Nvidia CEO Jensen Huang unveiled the company’s new AI supercomputer, Vera Rubin, designed to tackle one of the biggest challenges with Artificial Intelligence: the skyrocketing demand for computing power.
As AI models grow in size and complexity, the need for powerful hardware has surged. According to Huang, the largest AI models are increasingly driving an unprecedented demand for GPUs and infrastructure.
On average, Huang said, AI models are increasing in size by a factor of ten each year, emphasizing the need for systems that can keep pace with rapidly evolving workloads. Vera Rubin is Nvidia’s answer: a rack-scale supercomputer built from six completely new chips, each engineered to work together as one cohesive system. The platform is named in honor of Vera Rubin, the pioneering American astronomer who provided compelling evidence for dark matter in the late 1970s.
To design the new supercomputer, Huang says Nvidia had to break one of its internal rules.
“We have a rule inside our company and it’s a good rule. No new generation should have more than one or two chips changed.” Nvidia’s Vera Rubin kept none of the chips the same, coming out with six new chips.
Huang said Nvidia had no choice but to break the rule to keep up with Moore’s Law. Gordon Moore, co-founder of Intel, developed Moore’s Law, which theorized that transistor counts double every two years. Transistors are the processing tools for chips and originally were millimeters in size, but today, they are atomic-sized.
Given shrinking the size of the now almost atomic transistors would be extremely challenging, Nvidia co-designed six chips to work together, achieving higher performance without significantly increasing transistor count.
Huang says “each one of them [the chips] are revolutionary and the best of its kind.”
Nvidia claims that Vera Rubin delivers twice the computational power of Blackwell, its previous-generation system, while using less energy. The combination of liquid cooling, optimized chip design, and high-speed interconnects contributes to this efficiency. Beyond raw performance, the system also introduces confidential computing and is fully encrypted. This allows companies to deploy AI models in shared or public environments while ensuring that their data and models remain secure and inaccessible to others.
The completed system weighs nearly two tons and houses 1,152 GPUs, all cooled 100% by liquid, compared with Nvidia’s previous generation Blackwell system, which is 80% liquid cooled and 20% air-cooled. Liquid cooling is more efficient than air, as it absorbs heat more effectively, allowing the system to maintain peak performance with less energy.
Vera Rubin is already in full production and expected to be available for purchase in the second half of 2026. Notable leaders in tech have started praising Nvidia’s latest invention.
Elon Musk said the “NVIDIA Rubin will be a rocket engine for AI.” Mark Zuckerberg said the supercomputer would help create the next leap in artificial intelligence.
“NVIDIA’s Rubin platform promises to deliver the step-change in performance and efficiency required to deploy the most advanced models to billions of people.”
