Modal Adds Nvidia Vera CPU After Tests Show Doubled Performance for AI Agents

Modal Breaks from x86 Chips with Nvidia Vera CPU Adoption

Modal, a rapidly growing AI cloud computing startup valued at $15.75 billion, announced it will deploy Nvidia Vera CPU for hosting AI agents, marking its first move away from x86 chips made exclusively by Intel and AMD1

1

. The decision follows internal tests demonstrating that Nvidia's Arm-based Vera central processor doubled the work capacity that AI agents can handle compared to existing Intel and AMD chips already on the market2

2

.

The San Francisco-based company, which rents computing power to AI developers, raised $750 million last week in a funding round led by Accel and projects $1 billion in sales over the next year3

3

. Modal reported that usage of its cloud-computing services has tripled and revenue has grown more than 15 times year-over-year, driven largely by demand for its Sandboxes product, which generates $300 million annually2

2

.

Performance Gains Drive Strategic Shift for AI Agents

Modal's tests revealed significant performance advantages when running AI agents on Nvidia Vera CPU. The AI startup built a benchmark of 50 coding tasks, including compiling software and searching files, to evaluate the chip's capabilities2

2

. While Vera showed modest improvements when running tasks individually, the performance gap widened dramatically when Modal packed servers with containerized workloads.

According to Modal, Vera maintained consistent task completion speeds even under heavy loads, while rival chips hit performance ceilings2

2

. The company credits Vera's high memory bandwidth for this advantage, which proves particularly valuable for background coding agents and reinforcement learning workloads where expensive GPUs often sit idle waiting for CPU work to finish.

Mixed Fleet Strategy Maintains Flexibility

Chief Executive Erik Bernhardsson emphasized that Modal is not abandoning its existing suppliers despite adopting Nvidia central processors4

4

. The AI startup will run Nvidia chips alongside its current Intel and AMD offerings in a mixed fleet, assigning each job to whatever hardware performs best for that specific workload1

1

.

Bernhardsson noted that both Intel and AMD are developing newer models that could narrow the performance gap with Nvidia. "We'll test them when they're available," he said, adding that running a mixed setup means the company "can always deploy whatever hardware performs best"5

5

. The integration will require substantial software modifications on Modal's platform to support the Arm-based architecture4

4

.

Nvidia's Push into $200 Billion CPU Market Gains Traction

Modal's adoption represents another validation point for Nvidia's expansion beyond GPUs into the central processor market long dominated by Intel and AMD. Nvidia unveiled Vera in March as a CPU specifically built for hosting AI agents, featuring 88 cores optimized for these workloads2

2

. The company projects at least $20 billion in Vera chip sales this fiscal year as it targets a market it values at roughly $200 billion1

1

.

Nvidia has already secured commitments from major AI players including OpenAI, Anthropic, and Oracle to use Vera chips3

3

. Perplexity, another AI startup, previously reported 1.5 times faster performance for coding tasks on the chip2

2

. Nvidia hand-delivered Modal's first Vera server to its San Francisco office last week, though the company has not announced when Vera-powered AI sandboxes will reach customers2

2

.

The shift matters because AI agents require different computing characteristics than traditional workloads. Modal's Sandboxes provide isolated environments where AI agents run code they write, creating unpredictable demand patterns. An agent may sit idle while waiting for an AI model, then suddenly require ten times more computing power during a software install, with hundreds of these running simultaneously on each server2

2

. This volatility makes processor efficiency under variable loads increasingly important as AI developers scale agent deployments across industries.