The DeepSeek AI application is seen on a mobile phone in this photo illustration taken in Warsaw, Poland on 27 January, 2025.
Jaap Arriens | Nurphoto | Getty Images
Chinese artificial intelligence startup DeepSeek could face curbs from the U.S. government as it upends the U.S. AI ecosystem, though enforcing restrictions on an open-source technology could be a challenge, experts said.
DeepSeek’s sudden rise has questioned the effectiveness of Washington’s efforts aimed at curbing China’s access to high-end tech over national security concerns.
For DeepSeek, which relied heavily on open-source code, the additional export restrictions that the U.S. government could impose are limited, said Lawrence Ward, a partner at U.S.-based law firm Dorsey & Whitney with a specialization in national security law.
In terms of the company’s potential use of certain Nvidia chips, it might face civil and criminal penalties, but enforcing those penalties may be “difficult if not impossible,” Ward said.
On Monday, the U.S. lawmakers called for actions to slow down the Chinese tech startup, with some calling DeepSeek “a serious threat.”
There are no easy solutions to restrict the use of an open-source model, especially one that is being widely tested and used by organizations and individuals, said Paul Triolo, partner at Albright Stone Group.
One option would be for the Commerce Department to craft rules that require tech giants such as Apple and Google to take down DeepSeek’s app, restricting its downloads in the U.S. market, Triolo said, adding that it would, however, be challenging to pull the app off other platforms such as Github.
DeepSeek in December released a free, open-source large language model, which it claimed was built in just two months and at a fraction of the cost borne by other major players. Last week, the company released a reasoning model that reportedly outperformed OpenAI’s latest offerings in many third-party tests.
Deepseek, whose AI chatbot has become the most downloaded free application on Apple’s App Store in the U.S., said it created the model despite the sweeping controls imposed in 2022 by the U.S. on China’s access to the most advanced Nvidia chips.
The Hangzhou-based startup said it used the less-advanced Nvidia H800 chips to build the model, which was available until the U.S. government widened the ban in October 2023.
The situation is somewhat unprecedented, and it is not likely that anyone in Washington has a clear idea what to do about it, we are clearly in uncharted waters at many levels.
Paul Triolo
Partner, Albright Stone Group
“If DeepSeek had equipment that it shouldn’t have, it could certainly lead to further investigation,” said Daniel Newman, CEO of Futurum Group, who expects the U.S. government to look into DeepSeek’s hardware which might have violated export curbs.
U.S. President Donald Trump said Monday that the advancement of Chinese AI app DeepSeek “should be a wake-up call” for America’s tech companies, while lauding the low-cost model as “very much a positive development” for AI overall. “Instead of spending billions and billions, you’ll spend less, and you’ll come up with, hopefully, the same solution,” Trump said.
DeepSeek’s claimed cost efficiency and effectiveness of its model sent U.S. tech stocks into a tailspin as investors questioned the amount of money big tech firms have been investing in AI models and data centers. U.S. companies could now leverage the public codes to refine their own models and possibly bring down computation costs.
Rep. John Moolenaar, R-Mich., the chair of the House Select Committee on China, said on Monday the U.S. government should work to place stronger export controls on technologies critical to DeepSeek’s AI infrastructure.
“The situation is somewhat unprecedented, and it is not likely that anyone in Washington has a clear idea what to do about it … we are clearly in uncharted waters at many levels,” Triolo said.
The company said on Monday that it temporarily limited new sign-ups to users with mainland China’s phone numbers following a large-scale malicious cyberattack that caused outages on its website.
On Jan. 20, the day when the the company’s R1 reasoning model was launched to the public, DeepSeek’s founder Liang Wenfeng attended a closed-door symposium for businessman and experts, chaired by Chinese premier Li Qiang, according to state media Xinhua.
CNBC’s Dylan Butts, Hui Jie Lim contributed to this report.
