On Monday, ahead of the Consumer Electronics Show, Huang formally introduced Vera Rubin Architecture. This architecture is currently in production, with production expected to ramp up in the second half of this year. The move follows a year in which the company's Blackwell chips were a huge hit as demand for AI infrastructure continued to soar.
At a press conference prior to Huang's keynote, Dion Harris, Nvidia's senior director of HPC and AI infrastructure solutions, described Vera Rubin as “six chips that make up one AI supercomputer.”
“Vera Rubin is designed to address this fundamental challenge we have: the exponentially increasing amount of computation required for AI,” Huang told the audience during his CES presentation.
Huang added that compared to the Blackwell model, Rubin can dramatically improve performance by more than three times faster, run inference five times faster, and significantly increase the amount of inference computation per watt of energy.
Mr. Rubin was first announced in 2024 and has been slated to replace Mr. Blackwell ever since. This early debut comes several months earlier than NVIDIA's previously predicted late 2026 timeline.
Named after astronomer Vera Rubin, who discovered the existence of dark matter, NVIDIA said in a press release that the architecture is designed to support more complex agent-style AI workloads, as well as more networking and data movement.
The Rubin system is already ready for deployment across much of the cloud industry. Nvidia said its partners, including Amazon Web Services, OpenAI and Anthropic, all plan to use the new platform, along with Lawrence Berkeley National Laboratory's upcoming Doudna system.
The accelerated launch comes on the heels of Nvidia reporting record data center revenue, up 66% year-over-year, driven primarily by demand for Blackwell GPUs and Blackwell Ultra GPUs. These chips have become the benchmark for the current AI boom and are widely seen as a test of whether spending on AI infrastructure is sustainable.
Huang previously estimated that $3 trillion to $4 trillion could be spent on AI infrastructure worldwide over the next five years. Nvidia said products and services built on the Rubin platform will begin rolling out from partners in the second half of 2026.
