Nvidia (NVDA) stock dropped nearly 17% Monday, leading a sell-off across chip stocks and the broader market after a new AI model from China’s DeepSeek raised questions about AI investment and the rise of more cost-efficient artificial intelligence agents.
Chinese startup DeepSeek released a new AI model on Jan. 20 viewed as a threat to OpenAI. American venture capitalist Marc Andreessen called the model “one of the most amazing and impressive breakthroughs I’ve ever seen.”
The news came just a month after DeepSeek said one of its latest AI models cost just $5.6 million to train. OpenAI’s GPT model cost more than $100 million to train.
The announcements spurred fears that AI models may begin to require fewer chips and energy than they currently use. Nvidia has become the world’s largest company on the back of exploding demand for its high-end chips that help train AI models.
“If DeepSeek’s innovations are adopted broadly, an argument can be made that model training costs could come down significantly even at U.S. hyperscalers, potentially raising questions about the need for 1-million XPU/GPU clusters as projected by some,” wrote Raymond James semiconductor analyst Srini Pajjuri in a note to investors Sunday evening.
Chip stocks also dropped across the board premarket Monday, with Broadcom (AVGO) down over 18%, Micron (MU) off almost 13%, and Advanced Micro Devices (AMD) down more than 6%.
However, Pajjuri continued, “A more logical implication is that DeepSeek will drive even more urgency among U.S. hyperscalers to leverage their key advantage (access to GPUs) to distance themselves from cheaper alternatives.”
Bernstein analyst Stacy Rasgon also called into question the $5.6 million training cost for DeepSeek’s model, which “does not include all the other costs associated with prior research and experiments on architectures, algorithms, or data.”
Rasgon believes DeepSeek’s announcement was “not really worthy of the hysteria that has taken over the Twitterverse over the last several days.”
Tightened US export restrictions announced in former President Joe Biden’s final days in office could also add a threat to DeepSeek’s ability to continue training new models.
The new rules limit China’s ability to buy Nvidia’s AI chips through resellers and access chips in remote data centers. And China is restricted from importing the most advanced chipmaking machines required to make AI chips from the Dutch firm ASML (ASML).
At the same time, the US is ramping up investment in its own AI infrastructure.
President Trump last week announced the Stargate AI project, which plans to immediately invest $100 billion to build US data centers and other infrastructure — with funding from SoftBank, Oracle, OpenAI and UAE-based MGX — to power generative artificial intelligence. The project would include another $400 billion in spending over the next four years.