Nvidia CEO Jensen Huang said Monday that the company's next-generation chip is in “full production” and can deliver five times more artificial intelligence computing than the company's previous chips in delivering chatbots and other AI apps.
In a speech at the Consumer Electronics Show in Las Vegas, the leader of the world's most valuable company revealed new details about his company's chips. The chip is expected to arrive later this year, and as Nvidia faces increased competition from its own customers as well as competitors, Nvidia executives told Reuters it was already being tested by AI companies in its labs.
The Vera Rubin platform is made up of six separate Nvidia chips and is expected to debut later this year with flagship devices containing 72 of the company's flagship graphics units and 36 new central processors. Huang showed how they can be strung together into “pods” with more than 1,000 Rubin chips.
But to get the new performance results, the Rubin chip uses a unique type of data that the company wants the entire industry to adopt, Huang said.
“In this way, we were able to significantly improve performance despite having only 1.6 times as many transistors,” said Huang.
Nvidia still dominates the market for training AI models, but it faces much more competition in delivering the results of those models to hundreds of millions of users of chatbots and other technologies from traditional rivals like Advanced Micro Devices and customers like Alphabet Inc.'s Google.
Much of Huang's talk focused on how well the new chip will perform at that task, including adding a new layer of storage technology called “contextual memory storage” aimed at helping chatbots provide faster responses to long questions and conversations when used by millions of users at once.
Nvidia also touted a new generation of networking switches with a new type of connectivity called bundled optics. The technology is key to linking thousands of machines together and competes with products from Broadcom and Cisco Systems.
Among other announcements, Huang highlighted new software that can help self-driving cars decide which route to take and leave a paper trail that engineers can use later. Nvidia presented research on the software, called Alpamayo, late last year, and Huang said Monday it would make it more widely available, along with the data used to train the software, for automakers to evaluate.
“We not only open source our models, but we also open source the data we use to train those models, because only then can we truly trust how the models were created,” Huang said from a stage in Las Vegas.
Last month, Nvidia scooped up talent and chip technology from startup Groq, including executives who helped Alphabet's Google design its AI chips. Google is a major customer of Nvidia, but its own chips have emerged as one of Nvidia's biggest threats as Google works closely with Metaplatform and others to chip away at Nvidia's AI stronghold.
At the same time, NVIDIA wants to prove that its latest products can perform better than older chips such as the H200, which President Trump allowed to go to China. Reuters reports that the chip, a predecessor to Nvidia's current flagship Blackwell chip, is in high demand in China, alarming China hawks across the U.S. political spectrum.
