AI cloud company Modal is preparing to add Nvidia’s new Vera central processor to its infrastructure after early tests showed major performance gains for the isolated computing environments used to run AI agents.
The move is significant because Modal has historically operated on x86 processors from Intel and AMD, while Vera uses the Arm architecture. Supporting it therefore requires changes deeper in the company’s infrastructure than simply installing another accelerator.
Modal said its initial benchmarks showed the Nvidia Vera CPU delivered roughly twice the throughput for Modal Sandboxes, the isolated environments developers use to execute code and agent workloads.
AI agents are making CPUs strategically important again
The AI infrastructure conversation has largely centered on GPUs because those processors handle model training and inference.
Agents create another workload.
An AI model may generate instructions, but the resulting agent also needs somewhere to execute code, install packages, manipulate files or run longer processes. Those activities can depend heavily on conventional CPU infrastructure.
Modal’s engineering team said moving to Vera would require changes across its infrastructure and image-build system because of the shift from x86 to Arm.
Reuters reported that Modal plans to incorporate Vera specifically to help host long-running AI-agent workloads, after its testing showed a roughly twofold performance improvement compared with existing systems.
CEO Erik Bernhardsson stressed that Modal is not abandoning Intel or AMD and intends to maintain a mixed hardware strategy based on performance.
Modal has become a major AI infrastructure company
The hardware decision comes during extraordinary growth.
Data Center Dynamics reported that Modal recently raised a $750 million Series D at a $15.75 billion post-money valuation, led by Accel with participation from Kleiner Perkins and existing investors.
The company also says its annualized revenue run rate has passed $1 billion, while usage has tripled since its previous financing round.
Modal was founded by Erik Bernhardsson and Akshat Bubna and provides serverless infrastructure for model inference, training and software executed by AI agents.
Its Sandboxes product has become particularly important as developers need controlled environments where autonomous systems can safely run the code they generate.
The AI infrastructure stack is becoming broader
Nvidia’s Vera push also represents a strategic expansion for the chipmaker.
The company dominates AI accelerators but has historically had far less presence in general-purpose data-center CPUs, where Intel and AMD remain established suppliers.
Reuters reported that Nvidia expects at least $20 billion in Vera CPU sales during the current fiscal year and sees a potential $200 billion market for the technology.
Modal offers an early example of where that demand could come from.
As AI agents become longer-running and more autonomous, infrastructure providers need more than powerful GPUs. They need fast CPUs, isolated containers, storage, scheduling and networking capable of supporting everything the model does after it produces an answer.
For developers, the competitive frontier in AI infrastructure is therefore shifting.
The GPU still runs the model. But increasingly, the infrastructure surrounding the model determines how useful an AI agent can actually become.
