Nvidia’s Vera Rubin ships now as OpenAI readies Q3 rollout

Nvidia says its Vera Rubin platform is now in full production, with systems shipping to customers including OpenAI, CoreWeave, Google Cloud, Microsoft Azure, Meta, and Dell. At a briefing at Nvidia headquarters on Monday, Ian Buck, the company’s vice president of accelerated computing, said the hardware had moved into customer deployments.

According to Bloomberg, which first reported the briefing, OpenAI plans to adopt Vera Rubin at scale in the third quarter. CoreWeave, one of the first cloud providers to get the systems, told Bloomberg that its NVL72 racks are producing 10 times the token output of the previous generation.

The NVL72 is a full-rack system that combines 72 Rubin GPUs with Vera CPUs and uses liquid cooling to remove internal cabling. Nvidia demonstrated that design at the event. Buck said the cooling system lets Nvidia eliminate copper connections that had previously limited how densely components could be packed.

Recommended reading

Ural plant patents turbine pin that won’t jam in repairs

Nvidia also used the briefing to spotlight its new Vera CPU, which replaces processors it had previously sourced from other companies. The company drew a direct comparison with AMD, claiming Vera is nearly twice as fast as AMD’s Turin chip on Python workloads — a benchmark Nvidia chose because Python underpins much of today’s AI inference software stack. Anthropic, OpenAI, Perplexity, SpaceX, and Oracle are among the first groups receiving the processor.

The production milestone arrives during a weaker stretch for Nvidia’s stock. Its shares are up 9 percent this year, while the broader chip index has climbed 66 percent over the same period, according to Bloomberg. Intel, ARM, and AMD have each more than doubled.

Analysts expect Nvidia’s revenue to rise 82 percent to about $393 billion for the fiscal year. But the market response has been more muted than during the Blackwell cycle. Nvidia began building the Vera Rubin narrative over the past two months: Jensen Huang said at Computex in early June that the platform was in full production and named Anthropic, OpenAI, SpaceX, and Oracle as early recipients.

Monday’s update added concrete performance claims and customer references that were missing from that keynote. Still, the figures came from Nvidia and its customers, not independent testing. That includes CoreWeave’s 10x token-output claim and Nvidia’s Python benchmark against AMD, with broader volume shipments and third-party verification still to come.