NVIDIA's Vera CPU in evaluation

Nvidia has started putting a new signal into the market: Vera is no longer just a roadmap slide. The company said this week that first Vera CPU systems have gone out to early customers, with SpaceXAI, Oracle Cloud Infrastructure, Anthropic and OpenAI named among the organizations evaluating the platform. That matters because Vera is Nvidia’s attempt to extend its control of the AI stack beyond GPUs and into the CPU layer that helps run AI factories. Nvidia says Vera serves as the host CPU in its Vera Rubin systems and is built to handle orchestration, ETL and KV-cache management around large-scale AI workloads. The important word, though, is evaluating. Yahoo Finance reported that the named customers are evaluating Vera, not that all of them have committed to broad production rollouts. (finance.yahoo.com) If you’re trying to read demand from this, separate logo signal from deployment signal. A named customer tells you Nvidia has access and relevance. It does not tell you the size, timing or certainty of future revenue. (nvidia.com) In infrastructure sales, especially for new compute platforms, the useful progression is usually: sample or system delivery, technical evaluation, workload fit, deployment plan, then scaled purchase. The reporting around Vera supports the first two steps, but not a blanket conclusion that every evaluator is already a committed rollout. (finance.yahoo.com) Oracle is the clearest exception in the current reporting. A separate Yahoo Finance item said Oracle Cloud Infrastructure was described by Nvidia as the first cloud provider to deploy Vera at hyperscale and plans to roll out hundreds of thousands of CPUs beginning in 2026. That is a stronger commercialization signal than a generic evaluation mention because it points to a defined deployment path and timing. (finance.yahoo.com) Anthropic is also worth separating from the pack. In CNBC excerpts from Jensen Huang’s interview with Sara Eisen, Huang said Nvidia is “scaling very quickly” with Anthropic and has “big plans” there. That suggests a deeper relationship, but even there, those comments are broader than a disclosed Vera purchase commitment on their own. (finance.yahoo.com) The strategic backdrop is Nvidia’s Rubin push. Nvidia says the Vera Rubin NVL72 combines 72 Rubin GPUs with 36 Vera CPUs in a rack-scale system aimed at reasoning and agentic AI workloads. On Nvidia’s product pages, Vera is positioned as the CPU backbone for the tasks that keep an AI factory running while GPUs do the heavy model work. For anyone tracking this as a business-development or forecasting story, the clean way to score it is by milestones, not by logos. (cnbc.com) A named evaluator should count as strategic validation. A stated hyperscale rollout plan, like Oracle’s, should count as a materially stronger indicator. Public comments about broader account expansion, like Huang’s remarks on Anthropic, sit somewhere in between unless they are tied to disclosed deployment milestones. That last point is an inference from the reported facts, not a new company disclosure. (nvidia.com) The next useful evidence will be concrete: customer deployment dates, workload-specific announcements, volume commitments, or additional cloud partners saying Vera is moving from evaluation into production. Nvidia’s latest public materials place Vera inside the broader Vera Rubin rollout now in production, with more customer detail likely to surface through investor updates and product announcements during 2026. (investor.nvidia.com) (cnbc.com)