Nvidia has offered a view of its latest product, the Blackwell Ultra and subsequent generations of AI chips at its 2025 GTC AI conference.
Nvidia CEO Jensen Huang said in the conference—held between March 17-21 at San Jose, California—that Blackwell Ultra GB300 family of semiconductors will have 1.5 times the memory and deliver significantly higher performance than the current Blackwell line. Nvidia expects Blackwell Ultra to launch in the second half of 2025.
According to Huang, this sets the stage for Nvidia’s next-generation CPU-GPU platform, known as Vera Rubin, scheduled for the second half of 2026, and Rubin Ultra GPUs a year later.
Blackwell has 68 times the computing power of its Hopper line, he said, while Rubin represents a 900-fold increase. Huang also mentioned that Vera Rubin is named after the scientist who discovered dark matter, and a later line will be named for physicist Richard Feynman.
READ: Nvidia rises to second most valuable company after Q4 earnings surge (February 27, 2025)
Huang on Tuesday pointed to surging demand for Nvidia GPUs. The company shipped 3.6 million Blackwell GPUs to America’s four largest cloud service providers—Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), and Meta (META)—in its 2025 fiscal year, which ended in late January. That’s up from 1.3 million Hopper GPUs from the year before.
Nvidia also revealed a number of upcoming partnerships, including with General Motors, to train AI manufacturing models. The goal is to build the automaker’s “future self-driving car fleet,” Huang said. “The time for autonomous vehicles has arrived,” he added.
Another partnership that has got a lot of attention is with Yum Brands, which is partnering with Nvidia to use AI for restaurant service. The first AI-powered voice-ordering at the drive-through lane and on the phone was built by Yum’s developers using tools and development frameworks from Nvidia, and will begin rolling out at 500 restaurants across the portfolio in the second quarter of this year.