Microsoft unveils Maia 300 AI chip, plans rollout starting in 2027

Microsoft has unveiled Maia 300, a new in-house artificial intelligence (AI) chip. The company plans to begin rolling it out starting in 2027. Anadolu Agency and The Verge both reported the unveiling and the 2027 timeline. Anadolu's report is available only as a headline, so its details and sourcing could not be checked.

What is reported

According to The Verge, Maia 300 is the third generation of Microsoft's Maia chip. It succeeds Maia 200, which The Verge says was announced in January. The Verge reports that Maia 300 will start rolling out to production AI workloads in 2027.

According to The Verge, Maia 200 already runs production AI workloads across Microsoft. These include Microsoft Copilot, GitHub Copilot, Microsoft Foundry and Microsoft's own AI models.

The Verge reports that Maia 300 is built on TSMC's N2P process node. It uses a chiplet architecture that separates compute, memory controller and input/output functions. According to the outlet, Microsoft intends Maia to become the primary AI inference system for its products and services.

The Verge offered two assessments of its own. First, it said the efficiency gains will allow Microsoft to rely mainly on its own chips rather than chips from Nvidia and AMD. Second, it said Maia 300 could help Microsoft catch up with Google DeepMind, OpenAI and Anthropic. Microsoft did not make these statements.

What each side says

The Verge quoted Scott Guthrie, executive vice president of Microsoft's cloud group, as saying: "We expect Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems." He added that in a full 96-accelerator configuration, the chip delivers "up to 7x the rack-scale token throughput of Maia 200." The reports reviewed by Vocemundi do not describe any independent verification of these figures. They also do not say which GPU-based systems were used for comparison.

Guthrie was also quoted as saying: "We will scale our Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary AI inference system for Microsoft products and services."

According to The Verge, Microsoft's Azure hardware and infrastructure team said the chiplet design "increases the silicon area dedicated to compute performance."

The Verge quoted Mustafa Suleyman, Microsoft AI CEO, as saying: "At that scale we need a platform that we own, and that we co-design with our models."

What the reports do not include

The reports do not include a primary Microsoft statement. They do not describe the setting in which the executives made their remarks. They do not give details on memory, power or cost. They do not say whether external Azure customers will have access to Maia 300. They do not include responses from Nvidia, AMD or TSMC.

Analysis

Maia 300 is Microsoft's third-generation in-house AI chip, and The Verge describes it as making the company less reliant on Nvidia's GPUs. If Microsoft's efficiency claims hold up in independent testing, the chip could reduce the company's dependence on third-party GPU suppliers for inference. That outcome is not certain. The performance figures come from Microsoft, and the reports do not identify the comparison systems. Production rollout is also not due to begin until 2027, and the reports do not give a specific month. Key open questions include whether external Azure customers will get access to the chip and how Nvidia and AMD respond. Results from real-world use after 2027 will offer a clearer test.

Produced by the Vocemundi newsroom with AI assistance.