Microsoft has unveiled Maia 300, its next-generation AI accelerator, fabricated on TSMC's N2P process node, and confirmed plans to scale its Maia chip family to multi-gigawatt AI datacenter deployments, DigiTimes reported.

The announcement marks a significant step in Microsoft's in-house silicon strategy. Maia 300 follows the original Maia 100, moving from the previous node to TSMC's N2P — an enhanced version of the foundry's 2-nanometer platform. N2P sits behind only the initial N2 release in TSMC's nanosheet-transistor roadmap, giving Microsoft access to density and power-efficiency gains that matter directly for AI inference and training workloads.

How large is 'multi-gigawatt'?

The scale figure is the story's core number. Microsoft is not talking about pilot racks or single-cluster rollouts — the company frames Maia as infrastructure for datacenter capacity measured in gigawatts of power draw. That puts the chip family in the same conversation as the largest AI compute buildouts now underway across the hyperscaler ecosystem.

For context on magnitude: a multi-gigawatt footprint implies hundreds of thousands of accelerators per gigawatt, depending on per-chip power envelopes. Microsoft has not yet disclosed Maia 300's TDP, packaging details, memory configuration, or shipment volumes, so any per-gigawatt chip count remains an estimate rather than a confirmed figure.

What changes with the N2P node?

The move to TSMC N2P carries commercial weight beyond the spec sheet:

• N2P is a backside-power-ready refinement of TSMC's 2nm family, positioned for performance-per-watt gains over first-generation N2
• Power efficiency at the node level compounds across gigawatt-scale deployments, where even single-digit percentage improvements translate into large operating-cost and capacity differences
• Securing leading-edge capacity at 2nm-class nodes places Microsoft among TSMC's priority customers competing for early N2-family wafer allocation, alongside Apple, NVIDIA and other top-tier buyers

TSMC has not publicly confirmed volume allocations or wafer pricing tied to Microsoft's orders, and Microsoft has not stated when Maia 300 enters mass production.

Why does this matter for the supply chain?

Microsoft's Maia program exists to reduce dependence on merchant GPU suppliers, chiefly NVIDIA, for its Azure AI infrastructure. Each Maia generation deployed at scale shifts a slice of Microsoft's AI capex from purchased accelerators to in-house silicon, while still anchoring demand at TSMC — meaning the foundry wins either way.

The multi-gigawatt framing also signals that Microsoft expects its aggregate AI compute demand to keep expanding fast enough that custom silicon scales alongside, not instead of, merchant GPUs. DigiTimes' report did not specify deployment timelines, datacenter locations, or how Maia 300 capacity will split between training and inference.

What comes next?

Microsoft will need to confirm Maia 300 production timing, performance targets and deployment schedules; until then, the multi-gigawatt commitment stands as the announcement's hardest fact, and the competitive pressure it puts on merchant accelerator pricing and TSMC's 2nm-class allocation will be the dynamics to watch.