Microsoft plans to scale its Maia AI accelerator family to multi-gigawatt deployments and has unveiled the Maia 300, built on TSMC's N2P process node, DigiTimes reports.

The announcement signals a shift from pilot-scale silicon to infrastructure-class volumes. Multi-gigawatt deployments put Microsoft's custom accelerator program in the same capacity conversation as the largest AI compute buildouts now underway across the hyperscaler sector.

What do we know about the Maia 300?

The key technical detail is the process node. TSMC's N2P is an enhanced variant of the foundry's 2-nanometer family, and Microsoft's adoption of it places Maia 300 among the first accelerator designs targeting that generation of silicon.

Moving to a leading-edge node typically brings higher transistor density and improved performance-per-watt — the two variables that dominate the economics of AI inference and training at datacenter scale. For a company planning multi-gigawatt rollouts, even single-digit percentage gains in efficiency translate into large absolute savings in power and cost.

Why does multi-gigawatt scale matter?

Capacity figures frame the story. A multi-gigawatt AI footprint implies accelerator fleets measured in the millions of units, alongside matching investments in power provisioning, cooling and networking.

Microsoft's in-house Maia line exists to reduce the company's reliance on third-party suppliers, chiefly NVIDIA, for AI compute. Scaling Maia to this level would make Microsoft one of the largest producers of custom AI silicon by deployment volume.

What comes next?

DigiTimes' report did not specify a production timeline or volume targets for the Maia 300. TSMC's N2P node is expected to enter volume production in the coming process-cycle window, which positions the chip for the next major capacity phase of Microsoft's datacenter expansion.

The company has not yet disclosed pricing, power envelope or architecture details for the new part.