Why the Deepseek model requires “100 times more computing”

CEO Jensen Juan discussed impact New Artificial Intelligence Model Chinese Startup Deepseek in an interview CNBC’s CNBC Jim Kramer at NvidiaAnnual GTC Conference. Juan claimed to actually need the most important model More computing What many thought in the industry.

According to Huang, the Model R1 Deepseek- “fantastic” because it is “the first outdoor reasoning model”. He explained that the model destroys problems step by step, capable of thinking different answers, and it can make sure whether its correct answer.

“This reasoning II consumes 100 times more calculated than not disassembled,” he said. “It was exactly the opposite, it was the exact opposite of the conclusion that everyone had.”

At the end of January, the Deepseek model caused a large -scale AI stock sale because investors feared the model could perform as well as leading competitors using less energy and money. Nvidia fell 17% for one session to lose almost $ 600 billion, the largest in history One day drop for an American company.

Juan also told about some of what his company has disclosed So far, at its conference, including the new AI infrastructure for robotics and the enterprise, emphasizing partnerships with companies such as such companies Declaration. Hpe. Accent. Servicenow and Crowdstrike. He thought widely about AI BOOM, noting that the excitement had moved from purely generative II to models. He also predicted that world computing, it seems, by the end of the decade to reach a trillion dollars and that most of this money would be used for the II.

“Thus, our ability to do a percentage of dollars by the end of this decade is quite big,” Juan said. “We have a lot of infrastructure to build.”

Sign up now For the CNBC investment club to keep track of Jim Cremer’s step on the market.

Refusal CNBC Investing Club Club Trust takes Nvidia and Crowdstrike stocks.