WASHINGTON/SAN FRANCISCO — AI firm DeepSeek is aiding China’s military and intelligence operations, a senior U.S. official told Reuters, adding that the Chinese tech startup sought to use Southeast Asian shell companies to access high-end semiconductors that cannot be shipped to China under U.S. rules.
The U.S. conclusions reflect a growing conviction in Washington that the capabilities behind the rapid rise of one of China’s flagship AI enterprises may have been exaggerated and relied heavily on U.S. technology.
Hangzhou-based DeepSeek sent shockwaves through the technology world in January, saying its artificial intelligence reasoning models were on par with or better than U.S. industry-leading models at a fraction of the cost.
“We understand that DeepSeek has willingly provided and will likely continue to provide support to China’s military and intelligence operations,” a senior State Department official told Reuters in an interview.
“This effort goes above and beyond open-source access to DeepSeek’s AI models,” the official said, speaking on condition of anonymity in order to speak about U.S. government information.
The U.S. government’s assessment of DeepSeek’s activities and links to the Chinese government have not been previously reported and come amid a wide-scale U.S.-China trade war.
Among the allegations, the official said DeepSeek is sharing user information and statistics with Beijing’s surveillance apparatus.
The big three U.S. cloud providers Amazon, Microsoft and Alphabet’s Google offer DeepSeek to customers.
Chinese law requires companies operating in China to provide data to the government when requested. But the suggestion that DeepSeek is already doing so is likely to raise privacy and other concerns for the firm’s tens of millions of daily global users. The U.S. also maintains restrictions on companies it believes are linked to China’s military-industrial complex.
U.S. lawmakers have previously said that DeepSeek, based on its privacy disclosure statements, transmits American users’ data to China through “backend infrastructure” connected to China Mobile, a Chinese state-owned telecommunications giant.
DeepSeek did not respond to questions about its privacy practices.
The company is also referenced more than 150 times in procurement records for China’s People’s Liberation Army and other entities affiliated with the Chinese defense industrial base, said the official, adding that DeepSeek had provided technology services to PLA research institutions.
Reuters could not independently verify the procurement data.
The official also said the company was employing workarounds to U.S. export controls to gain access to advanced U.S.-made chips.
DeepSeek has access to “large volumes” of U.S. firm Nvidia’s NVDA.O high-end H100 chips, said the official. Since 2022 those chips have been under U.S. export restrictions due to Washington’s concerns that China could use them to advance its military capabilities or jump ahead in the AI race.
“DeepSeek sought to use shell companies in Southeast Asia to evade export controls, and DeepSeek is seeking to access data centers in Southeast Asia to remotely access U.S. chips,” the official said.
The official declined to say if DeepSeek had successfully evaded export controls or offer further details about the shell companies.
DeepSeek also did not respond to questions about its acquisition of Nvidia chips or the alleged use of shell companies.
When asked if the U.S. would implement further export controls or sanctions against DeepSeek, the official said the department had “nothing to announce at this time.”
China’s foreign ministry and commerce ministry did not respond to a Reuters request for comment.
“We do not support parties that have violated U.S. export controls or are on the U.S. entity lists,” an Nvidia spokesman said in a prepared statement, adding that “with the current export controls, we are effectively out of the China data center market, which is now served only by competitors such as Huawei.”
Access to restricted chips
DeepSeek has said two of its AI models that Silicon Valley executives and U.S. tech company engineers have showered with praise. DeepSeek-V3 and DeepSeek-R1, are on par with OpenAI and Meta’s META.O most advanced models.
AI experts, however, have expressed skepticism, arguing the true costs of training the models were likely much higher than the $5.58 million the startup said was spent on computing power.
Reuters has previously reported that U.S. officials were investigating whether DeepSeek had access to restricted AI chips.
DeepSeek has H100 chips that it procured after the U.S. banned Nvidia from selling those chips to China, three sources familiar with the matter told Reuters, adding that the number was far smaller than the 50,000 H100s that the CEO of another AI startup had claimed DeepSeek possesses in a January interview with CNBC.
Reuters was unable to verify the number of H100 chips DeepSeek has.
“Our review indicates that DeepSeek used lawfully acquired H800 products, not H100,” an Nvidia spokesman said, responding to a Reuters query about DeepSeek’s alleged usage of H100 chips.
In February, Singapore charged three men with fraud in a case domestic media have linked to the movement of Nvidia’s advanced chips from the city state to DeepSeek.
China has also been suspected of finding ways to use advanced U.S. chips remotely.
While importing advanced Nvidia chips into China without a license violates U.S. export rules, Chinese companies are still allowed to access those same chips remotely in data centers in non-restricted countries.
The exceptions are when a Chinese company is on a U.S. trade blacklist, or the chip exporter has knowledge that the Chinese firm is using its chips to help develop weapons of mass destruction.
U.S. officials have not placed DeepSeek on any U.S. trade blacklists yet and have not alleged that Nvidia had any knowledge of DeepSeek’s work with the Chinese military.
Malaysia’s trade ministry said last week that it was investigating whether an unnamed Chinese company in the country was using servers equipped with Nvidia chips for large language model training and that it was examining whether any domestic law or regulation had been breached.
(Reporting by Michael Martina in Washington and Stephen Nellis in San Francisco; Additional reporting by Fanny Potkin in Singapore; Editing by Don Durfee, Kenneth Li, Cynthia Osterman, Deepa Babington, Himani Sarkar and Emelia Sithole-Matarise)
