Microsoft has built its own AI chip, a move The Verge reports will make the company less reliant on Nvidia's GPUs, the hardware that currently dominates the market for training and running large AI models.

The announcement matters because of who is making it. Microsoft is one of Nvidia's largest customers, spending billions of dollars on GPUs to power its Azure cloud and the AI services built on top of it, including the systems behind its OpenAI partnership. By designing its own accelerator, Microsoft joins Google, Amazon and Meta in the effort to take control of a critical component of the AI supply chain.

Why would Microsoft design its own AI silicon?

The reasoning is partly commercial and partly strategic. Nvidia's data-center GPUs have been in tight supply since the generative-AI boom began, with demand consistently outstripping what the company can ship. Cloud providers that depend on those GPUs face long delivery timelines and premium pricing. Custom silicon gives a hyperscaler an alternative source of AI compute at scale — one it controls directly, from design specifications to deployment in its own data centers.

There is also a margin argument. When Microsoft buys Nvidia GPUs, a large share of the value of AI workloads flows to the chip supplier. When Microsoft designs its own chip, more of that value stays in-house, even if the manufacturing still goes to a third-party foundry.

What does this mean for Nvidia?

Nvidia remains the performance leader in AI accelerators, and a single custom chip program from one customer does not change that overnight. But the pattern is now unmistakable: the biggest buyers of Nvidia's GPUs — the hyperscale cloud companies — are each developing in-house alternatives. If Microsoft's own silicon proves capable of handling a meaningful share of its AI workloads, the volume of GPUs Microsoft needs from Nvidia could flatten even as its AI business grows.

For Nvidia, the risk is not losing Microsoft as a customer outright. It is watching a portion of future AI compute demand shift to chips it does not sell.

How does this fit the broader custom-silicon trend?

Microsoft is following a path its rivals cleared years ago. Google has designed its own tensor processing units for roughly a decade and now deploys them across its cloud. Amazon acquired chip designers and built its Trainium and Inferentia lines for machine-learning workloads. Meta has developed its own accelerator programs as well. Microsoft had largely relied on merchant silicon for AI compute, so its entry into custom accelerator design closes a gap with its closest competitors.

For the semiconductor supply chain, each new hyperscaler chip program adds design demand to foundries and advanced packaging suppliers, even as it redirects GPU demand away from Nvidia. The AI hardware market is no longer a single-vendor story — it is becoming a mix of merchant accelerators and captive silicon built by the cloud companies themselves.

Microsoft has not said its own chips will replace Nvidia GPUs across its infrastructure, and Nvidia hardware will remain central to Azure's AI offerings for the foreseeable future. But the message to the GPU maker is clear: its largest customers are now also its emerging competitors, and they are building the alternative one chip at a time.