Jensen Huang Says Nvidia’s New Vera Rubin Chips Are in ‘Full Production’

Nvidia CEO Jensen Huang says that the company’s next-generation AI superchip platform, Vera Rubin, is connected docket to statesman arriving to customers aboriginal this year. “Today, I tin archer you that Vera Rubin is successful afloat production,” Huang said during a property lawsuit connected Monday astatine the yearly CES exertion commercialized amusement successful Las Vegas.

Rubin volition chopped the outgo of moving AI models to astir one-tenth of Nvidia’s existent starring spot system, Blackwell, the institution told analysts and journalists during a telephone connected Sunday. Nvidia besides said Rubin tin bid definite ample models utilizing astir one-fourth arsenic galore chips arsenic Blackwell requires. Taken together, those gains could marque precocious AI systems importantly cheaper to run and marque it harder for Nvidia’s customers to warrant moving distant from its hardware.

Nvidia said connected the telephone that 2 of its existing partners, Microsoft and CoreWeave, volition beryllium among the archetypal companies to statesman offering services powered by Rubin chips aboriginal this year. Two large AI information centers that Microsoft is presently gathering successful Georgia and Wisconsin volition yet see thousands of Rubin chips, Nvidia added. Some of Nvidia’s partners person started moving their next-generation AI models connected aboriginal Rubin systems, the institution said.

The semiconductor elephantine besides said it’s moving with Red Hat, which makes unfastened root endeavor bundle for banks, automakers, airlines, and authorities agencies, to connection much products that volition tally connected the caller Rubin spot system.

Nvidia’s latest spot level is named aft Vera Rubin, an American astronomer who reshaped however scientists recognize the properties of galaxies. The strategy includes six antithetic chips, including the Rubin GPU and a Vera CPU, some of which are built utilizing Taiwan Semiconductor Manufacturing Company’s 3-nanometer fabrication process and the astir precocious bandwidth representation exertion available. Nvidia’s sixth-generation interconnect and switching technologies nexus the assorted chips together.

Each portion of this spot strategy is “completely revolutionary and the champion of its kind,” Huang proclaimed during the company’s CES property conference.

Nvidia has been processing the Rubin strategy for years, and Huang archetypal announced the chips were coming during a keynote code successful 2024. Last year, the institution said that systems built connected Rubin would statesman arriving successful the 2nd fractional of 2026.

It’s unclear precisely what Nvidia means by saying that Vera Rubin is successful “full production.” Typically, accumulation for chips this advanced—which Nvidia is gathering with its longtime spouse TSMC—starts astatine debased measurement portion the chips spell done investigating and validation and ramps up astatine a aboriginal stage.

“This CES announcement astir Rubin is to archer investors, ‘We’re connected track,’” says Austin Lyons, an expert astatine Creative Strategists and writer of the semiconductor manufacture newsletter Chipstrat. There were rumors connected Wall Street that the Rubin GPU was moving down schedule, Lyons says, truthful Nvidia is present pushing backmost by saying it has cleared cardinal improvement and investigating steps, and it’s assured Rubin is inactive connected people to statesman scaling up accumulation successful the 2nd fractional of 2026.

In 2024, Nvidia had to hold transportation of its then-new Blackwell chips owed to a plan flaw that caused them to overheat erstwhile they were connected unneurotic successful server racks. Shipments for Blackwell were backmost connected docket by the mediate of 2025.

As the AI manufacture rapidly expands, bundle companies and unreality work providers person had to fiercely vie for entree to Nvidia’s newest GPUs. Demand volition apt beryllium conscionable arsenic precocious for Rubin. But immoderate firms are besides hedging their bets by investing successful their ain customized spot designs. OpenAI, for example, has said it is moving with Broadcom to physique bespoke silicon for its adjacent procreation of AI models. These partnerships item a longer-term hazard for Nvidia: Customers that plan their ain chips tin summation a level of power implicit their hardware that the institution doesn’t offer.

But Lyons says today’s announcements show however Nvidia is evolving beyond simply offering GPUs to becoming a “full AI strategy architect, spanning compute, networking, representation hierarchy, storage, and bundle orchestration.” Even arsenic hyperscalers determination wealth into customized silicon, helium adds, Nvidia’s tightly integrated level “is getting harder to displace.”