Why DeepSeek model needs ‘100 times more computing’

CEO Jensen Huang discussed influence A new artificial intelligence model in the interview of DeepSeek of China’s CNBC on Wednesday Jim Cramer horse NvidiaAnnual GTC Conference. Huang claimed that the groundbreaking model would actually be required More calculation Many of the industry thought.

According to Huang, Deepseek’s R1 model is “fantastic” because “the first open picturesque justification model”. He explained that the model violates step by step problems, you can get acquainted with different answers and check if the answer is correct.

“This reasoning AI calculates 100 times more than the foolish AI,” he said. “It was in the opposite, the complete reflection that everyone did.”

In late January, Deepseek’s model, investors, less energy and money using the model and the highest competitors, caused a mass sale in AI shares. Nvidia plummeted In one session to lose 17%, about $ 600 billion, The largest A single day drop for a US company.

Huang also spoke about some of the things company obvious So far, lighten the partnership with new AI infrastructure, including robotics and enterprises, including robotics and enterprises Dell, Hpe, Mamur, Serviceenow and Community. The AI ​​was widely reflected in BOOM, the hype was purely from the Generative AI to models. It also forecasted the cost of the world’s computational capital expenditures to the end of the decade and the majority of this money would be used for AI.

“Thus, the opportunity of this decade is very great, as a trillion dollar percent,” Huang said. “We have a lot of infrastructure to build.”

Register now For the CNBC investment club, Jim Cramer to watch every move in the market.

Defender The CNBC investment club saves the shares of the philanthropist trust NVIDIA and Crowdstrica.