Microsoft to Scale In-House Maia AI Chips to Multi-Gigawatt Capacity by 2027

Microsoft is accelerating its push into proprietary hardware, announcing plans to deploy its Azure Maia AI chips at a multi-gigawatt scale starting in 2027. According to executive Scott Guthrie, the custom processors are set to become Microsoft's primary inference platform across its entire product lineup.

CEO Satya Nadella described the initiative as the company's "biggest bet yet on our own silicon," highlighting a broader industry trend among cloud providers to curb computing costs and decrease reliance on external semiconductor suppliers.

Maia 200 Powers Current Workloads

The current generation, Maia 200, is already operational and powering Microsoft Copilot workloads across data centers in the United States, Europe, and Australia. Deployment is scheduled to expand to GitHub Copilot and the Microsoft Foundry platform by the end of the year.

Maia 200 is designed to serve a broad range of architectures, including OpenAI models, Microsoft's internal MAI models, and third-party open-source systems. In operational benchmarks, Microsoft claims the processor delivers up to 1.7 times more tokens per watt than leading commercial GPUs when running workloads for Microsoft Teams.

Next-Gen Maia 300 Targets Agentic AI

Looking ahead to 2027, Microsoft plans to introduce Maia 300, a successor tailored specifically for demanding workloads such as long-context inputs, low-latency conversational queries, and autonomous agentic tasks.

Key technical specifications for Maia 300 include:

• Memory architecture: Eight HBM4 memory stacks offering 288 GB of total capacity.
• Bandwidth: 18 TB/s of memory bandwidth.
• Power envelope: Rated at 950 watts per chip.
• Efficiency: Delivers 40 percent more tokens per megawatt compared to leading GPU systems.

Density and throughput will also see significant jumps at the cluster level. A single 96-chip Maia 300 rack is expected to generate up to seven times the token output of an equivalent Maia 200 rack.

Microsoft has shared preliminary details in a technical report and plans to unveil further architecture disclosures at its Ignite conference.