Microsoft is announcing its third-generation Maia in-house AI part today, designed to be the “primary AI conclusion system” for Microsoft’s products and services. Maia 300 not lone improves the disbursal and effectiveness of operating AI workloads in the cloud, it’s additionally additional capable of handling the latest reasoning models and agentic tasks.
Maia 300 volition commencement being rolled out to manufacturing in 2027, and it has been designed alongside effectiveness in mind. “We anticipate Maia 300 to provision 40 percent additional tokens per megawatt compared to foremost GPU-based systems,” says Scott Guthrie, administrator vice chairman of Microsoft’s haze group. “And in a complete 96-accelerator configuration, it delivers up to 7x the rack-scale token throughput of Maia 200.”
These effectiveness improvements volition authorize Microsoft to depend chiefly on its in-house AI chips soon, alternatively of ones from Nvidia and AMD. “We volition measure our Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the chief AI conclusion scheme for Microsoft products and services,” says Guthrie.
Maia 300 is a successor to the 200, which was lone announced in January. Maia 200 is already operating on manufacturing AI workloads throughout Microsoft, including Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft’s own AI models.
“Maia 200 is the finest performing AI infrastructure in our fleet, measured in genuine earth metrics for AI services,” says Guthrie. “Maia 300 is a complete AI scheme designed for the new types of AI workloads emerging today: Long-context reasoning, latency-sensitive conversational experiences, and sustained, multi-step agentic tasks.”
Microsoft is using a new chiplet architecture to get its all-important effectiveness gains. Built on TSMC’s N2P procedure node, the latest Maia 300 architecture separates compute, recollection controller, and I/O functions. “This increases the silicon area dedicated to compute achievement and allows all function to use the innovation finest suited to its role,” says the Azure hardware and infrastructure team.
Maia 300 is additionally a key part of Microsoft’s own AI example work, and could assistance the business accomplish its goal of catching up alongside Google DeepMind, OpenAI, and Anthropic. “We have to activity throughout the complete stack, all the way downward to the silicon,” says Microsoft AI CEO Mustafa Suleyman. “At that measure we need a phase that we own, and that we co-design alongside our models. That’s why operating our Microsoft AI models on top of the Maia part is a crucial concentration for our team.”
Follow topics and authors from this narrative to see additional akin this in your personalized homepage nourish and to obtain email updates.
