Nvidia (NVDA) gave the world a sneak peek at its next-generation Vera Rubin AI superchip during its GTC event in San Jose, Calif., on Tuesday. CEO Jensen Huang took the stage at the SAP Center to show off the processor, which will succeed the company’s new Blackwell Ultra chip when it begins shipping in the second half of 2026.
Vera Rubin, like the company’s Grace Blackwell superchip, combines a central processing unit and graphics processing unit. In the face of Grace Blackwell, Grace is the CPU and Blackwell is the GPU. For Vera Rubin, Vera is the CPU and Rubin is the GPU.
Nvidia says Vera will feature 4.2 times the memory found in Grace and 2.4 times the memory bandwidth. Combine that with Vera’s 88 CPU cores and the company says the chip will provide 2x the overall performance of its predecessor.
Rubin, for its part, will include 288GB of high-bandwidth memory 4 (HBM4) versus the HBM3e found in the Blackwell Ultra.
Nvidia didn’t stop there, though. The chip giant also debuted the generation of chips that will come after Vera Rubin called Vera Rubin Ultra. Set to come to market in the second half of 2027, Nvidia says the Vera Rubin Ultra will combine a Vera CPU with its Rubin Ultra chip. And where each Rubin processor combines 2 GPUs to form a single chip, Rubin Ultra combines 4 GPUs.
Like the Blackwell and Blackwell Ultra, Vera Rubin and Vera Rubin Ultra will be available in a large server rack system. The Vera Rubin server, called the NVL144, consists of a total of 144 GPUs, and provides 3.3 times the performance of the GB300 NVL72.
As of 3:51:12 PM EDT. Market Open.
Read more: How does Nvidia make money?
Then there’s the Vera Rubin Ultra NVL576. Packing 576 Vera Ultra GPUs, Nvidia says the system is 14 times faster than the Blackwell Ultra-powered GB300 NVL72. And with 365TB of memory, it’s an absolute monster of a server.
According to Nvidia, the Vera Rubin NVL576 will be housed in a new liquid-cooled server rack design called the Kyber Rack.
Nvidia’s Blackwell chip is still its most important offering, bringing in $11 billion of the company’s $39.3 billion in total revenue in Q4, but the company is clearly setting the stage for Vera Rubin to help ensure its customers and developers are prepared for its next-generation systems.
Vera Rubin and Vera Rubin Ultra will appeal to hyperscalers like Amazon (AMZN), Google (GOOG, GOOGL), Meta (META), and Microsoft (MSFT), as well as research facilities and governmental organizations working on AI services.
But Nvidia’s stock price has been stung by fears that hyperscalers are overspending on artificial intelligence without notching sufficient returns on their investments. President Trump’s threat to enact a 25% tariff on semiconductors produced overseas and the potential for further export controls haven’t helped either.
