Nvidia CEO Jensen Huang surprised the audience at his keynote on Sunday for the Computex event by teasing the next chip architecture from his company, “Rubin.”
Nvidia
Nvidia CEO Jensen Huang disclosed in his keynote Sunday evening, at the Computex annual trade show in, Taipei, Taiwan, that the successor to the company’s “Blackwell” GPU and CPU family will be called “Rubin” named for US astronomer Vera Rubin, and will be made available in the first half of 2026.
The announcement reinforces Huang’s pledge for Nvidia to be on a one-year cadence of releasing new chip architectures.
Before clicking the slide to show the Rubin line, Huang quipped to his audience, “This is the very first time this next click has been made, and I’m not sure yet whether I’m going to regret this or not.”
Huang shows off the roadmap from prior generation, “Hopper,” through the newest, “Blackwell,” to Rubin.
Nvidia
“All of these chips are in full development, a 100% of them, and the rhythm is one year, at the limits of technology, all 100% architecturally compatible,” said Huang, “and all of the richness of software on top of it.”
Also: Nvidia CEO Jensen Huang unveils next-gen ‘Blackwell’ chip family at GTC
The surprise unveiling of Rubin comes hot on the heels of the formal announcement of Blackwell by Huang at the GTC conference in March.
According to a research note from Wall Street brokerage Jefferies & Co., citing news site Wccftech, the Rubin chips, including a GPU chip family, and a CPU chip named Vera, will move from using the current fastest computer DRAM memory, “HBM3e,” to using the next generation, “HBM 4.” The GPU chips will be designated as “R100” GPUs.
Nvidia
The chips will be fabricated by Taiwan Semiconductor Manufacturing in a 3-nanometer process technology, said Jefferies analysts, and will commence production in the fourth quarter of 2025.
Also: AI startup Cerebras unveils the WSE-3, the largest chip yet for generative AI
The Rubin uses what’s called a “4x reticle design,” the analysts note, which means that it takes up more of the surface area of a 12-inch silicon wafer than does the Blackwell, which is only 3.3x. The “reticle” is the area of a wafer that a typical photolithography system can address with one pass when laying down the circuits of a chip.
Huang also disclosed a forthcoming successor to the Blackwell GPU, Blackwell Ultra. The Rubin architecture, said Huang, will also have its “Ultra” version, following in 2027.
Check out all the Computex news from Nvidia on the company’s news site or this round-up blog post on the event from the company.