Tech in Asia reported on 20 October 2026 that Nvidia won CPU business from Modal, a New York startup that lets AI developers rent server capacity. Modal plans to run its AI agents on Nvidia's Vera CPUs, after using only Intel and AMD chips. Modal said early tests showed twice the work on Vera compared with its current chips. Nvidia said OpenAI, Anthropic and Oracle also plan to use Vera.
Chief executive Erik Bernhardsson said Modal will keep a mixed fleet, since newer Intel and AMD chips could narrow the gap. The report gives no deal value, start date or chip count. It does not say whether Modal's customers can choose Vera, how "work" was measured, or how Vera is built beyond calling it a data center CPU. Timing for the other three firms is not given.
As general context, GPUs usually run the AI model, while CPUs handle the surrounding jobs, such as coordinating steps, calling tools and running code. So the CPU choice can affect cost and speed. Tech in Asia says Nvidia is pushing beyond GPUs, and Modal's mixed fleet suggests hedging. If Modal and others deploy Vera, buyers could gain a third option. No timing or customer availability is reported. Buyers can ask providers: which workloads, which baseline chips, what counts as work, and what fleet mix?