Microsoft’s Satya Nadella unveils custom AI chip Maia 100

Like many of his peers, Microsoft CEO Satya Nadella is keen to ring in a new golden age of his technology thanks to AI. However, he’s not satisfied that the fate of this dream may lie outside his control.

Which is why Nadella has tasked his team to build Microsoft its own powerful silicon chip to disrupt a market which, thus far, Nvidia has largely tied up.

Fresh from watching India face off against New Zealand in the Cricket World Cup, Nadella took the stage at Microsoft’s Ignite conference to reveal the tech giant’s first ever custom silicon called Azure Maia 100. 

The in-house designed AI “accelerator”—a fancy name for a type of semiconductor— is tailored for use in ultra high-bandwidth data centers that power large language models (LLMs).

Maia poses a potential alternative to Nvidia’s A100 and H100 processors which currently dominate the AI chip market. 

“We’re already testing this with many of our own AI services, including the GitHub Copilot,” Nadella said on Wednesday. “We will roll out Maia accelerators across our fleet, supporting our own workloads first and we’ll scale it to third party workloads after.”

One of those potential customers is OpenAI’s Sam Altman. The CEO behind ChatGPT counts Microsoft as a major shareholder, and thus far he’s had to rely on the increasingly scarce silicon from Nvidia to power his AI chatbot.

“We were excited when Microsoft first shared their designs for the Maia chip, and we’ve worked together to refine and test it with our models,” Altman said in a statement on Wednesday. 

Nadella said the workforce will enter an “Age of Copilots” thanks to AI—defined by cutting-edge generative artificial intelligences with the performance of a supercomputer packaged in human-like behavior.

The Maia chip features 105 billion transistors using 5 nanometer nodes, the most advanced technology for miniaturizing semiconductor circuits that is commercially available. 

It also comes with its own “sidekick” thermal management system that replaces conventional fans with sophisticated liquid-cooling for higher performance.

“AI power demands require infrastructure that is dramatically different from other clouds,” reasoned Nadella.

Together with its sidekick, the chips are then mounted onto specially designed server racks which can be dropped into existing Microsoft datacenters rather than requiring new investment. 

Nvidia struggles to meet demand from buyers like Elon Musk

The Azure Maia 100 chip represents the final puzzle piece for Microsoft to deliver an entire system capable of tackling the heavy workloads needed for AI training and inference.

By controlling each facet—including the requisite software, server rack and cooling system—the bespoke design can better ensure that the whole is greater than the sum of its constituent parts. 

“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,” OpenAI’s Altman said. 

Nvidia founder and CEO Jensen Huang, who has been struggling to meet the voracious demand for his AI chips from buyers like Tesla boss Elon Musk, appears to have taken the news well.

He joined Nadella on stage at the Ignite conference to talk about their ongoing collaboration in various areas of the tech industry.

After all, Microsoft is not switching wholesale to Maia. Nadella made a pledge to offer Huang’s latest AI training chip, the H200 tensor core GPU revealed just this week, to Microsoft datacenter customers from next year as an alternative. 

“It’s hard to believe that it’s just been a year since ChatGPT first came out,” Nadella said. “We’re at a tipping point. This is clearly the age of copilots.”

Subscribe to the Eye on AI newsletter to stay abreast of how AI is shaping the future of business. Sign up for free.

link