your123 - stock.adobe.com

DeepSeek's AI breakthrough challenges Nvidia's chip dominance

DeepSeek's budget-friendly AI model challenges chip giants like Nvidia and could spark competition that lowers costs and expands access in the tech industry.

DeepSeek's cost-effective AI model development that rocked the tech world could spark healthy competition in the chip industry and ultimately make AI accessible to more enterprises, analysts said.

Chinese startup DeepSeek claimed to have trained its open source reasoning model DeepSeek R1 for a fraction of the cost of OpenAI's ChatGPT. Analysts were skeptical of DeepSeek's claim that training cost less than $6 million. Nevertheless, they were impressed with the company's development of a model that matches or exceeds ChatGPT despite using significantly less powerful Nvidia chips due to U.S. sanctions on the technology available to China.

DeepSeek's accomplishment shook the tech sector of the U.S. stock market Monday. Nvidia lost $589 billion in market value as investors grappled with whether cheaper hardware could topple sales of its expensive top products used by major customers like AWS, Google and Microsoft to train their cloud-based foundation models.

Doing more with less powerful and cheaper products could open the AI market to more startups and broaden the reach of AMD and Intel processors within enterprises, according to Jack Gold, principal analyst at J. Gold Associates. "Those guys could do really well."

More competition will benefit enterprises through more product choices and lower prices, said Sean Farney, vice president of data center strategy at Jones Lang LaSalle, a global commercial real estate services firm specializing in data centers.

Competition in any market makes it better.
Sean Farney Vice president of data center strategy, Jones Lang LaSalle

"Competition in any market makes it better," Farney said. "This will lead to more development here in the U.S. because there'll be a feeling of threat from other markets and other geographies, and the fear that other folks may be getting ahead in this race."

DeepSeek is also poised to change the dynamics that fueled Nvidia's success and left behind other chipmakers with less advanced products. Conventional thinking has been that models with 10s of thousands of the most powerful chips become more intelligent.

"It was throw more compute and throw more data at the problem, and you're going to magically hit reasoning at some point in time," Gartner analyst Arun Chandrasekaran said.

But that approach is no guarantee you'll achieve the goal of general intelligence. "More importantly, it's a very expensive way to actually scale, to seek reasoning," Chandrasekaran said.

Nvidia welcomes DeepSeek

Nvidia welcomed DeepSeek's accomplishment, calling it "an excellent AI advancement" and appeared confident that "significant numbers of Nvidia GPUs and high-performance networking" would still be needed.

Analysts were wary of DeepSeek's claims of training its model at a fraction of the cost of other providers because the company did not release technical details on its methods for achieving dramatic cost savings.

"The thing that's hard to determine is the [training] cost," Chandrasekaran said. "In some sense, you have to take what they're saying at face value here."

U.S. researchers in the AI market are familiar with DeepSeek's techniques for significantly lowering costs and maintaining model performance, analysts said. Now that DeepSeek has demonstrated that those techniques can be advanced, others in the industry will likely figure out how to do the same.

"There is some novelty, but it's not essentially transformative," Gartner analyst Chirag Dekate said. "This is not somebody coming up with a new technique that completely blindsided the industry."

Forrester Research analysts agreed.

"We expect these optimizations to be copied and improved upon by model builders worldwide," Forrester said in a blog post on Tuesday.

In the meantime, DeepSeek has reminded the tech industry of what researchers have never forgotten -- China is an "AI research powerhouse," Chandrasekaran said.

"Chinese companies have made enormous progress in the last few years in AI," he said.

Antone Gonsalves is an editor at large for Informa TechTarget, reporting on industry trends critical to enterprise tech buyers. He has worked in tech journalism for 25 years and is based in San Francisco. Have a news tip? Please drop him an email.

Dig Deeper on AI infrastructure