Nvidia chief executive Jensen Huang unveiled the company’s newest AI server systems at the Consumer Electronics Show in Las Vegas on 6 January 2026, Vera Rubin, which was not due to go on sale until the second half of this year.
Nvidia typically saves the serious chip detail for its spring developer event in Silicon Valley, but Huang said AI’s computing load and demand for processors are forcing the industry to move faster.
Huang said: “The amount of computing necessary for AI is skyrocketing. The race is on for AI. Everyone is trying to get to the next frontier.”
He said multiple shifts in computing are driving the surge, with inference, the process by which models respond to prompts, now becoming more like a “thinking process”.
Nvidia has long pitched its next phase as the “omniverse”, using simulations of reality to train models for real-world tasks, including refining autonomous vehicle systems without endless road miles.
The new Vera Rubin servers, named after astronomer Vera Rubin, are built to handle the massive computational loads required to create those simulations and use them in training.
Nvidia tested Rubin on the assumption that developers will soon train models with up to 10 trillion parameters, then claimed that Rubin GPUs can train that in a month with one-quarter the chips needed on Blackwell.
For inference, Nvidia said Rubin delivers a 10-fold cost reduction versus Blackwell, while bundling connectivity and memory products to speed up the whole box.
Huang said: “Our job is to create the entire stack so you can create the applications that change the world.”
Nvidia pushed new programming libraries and software aimed at robotics, autonomous vehicles and what Huang called “physical AI”, where models have to deal with messy reality.
Futurum Group chief executive Daniel Newman called Rubin an “incredible generational leap” based on the specs and said the early reveal signals production is on track and the servers will land quickly.
“The pace of innovation continues to impress,” Newman said.
Nvidia shares were roughly flat on Monday, including after-hours, which is what happens when the market treats wild AI promises like background noise.
