In the escalating technological cold war between Washington and Beijing, a new and deeply unsettling revelation has emerged, casting a long shadow over one of China’s most celebrated artificial intelligence darlings.
DeepSeek, the Hangzhou-based AI firm that sent ripples of excitement and consternation through the global tech community just months ago, now stands accused by a senior U.S. official of actively aiding China’s military and intelligence operations.
The firm is also accused of simultaneously employing elaborate schemes to circumvent stringent American export controls on advanced semiconductors.
This isn’t merely a rehash of familiar trade tensions; it paints a stark picture of a sophisticated, multi-layered strategy to acquire critical technology and leverage AI for strategic national objectives.
The allegations, previously unreported, suggest that DeepSeek’s meteoric rise might be less about pure innovation and more about an intricate dance with state interests and a determined evasion of international restrictions.
Remember the buzz?
It was only in January that DeepSeek captivated the AI world, boldly claiming its reasoning models were on par with, or even surpassed, U.S. industry leaders like OpenAI and Meta, all at an astonishingly low cost.
The narrative was compelling: a Chinese upstart challenging Silicon Valley’s dominance with efficiency and ingenuity.
But according to the anonymous senior State Department official, that narrative masks a far more complex and troubling reality.
“We understand that DeepSeek has willingly provided and will likely continue to provide support to China’s military and intelligence operations,” the official stated, emphasizing that this goes “above and beyond open-source access to DeepSeek’s AI models.” The implications are profound.
Beyond direct military assistance, DeepSeek is accused of sharing user information and statistics with Beijing’s pervasive surveillance apparatus. While Chinese law mandates data submission upon request, the suggestion that DeepSeek is already actively doing so raises immediate and serious privacy concerns for its tens of millions of daily global users.
Trust, a fragile commodity in the digital age, is precisely what such allegations erode.
For any user, anywhere, the thought that their interactions with a seemingly innocuous AI model could be funneled into a state surveillance system is chilling.
Further cementing these claims, the official pointed to procurement records, revealing DeepSeek’s name appearing over 150 times in connection with China’s People’s Liberation Army (PLA) and other entities linked to the Chinese defense industrial base.
The firm is said to have provided technology services directly to PLA research institutions, effectively embedding itself within the very core of China’s military-industrial complex. While Reuters could not independently verify this specific procurement data, the sheer volume of alleged connections, if true, paints a damning portrait.
Perhaps the most critical facet of the U.S. assessment revolves around DeepSeek’s alleged access to and acquisition of high-end U.S.-made chips, particularly Nvidia’s coveted H100s. These chips, the very backbone of cutting-edge AI development, have been under strict U.S. export restrictions since 2022, precisely to prevent China from utilizing them to advance its military capabilities or gain an insurmountable lead in the AI arms race.
The official detailed DeepSeek’s alleged use of shell companies in Southeast Asia to evade these controls, and its attempts to access data centers in the region to remotely tap into U.S. chips. This maneuver highlights the complex and often frustrating game of cat and mouse being played out on the global tech stage.
While importing restricted chips into China without a license is a clear violation, Chinese companies are still permitted to access these same chips remotely in data centers located in non-restricted countries.
It’s a loophole, albeit a narrow one, and one that U.S. officials are clearly scrutinizing. The only exceptions are if a Chinese company is on a U.S. trade blacklist, or if the chip exporter knows the Chinese firm is using its chips for developing weapons of mass destruction.
Notably, DeepSeek has not yet been placed on any U.S. trade blacklists, nor has Nvidia been accused of knowledge regarding DeepSeek’s alleged military ties. Nvidia, for its part, maintains a careful distance, stating it does not support parties that have violated U.S. export controls or are on U.S. entity lists.
An Nvidia spokesman also clarified that their review indicated DeepSeek used lawfully acquired H800 products, not H100s, adding another layer of complexity and potential obfuscation to the narrative.
The precise number of H100 chips DeepSeek might possess remains unverified, though sources familiar with the matter suggest it’s far less than the 50,000 once claimed by another AI startup CEO.
The ripple effects of these allegations are already being felt across Southeast Asia. Singapore, a key regional tech hub, has already charged three men with fraud in a case domestic media has linked to the movement of Nvidia’s advanced chips from the city-state to DeepSeek.
Malaysia, too, announced last week it was investigating an unnamed Chinese company using servers equipped with Nvidia chips for large language model training, examining potential breaches of domestic law.
These investigations underscore the regional complicity, witting or unwitting, in the intricate networks designed to bypass U.S. restrictions.
DeepSeek, for its part, has remained silent on the accusations regarding its privacy practices, the acquisition of Nvidia chips, or the alleged use of shell companies.
China’s foreign and commerce ministries also declined to comment on the Reuters inquiry.
This silence, in itself, speaks volumes in the charged atmosphere of U.S.-China tech rivalry. The U.S. government’s conclusions reflect a growing skepticism in Washington that the capabilities behind the rapid rise of some of China’s flagship AI enterprises may have been exaggerated, and that their true prowess is heavily reliant on circumvented access to cutting-edge U.S. technology.
The initial claims of DeepSeek’s models being trained for a mere $5.58 million in computing power were met with significant skepticism from AI experts, who argued the true costs were likely far higher.
This latest intelligence from Washington only compounds those doubts, suggesting that the emperor’s new clothes might indeed be woven from stolen threads.
As the U.S. weighs its next moves – with the State Department official stating “nothing to announce at this time” regarding further export controls or sanctions against DeepSeek – the saga of DeepSeek serves as a potent reminder.
It underscores the profound challenges of regulating dual-use technologies, the blurred lines between civilian innovation and military application, and the unrelenting global competition for technological supremacy.
In this high-stakes game, trust is a luxury, and every breakthrough, every claim, is now viewed through the lens of national security and geopolitical advantage.
