Modal Becomes the First Major Cloud to Run AI Agents on Nvidia's Vera CPU, Reuters Reports

Nvidia has been chipping away at the server CPU market for years without a marquee win. According to a Reuters report published Monday, it now has one: Modal, the fast-growing AI infrastructure cloud, says it plans to run AI agents on Nvidia's new Vera central processing unit — the first major deployment of its kind.

Modal said early testing showed Vera roughly doubled the amount of work its agents could complete, a claim that comes from the company and has not been independently audited. The startup is not abandoning its incumbent suppliers: it describes a hybrid setup that assigns workloads to whichever chip handles them best, keeping existing Intel and AMD hardware in the mix.

The agent angle is not incidental. Nvidia's Vera CPU team recently told Wccftech that agentic AI workloads — many small, latency-sensitive decisions rather than a few big batch jobs — are forcing CPU design priorities back toward single-threaded performance after years of chasing core counts. Modal's agents, which orchestrate GPU jobs for developers, are exactly that kind of workload.

For Modal, the deal lands two weeks after a $750 million Series D round led by Accel at a $15.75 billion valuation. The company says usage has tripled since May, revenue is up more than 15-fold year over year, and annualized revenue has crossed $1 billion — figures it disclosed itself.

For Nvidia, the CPU business is becoming a second growth engine alongside GPUs. The company expects CPU-related sales of at least $20 billion this fiscal year, Reuters reported, in a market it estimates at roughly $200 billion — dominated for decades by Intel and AMD's x86 duopoly. OpenAI, Anthropic and Oracle are also planning to adopt Vera, according to the report.

The timing is pointed. Intel shares climbed 4.5% on Monday after J.P. Morgan upgraded the stock, citing rising agentic AI workloads as a source of positive earnings revisions — the same trend Nvidia is now selling against. Two chipmakers, one thesis, opposite beneficiaries.

The deeper shift is that the AI buildout is no longer only about accelerators. Agents that call tools, parse documents and coordinate with each other spend real time on general-purpose computation, and whoever owns that layer owns a slice of every AI workload. If Vera's early numbers hold up at production scale, the x86 incumbents will face the first credible agent-shaped wedge into their stronghold — and Nvidia will have turned a side product into a $20 billion business line.