Nvidia’s AI CPU push could make agents a bigger threat to Intel and AMD

Nvidia is pushing deeper into server CPUs as AI agents create demand for the computing work around each model call. The important shift is that the agent boom may expand the market for processors, memory and virtual machines, while Nvidia tries to sell the whole rack rather than just the accelerator inside it.

NVIDIA Watch analysis

What happened

24/7 Wall St. reports that Meta’s Muse, which reached the top of the US App Store’s free iPhone chart within ten days of launch, offers a useful example of the workload Nvidia is targeting. Muse can research products, fill in forms, handle email and make reservations, with its activity running in a secured virtual machine after the user closes the app.

That kind of software still relies on GPUs for model inference, but it also needs CPUs to run browsers, direct tools, move data, manage permissions and keep tasks alive. Nvidia’s Vera CPU is designed for that surrounding work. The company positions it alongside its Vera Rubin platform and as a stand-alone server processor, using Arm architecture and 88 custom Olympus cores.

Nvidia has cited roughly $20 billion in server CPU demand and said CPU revenue could more than double in fiscal 2028. Those are company-stated opportunities, not booked sales. The wider market estimates cited by 24/7 Wall St. put AI CPU revenue at $38 billion in 2026, rising to $155 billion in 2030, although those forecasts are not company guidance.

Why it matters

The old AI hardware story was pleasingly simple: GPUs did the clever maths, and everyone else tried to secure enough of them. Agents complicate the bill. A task that browses websites, checks an inbox, calls a model several times and waits for approval consumes a mixture of accelerator time, CPU hours, memory, storage and networking.

That does not mean one agent needs one physical processor. Cloud systems share hardware between virtual machines. The useful measurements are concurrent sessions, virtual CPU hours and the amount of software work wrapped around each model request. If agents become persistent and popular, however, ordinary server computing could grow alongside accelerator demand.

The opportunity is not automatically AMD’s or Intel’s. Amazon, Google and Microsoft design their own Arm-based CPUs, while Nvidia can bundle Vera with its GPUs, networking, interconnects and software. AMD has a 256-core EPYC Venice family and Intel retains a large installed base, but Nvidia is changing the contest from an individual socket purchase into a platform decision. The socket, one might say, has been invited to join a much larger dinner party.

Our read

This is a meaningful infrastructure story, but not proof that Nvidia will sweep the CPU market or that every Muse user translates into a new server order. The strongest point is strategic: agentic software increases the amount of general-purpose computing surrounding AI, and Nvidia wants to capture that spend before rivals can make it somebody else’s problem.

For buyers and investors, the sensible question is not simply which chip has the most cores. It is which combination of CPU, accelerator, memory, networking and software delivers the lowest cost for a real workload. The forecasts are worth watching, but the eventual evidence will be shipments, utilisation, margins and independent performance data.

What to watch

• Whether Nvidia discloses actual Vera shipments and recognised CPU revenue.
• How much agent workload stays on Nvidia platforms rather than cloud providers’ own silicon.
• Independent comparisons of Vera with AMD EPYC Venice and Intel Xeon systems.
• Whether consumer agents generate sustained CPU demand or merely a brief launch spike.

Discussion spark: Will agentic AI create a broad new CPU market, or will Nvidia and the cloud giants capture most of the growth through tightly integrated platforms?

Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.