Why DeepSeek Models Need “100x More Computing”

CEO Jensen Huang discusses Influence Chinese startup DeepSeek’s new AI model in an interview with CNBC on Wednesday Jim Cramer exist NvidiaAnnual GTC Conference. Huang asserts that groundbreaking models actually require More calculations More than many in the industry think.

According to Huang, DeepSeek’s R1 model is “awesome” because it is “the first open source reasoning model”. He explained that the model gradually solves the problem, is able to come up with different answers, and can verify that its answer is correct.

“This inference AI has 100 times more computational power than non-conditioning AI,” he said. “The situation is exactly the opposite, and it is the exact opposite conclusion that everyone has come to.”

In late January, DeepSeek’s model triggered a massive sell-off in AI stocks as investors fear the model will perform well, while using less energy and money can also show the highest competitors. Nvidia plunges In one meeting, 17% of the losses were close to $600 billion. The largest ever Single-day drops for US companies.

Huang also talked about some of his company reveal So far, in conferences, new AI infrastructure for robotics and enterprises have been included Dell,,,,, HPE,,,,, Accenture,,,,, ServiceNow and Crowdstrike. He reflects the AI ​​boom extensively and points to the way hype has changed from purely generative AI to inference models. He also predicts that by the end of the decade, the world’s computing capital expenditure seems to be expected to reach a trillion dollars and that most of the money will go to AI.

“So, by the end of this decade, our chances are a million dollar percentage, which is huge,” Huang said. “We have a lot of infrastructure to build.”

Register now Let CNBC Investment Club follow Jim Cramer’s every move in the market.

Disclaimer CNBC Investment Club Charitable Trust holds shares in Nvidia and CrowdStrike.