Modal Receives First Nvidia Vera CPU Server, Moving Beyond Intel and AMD

Cloud-computing startup Modal said it has received its first server powered by Nvidia's Vera central processing unit, marking a shift from its previously exclusive use of Intel and AMD processors, according to Seeking Alpha Market News. Modal rents computing power to AI developers, and the arrival of the Vera-based machine signals that Nvidia's push into CPUs is reaching beyond the largest AI labs and into smaller specialized cloud providers.

What Modal received

Seeking Alpha's reporting is brief on specifics: Modal said it received its first Vera-powered server, without Modal or Nvidia detailing the exact rack configuration, unit count, or production timeline for Modal's deployment. That leaves open whether Modal's unit is a standalone Vera server or part of a larger rack-scale system, and when the company expects to put the hardware into live production for customers.

What is clear from other Nvidia disclosures is that Vera can be deployed two ways. According to ServeTheHome and SemiWiki, Vera is sold both as a standalone CPU server or rack and as the host processor inside Nvidia's integrated Vera Rubin systems, with Nvidia and its partners offering both rack-scale systems and individual servers. In the integrated configuration, the Vera Rubin NVL72 platform pairs 72 Rubin GPUs with 36 Vera CPUs, plus ConnectX-9 SuperNICs and BlueField-4 DPUs, in a single rack-scale platform, according to shattered.io. That is twice as many GPUs as CPUs in the rack, a 36-unit gap, by our calculation (72 minus 36 equals 36, a 100% relative difference). Our interpretation of that ratio: Nvidia weights its flagship rack toward GPU throughput while still dedicating substantial CPU capacity to supporting work.

Why Nvidia built a CPU for AI agents

Nvidia has framed Vera as purpose-built for the CPU-bound tasks that surround large-scale AI agent deployments rather than as a general-purpose chip. Shattered.io reported that Vera is built around 88 custom Olympus cores and that Nvidia describes it as the first processor designed specifically for AI agent workloads, including code sandboxes, reinforcement-learning environments, and data pipelines that support GPU-based training and inference. SemiWiki's reporting put it more bluntly, framing agentic sandboxes, tool calls, orchestration layers and long-context retrieval operations as fundamentally CPU work that Nvidia wants Vera to absorb.

Nvidia CEO Jensen Huang introduced the standalone Vera CPU at GTC San Jose in March, describing it as the company's next multi-billion-dollar business, according to SemiWiki's account. That framing positions Vera as a strategic expansion beyond Nvidia's GPU franchise rather than a minor accessory product.

Modal joins a short but growing list

Modal is not Nvidia's first publicized Vera recipient. SemiWiki reported that the first Vera CPUs arrived at Anthropic in San Francisco, OpenAI in Mission Bay and SpaceXAI in Palo Alto on the same day, followed by a delivery to Oracle Cloud Infrastructure in Santa Clara. Separately, a Facebook post attributed to Nvidia described the company's Ian Buck hand-delivering AWS's first Vera CPU server and Vera Rubin GPU to AWS executives in Seattle.

Of the disclosed customers, Oracle Cloud Infrastructure has made the largest stated volume commitment. SemiWiki quoted an OCI executive, identified as Batta, saying the company plans to deploy hundreds of thousands of Nvidia Vera CPUs beginning in 2026 because agentic AI demands sustained performance at massive scale. At CoreWeave, Vera is described as an addition rather than a wholesale replacement: shattered.io reported that Vera joins CoreWeave's existing AMD EPYC and Intel Xeon CPU options rather than replacing them, giving customers a specialized choice for agent-heavy, GPU-coupled workloads alongside general-purpose compute. CoreWeave's rack-scale Vera deployment is reported to fit 128 CPUs and 11,264 cores into a single rack, which the company says supports more than 11,000 concurrent agent environments, according to shattered.io's reporting, based on an assumption of roughly one core per environment.

What remains unknown

Several questions about Vera's broader rollout remain unanswered in available reporting. Shattered.io reported that CoreWeave's own blog post describes its Vera CPU service as "coming soon," without naming a launch date or pricing, citing TheEnergyMag — meaning neither Nvidia nor CoreWeave has disclosed Vera pricing or a firm standalone-CPU availability date. Nvidia has also reportedly pitched Vera CPUs to customers in China, according to Reuters as summarized by Schwab Network. Bottom line: Modal's single server is a small data point, but taken with the disclosed deliveries to Anthropic, OpenAI, SpaceXAI, Oracle Cloud Infrastructure and AWS, it suggests Nvidia's CPU line is spreading from flagship AI labs to smaller specialized cloud providers — while pricing, deployment timelines and real-world performance remain undisclosed.