Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs

THE NEWS

What happened

Microsoft announces its third-generation Maia 300 AI chip, designed to serve as the primary inference system for its products starting in 2027. Built on TSMC's N2P process with a new chiplet architecture, the chip aims to deliver 40% more tokens per megawatt and up to 7x rack-scale throughput compared to previous generations, reducing reliance on Nvidia GPUs.

CONTEXT

Why it matters

Microsoft is shifting its AI infrastructure strategy with the upcoming Maia 300 chip. Scheduled for production rollout in 2027, this third-generation silicon aims to become the primary inference engine for Microsoft's entire ecosystem, including Copilot and GitHub Copilot. Built on TSMC's N2P process node and utilizing a new chiplet architecture, the Maia 300 is designed to deliver significant efficiency gains, offering up to 40% more tokens per megawatt and seven times the rack-scale throughput of its predecessor. This move signals Microsoft's commitment to owning its AI stack from silicon to software, potentially reducing dependence on Nvidia GPUs while targeting advanced workloads like long-context reasoning and agentic tasks.

AT A GLANCE

Key facts

• Maia 300 is Microsoft's third-generation in-house AI inference chip.
• Production rollout for Maia 300 begins in 2027.
• The chip is built on TSMC's N2P process node using a new chiplet architecture.
• Microsoft expects 40% more tokens per megawatt compared to leading GPU-based systems.
• In a 96-accelerator configuration, Maia 300 delivers up to 7x the rack-scale token throughput of Maia 200.
• Maia 200 is currently running production workloads for Copilot, GitHub Copilot, and Microsoft Foundry.
• The chip targets long-context reasoning, latency-sensitive experiences, and multi-step agentic tasks.
• Microsoft aims to scale Azure Maia to multi-gigawatt deployments by 2027.