DeepSeek who?
Nvidia said demand for its AI chips remained strong in its first quarterly earnings report since DeepSeek’s cheaply trained models raised questions about the need for Nvidia’s state-of-the-art chips. The Nvidia results—often read as tea leaves for broader AI momentum—came after a post-DeepSeek earnings season left the rest of Big Tech largely unscathed.
So what happened to those fears that the upstart Chinese lab’s innovation could mark a new era of shoestring budgets? Analysts say they may have been overblown—and the real story is a bit more complicated.
To recap: DeepSeek’s ability to train AI that performs on par with leading models despite export controls curbing access to the latest chips triggered a sell-off of Nvidia and other Big Tech stocks earlier this year.
As the company that’s cornered the market on AI chips, Nvidia has a lot riding on the tens of billions that tech giants are planning to plow into AI infrastructure this year. But DeepSeek’s efficiency measures—if its budget disclosures indeed hold water—raised questions about whether that scale was even necessary.
“It was a shock to the system in terms of realizing that you didn’t have to maintain the current course of spending more, faster,” Alvin Nguyen, senior analyst at Forrester Research, told Tech Brew.
Except that: One issue with that narrative, though, according to Nguyen, is that Nvidia’s customers are largely hyperscalers or other deep-pocketed buyers ready to shell out for the best of the best when it comes to AI infrastructure. These companies are in a tight race where any advantage coming from better and faster chips makes a difference, Nguyen said.
“The reason why [DeepSeek] didn’t have that impact is people realized that if you’ve got the best-performing GPUs, if you’ve got the Nvidia AI accelerators, it does give you an advantage. Having faster means better results or faster results,” Nguyen said. “They’re selling to hyperscalers, not enterprises. They’re selling to people who can, who do, benefit from finishing first, buying more, having better performance.”
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
Because Nvidia sees more demand than it can meet, it’s essentially bound by the “golden handcuffs” of shareholder obligation to serve products with the highest profit margins—top-shelf chips for upmarket buyers, he said.
Demos democratized: The impact is more likely to hit the lower end of the market, where more cost-conscious companies might have newfound access to experimentation with reasoning models like DeepSeek’s R1, Nguyen said.
“It democratizes that form of AI, the reasoning models where you can use the CPU, not a GPU; the performance might be less, but if you’re experimenting, that’s fine,” Nguyen said.
Not ready for prime time: Chirag Dekate, VP analyst at Gartner, said markets are prematurely betting on a shift from AI training to inferencing—or running already-trained AI. That evolution could result in more competition for Nvidia from companies with lower-performance offerings, but Gartner believes it’s still a way off.
“We are not done with the training yet,” Dekate told us. “GenAI models today are the worst they are going to be, and these models will only keep getting better. Frontier model innovators continue to drive investments in leadership class AI supercomputers, which involve extreme scaling of model performance, capabilities, and accuracy. Most of the industry today relies largely on Nvidia GPUs for these efforts, barring a few exceptions.”
While DeepSeek exhibits some promising efficiency measures, reproducing them at production scale will still require top chips, Dekate said. “There is no such thing as a free lunch.”