Nvidia has won central processor business with Modal, a fast-growing cloud startup that rents computing power to AI developers and until now ran that work entirely on chips from Intel and AMD. Reuters reported the win on Monday, October 19 — an early sign that Nvidia’s push into the CPU market those two rivals have long controlled is gaining traction.
Modal said it plans to start using Nvidia’s new Vera central processor for the first time, for hosting agents — the AI systems that carry out long-running tasks for human users. Until now, Modal has run such work only on Intel and AMD chips.
Vera At Modal: A First CPU Win Outside Intel And AMD
The workload is the story. Vera is not being bought to train models — Modal wants it for hosting AI agents, whose long-running, tool-calling sessions put unusual weight on the central processor that orchestrates them. That a specialist AI cloud, built entirely on Intel and AMD CPUs, is adding Nvidia’s processor for exactly this work says something about where agent infrastructure is heading.
Modal brings scale to the endorsement. The company last week raised $750 million in a funding round led by Accel at a $15.75 billion valuation, saying it is on track to record $1 billion in sales over the next year. Modal said usage of its services has tripled and revenue has grown more than 15 times year-over-year.
Adopting Vera is not a drop-in change: Modal said supporting Nvidia’s chip design for the first time will require deep software changes on its side — which makes the decision a considered bet rather than a trial of convenience.
The Test Result: Twice The Agent Work
Modal said Monday that early tests showed the new Nvidia chip doubled the amount of work that AI agents can get done, and that this result is what prompted it to support Nvidia’s chip design. The comparison was against the Intel and AMD chips already on the market that Modal runs today.
Chief Executive Erik Bernhardsson was careful to bound the claim. Both Nvidia rivals are developing newer models that could narrow the gap with Nvidia, he noted. “We’ll test them when they’re available,” he said. The doubling is a result on Modal’s own agent-hosting workload, not a general benchmark — but for a company whose product is agent compute, its own workload is the benchmark that decides purchasing.
A Mixed Fleet, Not A Switch
Bernhardsson stressed that Modal is not abandoning its existing suppliers. The company will run the Nvidia chips alongside its current Intel and AMD offerings in a mixed fleet, putting each job on whatever hardware runs it best. Running a mixed setup, he said, means the company “can always deploy whatever hardware performs best.”
That framing matters for how to read the win. Nvidia has not displaced Intel and AMD at Modal; it has earned a place in a fleet where every job is routed by performance. CPU share in AI clouds will be contested job by job, and Vera’s first foothold comes with a standing invitation for its rivals’ next chips to win the work back — the same agent-infrastructure race our guide to the agent platforms to know in 2026 tracks from the software side.
Nvidia’s $20 Billion Vera Ambition
Nvidia has told investors it expects to generate at least $20 billion from Vera chip sales this fiscal year — its opening bid for a CPU market it values at roughly $200 billion. Modal joins a named list: Nvidia has previously said OpenAI, Anthropic and Oracle also plan to use the chips, along with AI startups such as Perplexity.
The strategy extends the platform story in our explainer on why Nvidia powers the AI boom: having won the accelerator, Nvidia is moving into the processor that surrounds it, so an AI data centre can be Nvidia at every layer. Its newest open models, covered in our report on Nemotron 3.5 Super VL and Bolt, target the same always-on agent workloads Vera is now being bought to host.
Why It Matters
The central entity is Nvidia’s Vera CPU, and the attribute on display is traction: a first design win at a fast-growing AI cloud that had every reason to stay with its incumbents. Its neighbours are the story — Intel and AMD, whose decades-long hold on data centre CPUs now faces a challenger arriving from the AI side rather than the traditional server market; agent-hosting clouds like Modal, whose purchasing is being driven by agent throughput rather than generic compute; and Nvidia itself, for which Vera converts GPU dominance into a shot at the whole machine. A doubled agent workload in a customer’s own tests is the kind of evidence that travels between infrastructure teams faster than any launch keynote.
Background
Central processors were the one layer of the AI data centre Nvidia did not own: its GPUs dominated training and inference while Intel and AMD CPUs hosted the systems around them. Vera is Nvidia’s entry into that layer, aimed at the orchestration work AI agents generate, and Nvidia has priced its ambition openly — at least $20 billion in sales this fiscal year from a market it sizes at roughly $200 billion. Modal, founded as a developer-first cloud for running AI workloads, has grown into one of the largest independent agent hosts, which is why its fleet decision on October 19, 2026 carries weight beyond one contract.
Source: Reuters
0 Comments