Nvidia CEO Jensen Huang says its next-generation AI superchip platform, Vera Rubin, will begin shipping to customers later this year on schedule. “Today I can say that Vera Rubin is in full production,” Huang said during a Monday press event at the annual CES technology trade show in Las Vegas.
Rubin will reduce the cost of running AI models to about one-tenth of Nvidia’s current leading chip system, Blackwell, the company told analysts and journalists on a Sunday earnings call. Nvidia also said Rubin can train some gigantic models using about a quarter of the number of chips Blackwell needs. Taken together, these benefits could make advanced AI systems much cheaper to run and harder for Nvidia’s customers to justify moving away from its hardware.
Nvidia said during the call that two of its current partners, Microsoft and CoreWeave, will be among the first companies to begin offering services based on Rubin chips later this year. Nvidia added that the two major AI data centers Microsoft is currently building in Georgia and Wisconsin will eventually house thousands of Rubin chips. The company said some Nvidia partners have started running their next-generation AI models on early Rubin systems.
The semiconductor giant also said it is working with Red Hat, which creates open-source enterprise software for banks, automakers, airlines and government agencies, to offer more products that will run on the current Rubin chip system.
Nvidia’s newest chip platform is named after Vera Rubin, an American astronomer who changed the way scientists understand the properties of galaxies. The system consists of six different chips, including a Rubin GPU and a Vera processor, both of which are built using Taiwan Semiconductor Manufacturing Company’s 3-nanometer manufacturing process and the most advanced wideband memory technology available. Nvidia’s sixth-generation interconnect and switching technologies connect different chips together.
Every part of this chip is “completely revolutionary and the best of its kind,” Huang announced during the company’s press conference at CES.
Nvidia has been developing Rubin for years, and Huang first announced the chips’ arrival during a keynote in 2024. Last year, the company said systems built on Rubin would start arriving in the second half of 2026.
It’s unclear what exactly Nvidia means when it says Vera Rubin is in “full production.” Typically, production of such advanced chips – which Nvidia builds with longtime partner TSMC – starts with miniature volumes while the chips undergo testing and verification, then increases at a later stage.
“This CES announcement for Rubin is intended to tell investors, ‘We’re on the right track,’” says Austin Lyons, an analyst at Imaginative Strategists and author of the semiconductor industry newsletter Chipstrat. There have been rumors on Wall Street that the Rubin GPU is lagging, Lyons says, so Nvidia is backtracking, saying it has completed key development and testing milestones and is confident Rubin is still on track to start ramping up production in the second half of 2026.
In 2024, Nvidia had to delay the shipment of its then-new Blackwell chips due to a design flaw that caused them to overheat when connected together in server racks. Deliveries to Blackwell returned on schedule in mid-2025.
As the AI industry continues to grow rapidly, software vendors and cloud service providers have had to compete fiercely for access to the latest Nvidia GPUs. Demand for Rubin is likely to be equally high. However, some companies are hedging their bets by investing in their own custom chip designs. For example, OpenAI said it is working with Broadcom to create custom silicon for the next generation of artificial intelligence models. These partnerships underscore the long-term risk for Nvidia: Customers designing their own chips could gain control over their hardware that the company doesn’t offer.
But Lyons says today’s announcements show how Nvidia is evolving beyond simply offering GPUs and becoming “a full-system AI architect, spanning compute, networking, memory hierarchy, storage and software orchestration.” He adds that even as hyperscalers pour money into custom silicon, Nvidia’s tightly integrated platform “is becoming increasingly difficult to displace.”