DeepSeek Shakes the World!
On January 10, 2025, the Chinese artificial intelligence company DeepSeek officially launched a free chat application based on its in-house R1 model. The application was made available on major platforms, including iOS and Android. By January 27, DeepSeek had surpassed ChatGPT in download volume on the U.S. iOS App Store, making it the most popular free application at the time.
DeepSeek is not only capable of answering questions and handling various logical computations like ChatGPT, but it can also generate code. Users have noted that its performance is on par with, if not superior to, other AI models currently available on the market.
What is most shocking is that, according to DeepSeek’s official statement, they successfully trained their V3 model using only around 2,000 Nvidia H800-series GPUs over a period of 55 days, with a total cost of no more than $5.58 million. This figure is astonishing when compared to the usual training expenses for large language models, which often run into tens or even hundreds of millions of dollars.
The media has even referred to DeepSeek’s rise as a "Sputnik moment" for the United States, highlighting the profound impact it has had on the American AI industry. Even President Trump has acknowledged its significance, stating that "DeepSeek has sounded the alarm for the U.S. AI sector."
DeepSeek’s success has sent shockwaves through the global technology stock market, as investors and industry experts alike grapple with the realization that AI training may not be as expensive or as dependent on high-end chips as previously assumed.
As a result, by the close of trading on January 27, Nvidia's stock price had plunged by approximately 17%, wiping out $600 billion in market capitalization overnight. Other tech giants such as Broadcom, Microsoft, Alphabet (Google's parent company), and Dutch semiconductor equipment manufacturer ASML also saw their stock prices decline. By January 28, the global AI and computer hardware sector had collectively lost approximately $1 trillion in market value overnight.
Behind the Shock: The Myth Shattered – False Cost Reports and Intellectual Property Theft
As news of DeepSeek’s success spread widely, the prominence of the DeepSeek team surged. The company's founder, Liang Wenfeng, was hailed by CNN as "China's Sam Altman," positioning him as a pioneer in China's AI development. Several Chinese media outlets began publishing extensive coverage praising DeepSeek, branding it as a "national asset" of China. Even Chinese Premier Li Qiang invited Liang Wenfeng to participate in an expert roundtable discussion, requesting his input and suggestions for the 2024 Government Work Report (draft for public consultation).
Most people believe that DeepSeek's impressive training results, achieved with limited resources, demonstrate the futility of U.S. sanctions aimed at restricting China's AI development, including export controls on advanced AI chips.
However, DeepSeek's achievements quickly came under scrutiny, prompting a wave of investigations. The first shot was fired by the semiconductor research firm SemiAnalysis, which released a report challenging DeepSeek’s claims. According to the report, DeepSeek had actually utilized Nvidia's Hopper architecture chips, including 10,000 units of H800, 10,000 units of H100, and even the China-exclusive H20 model. The total capital expenditure was estimated to exceed $1.6 billion, with more than $500 million specifically allocated for GPUs.
The $6 million cost mentioned by DeepSeek only accounts for the GPU expenses of the pre-training run and does not reflect the full expenditure. Research and development, as well as hardware costs, were not included in the calculation.
Even more critically, DeepSeek's primary investor, the hedge fund "Lingfeng Quant" , had strategically stockpiled a large quantity of Nvidia's high-end A100 chips before the semiconductor export restrictions took effect, ensuring ample computing resources for the company's AI training efforts.
These facts indicate that DeepSeek's success remains highly dependent on Nvidia's chip computing power and is not the self-sufficient "Chinese achievement" that Chinese media have portrayed it to be.
Additionally, the source of DeepSeek’s chips raises further suspicions. It is possible that DeepSeek circumvented U.S. export controls on advanced chips by acquiring a large number of Nvidia chips through third parties to support its AI model development. In response, the U.S. Department of Commerce has launched an investigation into whether DeepSeek utilized American chips that were banned from being shipped to China.
DeepSeek has stated that it used Nvidia’s H800 chips, which were still legally available for purchase in 2023. However, the U.S. government is also investigating whether DeepSeek obtained high-performance Nvidia chips through third-party companies based in Singapore.
The Financial Times further added to the controversy with a critical report. OpenAI issued a statement claiming that it had evidence suggesting DeepSeek used OpenAI’s models to train its own system—a practice known as "data distillation." This technique involves leveraging outputs from a larger AI model to train a smaller-scale model, thereby enhancing computational efficiency, achieving higher performance, and significantly reducing training costs.
This contradicts Liang Wenfeng's claims. Liang has consistently stated that DeepSeek’s models were built from the ground up. However, if OpenAI's accusations prove to be true, then DeepSeek is merely standing on the shoulders of giants—without permission.
Trump’s AI czar, David Sacks, who also serves as the Director of AI and Cryptocurrency Policy, further warned that "intellectual property theft is now a very real and likely scenario."
Even users within China have begun to express skepticism about DeepSeek's capabilities. A user from Xinjiang reported an unusual incident: he submitted a research paper co-written with ChatGPT and Claude to DeepSeek, requesting revision suggestions. However, to his surprise, DeepSeek provided feedback that was identical to ChatGPT’s suggestions.
What’s even more astonishing is that the feedback appeared to be based on a previous version of the paper—a version that the user had never submitted to DeepSeek. This raised concerns over whether DeepSeek had access to external AI models or if it was engaging in unauthorized data collection.
Some users have also discovered that when they ask DeepSeek about the model it is based on, it responds by stating that it is built on GPT-3.5.
These inconsistencies and anomalies strongly suggest that DeepSeek's training process is closely tied to OpenAI.
A Classic Example of China's Narrative Warfare Deception
As the myth surrounding DeepSeek crumbled, Wall Street swiftly corrected and rebounded. The market realized that AI training still heavily relies on high-end chips, leading Nvidia’s stock price to quickly recover to $129 the following day. By February, it once again touched the $130 mark. JPMorgan stated that Nvidia remains a key beneficiary of low-cost AI models, while Bank of America reaffirmed its "Buy" rating for Nvidia, setting a target price of $190, suggesting a potential 50% upside.
The shockwave caused by DeepSeek is, in fact, a textbook example of China’s cognitive warfare strategy. By generating massive hype and posturing, China aims to undermine its opponents' confidence and force them into submission. The Chinese government is leveraging the launch of DeepSeek to push two key narratives: first, that U.S. chip export controls are ineffective, and second, that China's domestic chip industry is thriving.
As investors lose confidence and the public becomes unsettled, China gains valuable material to reinforce its political narrative. At the same time, it inflicts significant damage on its adversaries' financial and investment markets, ultimately undermining economic stability in the U.S.
In reality, DeepSeek's success remains heavily dependent on high-end software and hardware suppliers from the United States. This includes OpenAI’s pioneering advancements in AI model development, as well as Nvidia’s cutting-edge GPU design and manufacturing.
A London-based biotech executive issued a ban on DeepSeek, and his reasoning serves as a fitting conclusion to this incident. A biotech employee in London shared a message from his boss: "It is almost certainly harvesting everything for the Chinese state." The Chinese Communist Party aims to leverage DeepSeek to ruthlessly harvest data and intelligence from across the globe.
Only by exposing and dismantling these deception tactics and narrative strategies can we effectively counter the Chinese Communist Party’s cognitive warfare. Recognizing these methods allows industries to fortify themselves, safeguard their competitive edge, and ultimately protect national interests in the ongoing global competition.
