Two 59%+ winners, four above 25% in Aug – How this AI model keeps picking winners
NVIDIA (NASDAQ:NVDA) CEO Jensen Huang discussed the elephant in the room – DeepSeek – in a DDN interview with Alex Bouzari on Thursday. Huang said DeepSeek and the energy around its open-source reasoning model are “incredibly exciting,” but investors “got it wrong,” assuming it would be a negative for his company and the AI industry.
NVIDIA’s stock sold off 17% on January 27th, the Monday after DeepSeek took the world by storm. It has since recovered most of those losses.
Huang explains that there is a different paradigm than what investors were considering. He said from an investor perspective, there was a mental model that the world was pre-training and then inference, which is not the case. “I don’t know whose fault it is, but obviously that paradigm is wrong.”
In explaining the correct paradigm, he highlights that there is “pre-training…because we want to have foundation…in order to the second part, which is post-training.” Huang said post-training is going to continue to be important.
Huang said the second layer, the most important, is post-training. This is where “you learn to solve problems.”
The CEO said people got worried that “oh my gosh, pre-training is a lot less…they forgot that post-training is really quite intensive.”
He said the third scaling law is that the more reasoning you do before answering a question, the better. “Reasoning is a fairly compute-intensive part of it," he said.
Overall, Huang points out that the stock market responded to DeepSeek’s R1 as “oh my gosh, AI is finished… it’s exactly the opposite.”