Nvidia CEO Jensen Huang showing Vera Rubin at GTC.
Nvidia CEO Jensen Huang showing Vera Rubin at GTC.
Nvidia might be best known as a graphics card company, but its business focus is decidedly elsewhere. With most of its revenue coming from AI and data center sales, it's no wonder that we're already getting a look beyond existing Blackwell chips. At Nvidia's GTC conference this week, it showed off two new AI chip designs: Vera Rubin and Feynman. These AI-first processors are designed to be the future of AI and robotics in 2027 and 2028.
Vera Rubin is the first of the new chip designs and was originally announced at Computex 2024. It's slated to launch sometime in the second half of 2026, as per Ars Technica, and will reportedly support up to 75 terabytes of memory—fast memory, too. It's able to leverage the power of HBM4 to great effect, delivering an incredible 12TB per second of bandwidth.
Nvidia CEO Jensen Huang showing Vera Rubin at GTC.
Nvidia CEO Jensen Huang showing Vera Rubin at GTC.
Credit: Nvidia
The Vera Rubin die will have two GPUs sandwiched together, delivering over 50 petaflops of FP4 performance per chip. Nvidia claims that a full rack of these will deliver over 3.6 exaflops of performance—that's more than three times what's possible on existing Blackwell hardware.
Shipping alongside the Vera Rubin graphics hardware will be a companion CPU, itself called Vera. It will be made up of 88 custom ARM cores, delivering as many as 176 simultaneous threads. The CPU and GPU will be connected directly using Nvidia's high-speed NVLink interface, which Nvidia claims will offer up to 1.8 terabytes per second of bandwidth.
That will be followed by an advanced version of the Vera Rubin GPU in 2027, called Rubin Ultra. This will take advantage of a faster HBM4e standard, delivering even greater memory performance with support for up to 365TB. All in all, we're told it will deliver up to four times the performance of the original Vera Rubin stacks.
Nvidia CEO Jensen Huang showing the future Nvidia hardware at GTC.
Nvidia CEO Jensen Huang showing the future Nvidia hardware at GTC.
Credit: Nvidia
If that wasn't enough to get AI developers salivating, Nvidia also teased the next-generation chip designs beyond even Rubin Ultra's capabilities. It's known as "Feynman," named after American theoretical physicist Richard Feynman, and will appear sometime in 2028. Nvidia said it would use an advanced version of the Vera CPU introduced with the Vera Rubin-era graphics hardware.
Nvidia CEO Jensen Huang also spoke wistfully about the future, calling datacentres of the future "AI Factories," which would effectively produce the computational power required to run some of the most advanced AIs. Huang said these AIs would find their way into future assistive tools and humanoid robotics.