Microsoft가 Maia 200을 가동해 Copilot을 구동하는 가운데, 3세대 AI 칩 Maia 300을 공개했다.

Microsoft on Monday (Oct. 12) unveiled Maia 300, its third-generation in-house AI accelerator, and said that Maia 200 is now in production, powering Copilot and other Microsoft cloud services.

The company said it plans to scale deployments of its Azure Maia processor family at multi-gigawatt scale in 2027, positioning Maia as a primary inference system for Microsoft products and services, according to a blog post by Scott Guthrie, Microsoft's executive vice president, cloud + AI.

Microsoft said Maia 200 is deployed across its data centers in the US, Europe, and Australia, delivers up to 1.7 times more tokens per watt than leading GPU infrastructure, and powers built-in AI capabilities in Microsoft Teams, Copilot, and Cowork. By the end of the year, the chip is also expected to power GitHub Copilot and Microsoft Foundry, a platform for developing, deploying, and managing AI applications and agents, according to Guthrie's blog post.

Maia 300, which Microsoft said will be deployed in production in 2027, is designed with an inference-first architecture optimized "for long-context, memory-intensive, and decode-heavy AI workloads," according to the blog post. The company said the chip is expected to deliver 40% more tokens per megawatt than leading GPU-based systems and, in a full 96-accelerator rack, up to seven times the rack-scale token throughput of Maia 200.

Like other hyperscalers, Microsoft continues to rely heavily on Nvidia and AMD GPUs for AI workloads. At the same time, it has spent the last few years building its own custom silicon to improve performance and efficiency across its infrastructure and cloud services.

Unlike Amazon Web Services, which offers Trainium and Inferentia chips via public cloud instances, and Google Cloud, which provides access to its TPUs, Microsoft has not made Maia available to customer workloads via Azure instances. In Monday's blog post, the company did not say whether that could change in the future.

Because Microsoft is using Maia primarily for internal workloads, the competitive dynamic is different from a traditional chip war among hyperscalers, said Matt Kimball, vice president and principal analyst at Moor Insights & Strategy. "This is about Microsoft and other cloud providers delivering services faster, more efficiently, and more reliably."

"That's good for the customer because they're getting better quality of service, and it's good for the cloud provider because they're doing it at lower cost, which means they're more profitable," Kimball said.

Kimball said that Maia 300 appears to represent a significant step forward in Microsoft's custom silicon strategy. "Maia 100 was a proof of concept. Then 200 came out, and they solved the issues they had seen in the 100. They are powering GitHub and Copilot," he said. "It validated the design choices they were making, and 300 was where they said, 'Let's go big,' and that's what you are seeing with the Maia 300."

Microsoft's plan to deploy the Azure Maia family at multi-gigawatt scale suggests broad global adoption across its infrastructure, Kimball noted.

"It's going to be used very broadly to support their workloads across the globe," he said.