Microsoft to make Maia its main AI chip at multi-gigawatt scale in 2027

Microsoft's Maia program is moving from a supporting role to the center of how the company serves AI answers. According to executive Scott Guthrie, the Azure Maia chips will become the company's main inference system at multi-gigawatt scale starting in 2027, a shift from today's mix.

The current Maia 200 already runs Microsoft Copilot in the US, Europe, and Australia, with GitHub Copilot and Foundry following by year's end, so the installed base and software support exist before the larger deployment. Microsoft also says Maia 200 handles OpenAI, MAI, and open-source models and pushes up to 1.7 times more tokens per watt than leading GPUs in Teams, which suggests the company has been testing the chips on real workloads.

The next chip, Maia 300, is where the economics get more ambitious. Microsoft claims it offers 40 percent more tokens per megawatt than leading GPU systems and that a 96-chip rack should produce up to seven times the tokens of a Maia 200 rack. Each chip carries eight HBM4 memory stacks totaling 288 GB with 18 TB/s of bandwidth and is rated at 950 watts. With the technical report out and more detail promised in November at Ignite 2026, the direction is clear: Microsoft intends its own silicon, not purchased GPUs, to carry the bulk of its AI serving load.