Nvidia has won CPU business from Modal, a fast-growing AI cloud startup that until now ran only Intel and AMD processors, Reuters reported Monday, Oct. 19. Modal will use Nvidia's new Arm-based Vera CPU to host AI agents after its tests showed up to double the task throughput.
Why it matters: AI agents, the kind of software that runs long tasks on its own, need lots of ordinary CPU work, not just GPUs. Nvidia has told investors it expects at least $20 billion in Vera sales this fiscal year, in a market it values at roughly $200 billion.
What Modal found
- Modal built an "Agentless Agent Bench" of 50 Terminal-Bench tasks: compiling, file searching, git operations and more.
- On a quiet machine, Vera was often faster, "but not always, and not by a particularly large margin," Modal engineer Erik Dunteman wrote.
- Under heavy concurrent load, Vera delivered up to 2x the throughput of an equivalent x86 Granite Rapids machine, with per-task latency equal or lower.
These are Modal's own early tests, not an independent benchmark. CEO Erik Bernhardsson told Reuters Intel and AMD have newer chips coming: "We'll test them when they're available."
The company
Modal raised $750 million last week in a round led by Accel at a $15.75 billion valuation, and says it is on track for $1 billion in sales over the next year. Nvidia has said OpenAI, Anthropic and Oracle also plan to use Vera.
More Nvidia: Nvidia's German physical-AI partnership and GTC Berlin on Oct. 21.