Jensen Huang, co -founder and director of Nvidia Corp., during a press conference in Taipei, Taiwan, on Tuesday, June 4, 2024.
Annabelle Chih | Bloomberg | Getty Images
Nvidia Executive director Jensen Huang said that the next generation AI would need 100 times more than the older models as a result of new remedies approachs that think “about how best to answer” questions step by step.
“The amount of calculation needed for this explanation process is 100 times more than what we used to do,” Huang told CNBC Jon Forttt in an interview on Wednesday after Chip -chip Earnings in the Fourth quarter.
He cited the models, including Deepseek’s R1, Openai’s GPT-4 and Xai’s Grok 3 as models that use the explanation process.
Nvidia informed the results that are at the top of the analysts throughout the board, with revenue of 78% compared to the year earlier to $ 39.33 billion. The revenue of the Data Center, which includes graphic units for the NVIDIA market graphic processing, or GPUs, for the load of artificial intelligence, has increased by 93% to $ 35.6 billion and now makes over 90% of total revenue.
The company’s shares have not yet recovered after Losing 17% of its value January 27, his worst fall since 2020. This decline was due to the concern encouraged by the Chinese AI laboratory Deepseek that companies could potentially achieve greater performance in AI at far lower infrastructure costs.
On Wednesday, Huang pushed the idea on Wednesday, saying that Deepseek would popularize the resonance models that would need more chips.
“Deepseek was fantastic,” Huang said. “It was fantastic because it opened a model of explanation that is an absolute world class.”
Nvidia is limited from China’s business due to export control increased at the end of the Biden administration.
Huang said the percentage of the company’s revenue in China has declined by about half due to export limit, adding that there are other competitive pressures in the country, including Huawei.
The developers are likely to look for ways to export controls through software, whether it is a supercomputer, a personal computer, a phone or a playing console, Huang said.
“Ultimately, the software finds a way,” he said. “In the end, make that software on any system that you aim and create great software.”
Huang said that Nvidia GB200, which is sold in the United States, can generate AI content 60 times faster than the versions of chips selling to China under export controls.
Nvidia counts billions of dollars of infrastructure a year from the largest technological companies in the world for a large amount of revenue. The company was the largest user of AI boom, with the revenue greater than doubled in five flat districts to mid -2024. Before the growth slowed down slightly.
