Microsoft has debuted a new homegrown AI chip and plans to deploy multiple gigawatts of capacity built on it next year, according to The Information, marking the company's most aggressive move yet to reduce reliance on Nvidia for its data center buildout.

The announcement confirms what supply chain watchers have suspected since the company first stepped into custom silicon: Redmond now treats its own accelerators as a pillar of its AI infrastructure strategy, not a side experiment. Deploying multiple gigawatts of compute means retrofitting and expanding entire data center campuses around silicon Microsoft designs itself.

Why does a homegrown chip matter for Microsoft?

Custom silicon shifts the economics of AI infrastructure. When Microsoft designs its own accelerator, it pays for silicon and packaging rather than Nvidia's margin on top of them. At gigawatt scale, even modest per-chip savings compound into billions of dollars across a deployment cycle.

It also changes Microsoft's negotiating position. A credible internal alternative gives the company leverage in every procurement conversation with merchant GPU suppliers, at a moment when AI accelerators remain supply-constrained and pricing power sits heavily with the vendor.

The move parallels strategies already visible across the hyperscaler segment. Google runs its TPU family at scale for both internal workloads and, increasingly, external cloud customers. Amazon has iterated through multiple generations of Trainium and Inferentia. Microsoft's earlier in-house effort, the Maia series, put the company on the same path; a new debut signals that internal roadmap is advancing rather than stalling.

What does 'multiple gigawatts' actually mean?

Gigawatts, not chip counts, have become the standard unit for AI infrastructure planning, because power — not wafer supply — is now the binding constraint on data center expansion. A multi-gigawatt deployment implies dozens of new facilities or major expansions at existing campuses, each with dedicated power contracts, cooling infrastructure, and network backbones sized for accelerated computing.

The Information reports this capacity is planned for next year. That timeline is tight. Chips destined for 2026-era deployments are already in production or late-stage design at the foundry today, which means the new part is likely taping out or manufacturing now through Microsoft's foundry partners.

Analysts tracking the custom silicon market generally expect hyperscaler internal accelerators to take a growing share of AI compute within their own clouds, though merchant GPUs will remain dominant for training frontier models in the near term. Microsoft has not disclosed specific performance figures, wafer allocations, or process node details for the new chip, and The Information's report did not include them.

Who loses if this works?

The immediate commercial pressure falls on Nvidia, which still supplies the majority of AI accelerators across Microsoft's Azure fleet. Every gigawatt Microsoft fills with its own silicon is a gigawatt of demand that never reaches Nvidia's order book — meaningful at a time when hyperscalers represent the largest concentration of AI chip spending in the industry.

AMD and other merchant vendors chasing hyperscaler business face a similar dynamic: the biggest customers are increasingly competitors. For foundries and advanced packaging suppliers, however, Microsoft's expansion is pure incremental demand, whichever designer's logo ends up on the die.

Microsoft has framed its silicon program around giving customers infrastructure choice, positioning custom accelerators as one option alongside merchant GPUs in the Azure catalog.

What comes next?

With multiple gigawatts of its new chip planned for next year, Microsoft's deployment pace will be one of the clearest signals of whether hyperscaler silicon can genuinely displace merchant GPU demand at scale — or simply add to it.