Many in Silicon Valley believe that the large-scale stock sell-off is an overreaction to the Chinese lab's DeepSeek-R1 AI model, the Financial Times reports.
NVIDIA became the most valuable company in the world last year as investors bet on the great need for artificial intelligence processors by technology companies. The chipmaker's CEO Jensen Huang predicted that $1 trillion worth of AI data centers will be built in the next few years.
Underlying this confidence was the concept of the "law of scaling" of artificial intelligence, popularized by the leaders of AI startups such as OpenAI and Anthropic, which suggested that AI models become smarter as they are given more data and computing resources.
The release of the DeepSeek-R1 model and the accompanying research article that openly explains how it was created broke the spell of the law of scaling, as it seems to have used significantly fewer resources to train the model.
In December, DeepSeek released the V3 model, which it claims has similar performance to OpenAI and Google's models, but was allegedly built at a much lower cost of $5.6 million. The Chinese company said it used only 2,048 NVIDIA chips that could be obtained without violating US export controls, which limit China's access to the latest products from US chipmakers.
Last week, DeepSeek unveiled its latest model, the R1, which performs on par with OpenAI's o1. Investors were even more frightened by the fact that DeepSeek was able to create the model without relying on NVIDIA's Cuda software platform, which is widely seen as key to the Silicon Valley chipmaker's dominance in AI development.
The largest U.S. tech companies "had monopolistic access to artificial intelligence - the entry ticket was billions of dollars, otherwise there was no chance of challenging the status quo," said Stephen Yu, chief investment officer at Blue Whale Growth. This made the emergence of DeepSeek "a very positive development for the adoption, development, and penetration of artificial intelligence," he added.
Shorts sellers, who have made a flurry of bets against NVIDIA's sky-high share price in recent weeks, were delighted on Monday. The 17% drop in the company's share price brought shortsellers $6.75 billion in profits, according to calculations by data group S3 Partners.
"A Chinese organization released its open-source AI model right before the earnings release of all the major US tech companies," said one short-seller with interests in a number of major AI companies. "They tell you that [these companies' AI models] have no value, they have become a commodity."
Dylan Patel of the consulting firm SemiAnalysis estimates that DeepSeek and its parent company, the hedge fund High-Flyer, have access to tens of thousands of NVIDIA GPUs that were used to train R1's predecessors.
"DeepSeek has spent well over $500 million on GPUs in the company's history," said Patel. "While their launch was very effective, it required a lot of experimentation and testing."
Dan Hutcheson of TechInsights says that the market reaction doesn't reflect who is most affected by the DeepSeek breach. "I don't see this as a big blow to NVIDIA, I see it as a bigger problem for companies like OpenAI that are trying to sell these services," he explains.
NVIDIA, at least publicly, also says that DeepSeek's developments will benefit rather than undermine its business.
"DeepSeek is a remarkable achievement in AI and a great example of Test Time Scaling," said an NVIDIA spokesperson. "DeepSeek's work illustrates how new models can be created with this technique, using widely available models and computing that are fully compliant with export control requirements.
The implication of NVIDIA's announcement is that by pushing the boundaries of what is possible with open-source AI models, DeepSeek has actually increased demand for the chips used to run them.
"The market doesn't quite realize that this is great for NVIDIA," says Dmitry Shevelenko, chief business officer of Perplexity, a San Francisco-based AI search startup that counts NVIDIA as one of its investors. "No matter what happens, Jensen will win."
However, the high-profile launch of DeepSeek risks escalating into a new stage of confrontation between the US and China, as OpenAI has already stated that after examining DeepSeek-R1, they found evidence of the use of their AI models. Therefore, there is a high probability that export controls on American chips will be tightened, and this may already play against NVIDIA.