NVIDIA Vera adopted by Anthropic, OpenAI

Nvidia said on Monday that Anthropic, OpenAI and SpaceX are among the first major users of its upcoming Vera central processing unit, extending the company’s reach in AI data centres beyond graphics processors into the host-CPU layer. Chief Executive Jensen Huang named the companies during a presentation tied to Computex in Taiwan, according to Bloomberg and The Hindu BusinessLine. Nvidia has positioned Vera as the CPU inside its broader Vera Rubin platform and also as a standalone processor for AI factory workloads. The company says the chip is built for the CPU-side work around AI systems, including orchestration, data pipelines, analytics, tool use and reinforcement learning. ### Why is Nvidia talking about a CPU, not another GPU? Nvidia’s Vera product page says the chip is designed for “host CPU” tasks in accelerated systems and for standalone AI factory workloads. Those jobs include keeping GPUs supplied with data, handling scheduling and orchestration, and running code execution and sandboxing workloads around model inference and training, according to Nvidia. (bloomberg.com) The Rubin platform announcement in January described Vera as one of the core components in a tightly integrated system that also includes Rubin GPUs, NVLink, networking and other infrastructure. Nvidia said at the time that the platform’s design combined several chips and interconnect technologies to cut inference token costs and reduce the number of GPUs needed for some training workloads. (nvidia.com) ### Which companies did Huang name? Bloomberg reported that Huang said Anthropic, OpenAI and SpaceX would be among the first big users of Vera in their data centres. Nvidia’s own March Rubin platform release separately said Anthropic, Meta, Mistral AI and OpenAI were looking to use the Vera Rubin platform for larger model training and lower-latency, lower-cost serving. (investor.nvidia.com) Nvidia’s blog said on May 18 that Ian Buck, the company’s vice president of hyperscale and high-performance computing, hand-delivered the first Vera CPU systems to Anthropic, OpenAI, Oracle Cloud Infrastructure and SpaceXAI. That post described the deliveries as the move from announcement to production. ### What does Vera actually do inside an AI datacentre? (bloomberg.com) Nvidia says Vera is built for “the CPU work behind agentic AI and reinforcement learning,” including analytics, orchestration and data pipelines. The company says the CPU serves as the host processor for Vera Rubin NVL72, where it connects to Rubin GPUs through second-generation NVLink-C2C, and can also run as a standalone CPU system. (blogs.nvidia.com) Nvidia’s technical blog said the Rubin platform uses multiple chips as one system, with Vera handling CPU duties alongside Rubin GPUs and Nvidia networking components. That architecture gives Nvidia a role in both the accelerator layer and the general-purpose compute layer around AI workloads. That last point is an inference from Nvidia’s product description and platform design, not a direct company quote. (nvidia.com) ### What performance claims has Nvidia made? Nvidia-related coverage on June 1 said Vera is aimed at AI-specific workloads rather than conventional server tasks. Techmeme’s summary of Bloomberg’s report said Huang described Vera as 1.8 times faster at AI workloads than x86 chips, while NDTV Profit reported the same figure in coverage of Nvidia’s Computex announcements. (developer.nvidia.com) Nvidia’s investor release in January did not use that exact 1.8-times figure in the material surfaced here, but it said the broader Rubin platform could deliver up to a 10-fold reduction in inference token cost and cut the number of GPUs required to train mixture-of-experts models by a factor of four versus Blackwell. (techmeme.com) ### When will Vera be widely available? Bloomberg reported that Vera will go into full production in the third quarter of 2026. Nvidia’s May 18 blog post said the first systems had already been delivered to early customers including Anthropic, OpenAI, Oracle Cloud Infrastructure and SpaceXAI. (investor.nvidia.com) Nvidia’s next public milestones are tied to the broader Rubin rollout, which the company has already framed as its next-generation AI platform. The named early users now include Anthropic, OpenAI, SpaceX and Oracle Cloud Infrastructure, according to Nvidia and Bloomberg. (blogs.nvidia.com) (bloomberg.com)