Microsoft has revealed the Maia 300, its next-generation custom AI chip, positioning the accelerator for what the company describes as widespread deployment across its datacenter footprint.

The announcement, surfaced through Yahoo Finance on Thursday, confirms that Microsoft is pushing ahead with its in-house silicon strategy rather than relying solely on Nvidia's GPUs for AI training and inference workloads. That shift carries real commercial weight for the supply chain: Microsoft operates one of the world's largest cloud infrastructures, and every Maia wafer that displaces an Nvidia order reshapes demand at TSMC, where both companies fabricate their flagship AI accelerators.

What do we know about the Maia 300?

The source report is thin on specifications. Microsoft has not yet published performance figures, process node details, memory bandwidth numbers, or a firm production date for the Maia 300. What the company has signaled is intent: the "widespread deployment" language marks a step beyond the limited pilot placement of the first-generation Maia 100.

That framing matters. First-generation custom chips typically land in a handful of servers for validation, power and thermal testing, and software-stack hardening. A second generation announced with deployment-scale ambition suggests Microsoft's engineers cleared those internal gates with the initial design.

Why does custom silicon change Microsoft's cost picture?

Building your own accelerator is a classic hyperscaler lever. Amazon has done it with Trainium and Inferentia at AWS; Google has run Tensor Processing Units through four-plus generations. Custom silicon lets a cloud operator tailor the chip to its own workloads, its own networking, and its own cooling — and it gives the operator negotiating leverage against merchant GPU suppliers whose top-end parts have carried steep premiums during the AI buildout.

For Microsoft, the calculus is straightforward. The company has committed tens of billions of dollars to AI infrastructure, and CFO commentary over recent quarters has repeatedly flagged capacity constraints as a limit on revenue growth. An in-house accelerator that scales gives Microsoft a second supply lane — one it controls — alongside its purchases of Nvidia's Blackwell-family GPUs.

What are the open questions?

Plenty, for now. Microsoft has not said which foundry will build the Maia 300, what node it targets, or when volume shipments begin. The company also hasn't quantified how many Maia 300 units it plans to deploy, nor whether the chip will serve training, inference, or both.

The competitive read is still clear. Nvidia retains the merchant-market lead, and Microsoft will keep buying its GPUs in volume. But the Maia 300 signals that Redmond intends to own a growing slice of its AI compute stack outright — and as deployment scales, expect tighter scrutiny of what share of Microsoft's AI capacity runs on silicon it designed itself.