In This Article:
Nvidia (NASDAQ: NVDA) is known for its dominance in artificial intelligence (AI) chips, holding about 80% of the market. The company sells the world's most expensive AI chips -- or graphics processing units (GPUs) -- but customers are willing to pay the price thanks to these products' top performance. Nvidia chief Jensen Huang has even said that, over time, Nvidia's chips are the best deal as they save on total cost of ownership.
For Nvidia to keep on winning, customers must continue to flock to the company for their AI needs -- and get in on the company's latest and highest-priced chips. The tech giant, promising AI customers updated chips on an annual basis, has used innovation to keep this momentum going.
But late last month, something happened that called into question Nvidia's ability to keep revenue growth soaring over the long term. Chinese startup DeepSeek announced that it had trained its R1 model in two months for less than $6 million. That compares with the billions of dollars top U.S. tech companies have spent on Nvidia AI chips. As a result, Nvidia stock sank 17% in one trading session, losing nearly $600 billion in market value.
DeepSeek's news clearly shocked the market -- but Nvidia just did something that should crush the panic.
DeepSeek's "excellent AI advancement"
Before we get to that, though, let's take a closer look at the DeepSeek news. As mentioned, the startup trained a model for much less than American tech companies have been spending on their programs. And DeepSeek says its R1 model rivals OpenAI's model o1, suggesting companies could slash their current AI budgets and still obtain pretty decent results. Nvidia even called DeepSeek's accomplishment "an excellent AI advancement."
Investors, concerned that companies actually would reduce spending on Nvidia's most powerful GPUs, rushed to sell Nvidia stock. The idea was revenue could decline significantly if customers decide they don't need the tech giant's latest innovations and instead could opt for less expensive GPUs from Nvidia or rivals.
But it's important to keep in mind that when things look too good to be true, they often are. First, we can't be certain the R1 training cost less than $6 million. Experts have said that, considering all of the steps involved in bringing a model to the launch stage, the figure looks much too low. For example, a report by consulting firm Semianalysis estimates DeepSeek spent more than $500 million on the project.
Another important point: DeepSeek has said it used Nvidia chips designed for the Chinese market. These are meant to be less powerful than Nvidia's main line of chips to comply with the U.S. government's export controls. But some experts have questioned whether DeepSeek also relied on other Nvidia chips potentially purchased before the export controls kicked in. We don't have answers to these questions, of course, but what this shows us is the picture isn't detailed enough to prove that cheaper chips can do the job.