Jensen Huang says the market got DeepSeek all wrong

When DeepSeek dropped its R1 model a month ago, Wall Street panicked. The sudden fear was straightforward: if an open-source model can match frontier reasoning with far less hardware, why would anyone keep buying Nvidia's expensive GPUs?

That knee-jerk reaction triggered a historic rout. Nvidia’s stock crashed 16.9% in a single day, erasing roughly $600 billion in market value over just three trading sessions. But Nvidia CEO Jensen Huang thinks investors got the story completely backwards.

Speaking with DataDirect Networks CEO Alex Bouzari in an interview released this Thursday, Huang pushed back against the doom-and-gloom narrative. Instead of viewing R1 as a threat, he called the open-source reasoning model “incredibly exciting.”

"Complete opposite" of what the market thinks

Huang didn't hold back on mocking the market's initial freakout. He noted people reacted as if AI was suddenly finished and models had magically dropped out of the sky without needing compute anymore.

“It’s exactly the opposite,” Huang said. R1 proved that models can be far more efficient than previously thought. But in his view, higher efficiency doesn't kill demand—it broadens the entire market and accelerates widespread AI adoption.

More importantly, Huang highlighted where the compute is actually heading: post-training and inference. While DeepSeek found smarter ways to train base models, refining models through reasoning is still a massive computational sponge. As Huang put it, reasoning remains an extremely compute-intensive process.

A $600B dip that vanished in a month

The market seems to be coming around to that view. On January 24, Nvidia closed at $142.62 per share. Following the DeepSeek news, it plunged to $118.52 by Monday, January 27.

Fast forward roughly a month, and the stock opened at $140 on Friday, effectively reclaiming almost all the lost ground. Nvidia is set to report its Q4 earnings on February 26, where Huang will likely give a deeper breakdown of ongoing GPU demand.

Meanwhile, the pressure from open source isn’t slowing down. DeepSeek announced on Thursday that it plans to release five new code repositories during an "open source week" next week.

Why it matters

Huang’s defense touches on a classic tech principle known as Jevons' Paradox: making a resource cheaper and more efficient usually causes people to consume far more of it, not less.

For developers and startups, DeepSeek-R1 made high-end reasoning cheap enough to build into everyday products. But as millions of end users start running complex reasoning loops at scale, the aggregate demand for inference compute skyrockets.

We think Huang made the right call here. The narrative that AI training would freeze overnight was always flimsy. While tech giants might rethink brute-force pre-training budgets, the shift toward test-time compute and intensive post-training plays right into Nvidia’s wheelhouse. The $600 billion dip wasn't a warning shot for Nvidia’s business—it was just Wall Street misunderstanding how software efficiency fuels hardware demand.