling (MX) formats — a genuinely open, royalty-free standard co-developed with six other major chip and AI companies through the Open Compute Project, rather than a Microsoft-only proprietary format 3 4.

11.Sources

6 sources, checked September 2026. Where NVIDIA or another maker has not published a figure, this page says so rather than estimate it.

  1. Microsoft Source: With a systems approach to chips, Microsoft aims to tailor everything ‘from silicon to service’ (15 Nov 2023)Official
  2. Microsoft at Hot Chips 2024: Inside Maia 100 (Xu and Ramakrishnan)Conference
  3. Microsoft Azure Blog: Azure Maia for the era of AI: from silicon to software to systems (3 Apr 2024)Official
  4. Open Compute Project: AMD, Arm, Intel, Meta, Microsoft, NVIDIA and Qualcomm standardize narrow-precision data formats (MX)Standard
  5. Official Microsoft Blog: Maia 200: the AI accelerator built for inference (26 Jan 2026)Official
  6. Tom’s Hardware (press, citing The Information): Microsoft’s own AI chip delayed six months (Jun 2025)Press

Ask an AI about this page

Opens your assistant with this page as the source, and a question rather than a summary. It will ask what you are building before it answers.

ChatGPTClaudeGeminiPerplexityGrok

Nothing is sent from here. The link carries only this page’s title and address.