Nvidia CEO Jensen Huang said Monday that the company’s next-generation chips are in “full production” and that they can provide five times more artificial-intelligent computing than the company’s previous chips when serving 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 its chips, which will arrive later this year and are being tested by AI firms in the company’s labs, Nvidia executives said, as Nvidia faces increasing competition from rivals as well as its own customers.
The Vera Rubin platform, made up of six different Nvidia chips, is expected to launch later this year, with flagship servers featuring the company’s 72 graphics units and 36 new central processors.
Huang showed how they could be linked together into “pods” with more than 1,000 Rubin chips and said they could improve the efficiency of generating what are known as “tokens” – the fundamental unit of AI systems – by 10 times.
However, to achieve new performance results, Huang said Rubin chips use a proprietary type of data that the company hopes the broader industry will adopt.
“That’s how we were able to make such a huge step up in performance, even though we only had 1.6 times the number of transistors,” Huang said.
While Nvidia still dominates the market for training AI models, it faces far more competition from traditional rivals like Advanced Micro Devices as well as customers like Alphabet’s Google in delivering the fruits of those models to millions of users of chatbots and other technologies.
Much of Huang’s speech focused on how well the new chips will work for that task, including adding a new layer of storage technology called “context memory storage,” which is aimed at helping chatbots provide quicker responses to longer questions and conversations.
Nvidia also announced a new generation of networking switches with a new type of connection called co-packaged optics. This technology, which is key to linking thousands of machines together, competes with offerings from Broadcom and Cisco Systems.
In other announcements, Huang highlighted new software that could help self-driving cars decide which path to take — and leave a paper trail for engineers to use later. Nvidia showed research late last year about the software, called AlpaMyo, which Huang said Monday would be released more widely, along with the data used to train it so automakers can evaluate.
“Not only do we open-source the models, but we also open-source the data that we use to train those models, because only that way can you really trust how the models were built,” Huang said from a forum in Las Vegas. Last month, Nvidia acquired talent and chip technology from startup Grok, which also included executives who played a key role in helping Alphabet’s Google design its own AI chips.
While Google is a major Nvidia customer, its own chips have emerged as one of the biggest threats to Nvidia as Google teams up with Meta Platform and others to chip away at the company’s lead.
During a question-and-answer session with financial analysts after his speech, Huang said the Groke deal “will not impact our core business” but could result in new products that would expand its lineup. At the same time, Nvidia is eager to show that its latest products can outperform older chips like the H200, which Donald Trump has allowed to flow into China.
The chip, which was the predecessor to Nvidia’s current “Blackwell” chip, is in high demand in China, which has alarmed people across the US political spectrum.
