Microsoft has announced its third-generation custom artificial intelligence (AI) chip and inference platform, the Maia 300, marking an aggressive push to run more of its AI infrastructure on homegrown hardware after years of heavy reliance on Nvidia’s market-dominant processors. Production of the Maia 300 is scheduled to begin in 2027, and Microsoft plans to use the platform as its primary inference engine to process incoming AI queries across its ecosystem, powering tools such as Copilot, Microsoft Teams, GitHub, Foundry and Dragon Copilot.
Executive leadership described the rollout as the company’s “biggest bet yet on our own silicon,” positioning the processor as the future operational backbone for its consumer, developer, and enterprise services.
And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems. We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot. Microsoft plans to install 96 Maia 300 chips into every server rack, effectively doubling the 48-chip density of the Maia 200. The platform achieves up to seven times higher token throughput per rack compared to the Maia 200. Moreover, the hardware is claimed to produce 40% more tokens per megawatt, delivering notable power savings compared to standard GPUs.