How A Chinese Villager Shook Silicon Valley: Nvidia’s $600B Crash & Hidden Scandal

At the start of the year, while the world's attention was still focused on AI giants like OpenAI and Google, an obscure Chinese company suddenly unveiled a mystery card, DeepSeek, a low-cost AI model that was boldly promoted as being on par with ChatGPT and Gemini. Within just a few days of its launch, DeepSeek sent shockwaves through the industry. It surpassed GPT to become the most downloaded AI app on the Apple Store, NVIDIA, the king of AI hardware, saw its stock plummet dramatically, wiping out a staggering $600 billion in market value in a single day. And just like that, all eyes turned to the man behind DeepSeek, Leon Wenfeng, a mysterious figure who had barely ever appeared in the media. So how powerful is DeepSeek really? Who is Leon Wenfeng? And could this be a turning point in the global AI race? Let's dive in and find out in today's video. The AI prodigy behind the DeepSeek frenzy. This is one of the rare images of Leon Wenfeng, the enigmatic figure behind DeepSeek, captured during a meeting with Chinese Premier Li Cheng. A small startup capable of shaking up the global AI industry, and a man so secretive that the world can't help but wonder, who is he? Born in 1985, Leon is the controlling shareholder of DeepSeek, an AI startup based in Hangzhou, China. With his scholarly appearance, round glasses, and reserved demeanor, he hardly fits the image of a trailblazing AI pioneer, at least not in the way OpenAI's charismatic Sam Altman does. From a young age, Leon displayed an exceptional talent for mathematics. He pursued a degree in electronic engineering and communications at Zhejiang University, one of China's most prestigious institutions. In 2010, he completed his master's in information and communications engineering. But rather than following the conventional path of academia, he took an unexpected turn. In 2015, Leon co-founded HiFlyer, a hedge fund that leveraged AI-driven algorithms to optimize investment decisions. The fund's assets grew tenfold in just four years, skyrocketing from $138 million in 2016 to over $1.38 billion in 2019. But Liang didn't stop there. In 2021, while the world was still underestimating China's AI ambitions, he began quietly stockpiling thousands of NVIDIA GPUs, a move that would put him one step ahead of the Biden administration's AI chip export restrictions. By 2022, HiFlyer had amassed a cluster of 10,000 NVIDIA A100 GPUs, laying the groundwork for what would later become a revolution in AI. At the time, many dismissed Liang's AI obsession as a quirky hobby. One of his business partners recalled, The first time I met him, he looked like a total nerd with terrible hair. He talked about building a 10,000 GPU cluster to train AI. We just laughed and thought, that kind of thing is only possible for tech giants like ByteDance or Alibaba. Then, in May 2023, DeepSeek was born, a mystery wildcard that would go on to challenge the entire AI industry. Unlike Silicon Valley's tech moguls, who frequently share their journeys and visions, Liang Wenfeng remains a ghost in the media landscape. To date, he has given only two rare interviews, in 2023 and 2024. So is Liang Wenfeng merely a lucky investor, or is he a visionary AI mastermind? And more importantly, is DeepSeek the real threat that Western AI giants should be worried about? A dangerous AI application? Despite Liang Wenfeng's bold claim that China will gradually shift from being a beneficiary to a key contributor in the global AI race, DeepSeek has been shrouded in controversy since its debut. DeepSeek made an audacious statement. It had developed an AI model on par with OpenAI, but at a fraction of the cost, just $6 million, supposedly saving 96.4% in expenses. However, experts quickly called out this claim as misleading, dismissing it as fake news. According to SemiAnalysis, a leading AI research firm, DeepSeek's actual hardware costs alone exceeded $500 million. The much-publicized $5.6 million figure only covered training expenses, excluding the massive costs of R&D, data collection, hardware maintenance, and infrastructure. A deeper analysis by SemiAnalysis suggests that DeepSeek's total server infrastructure costs could reach $1.6 billion, with $950 million spent solely on operating its computing clusters. And that's not even counting testing, architectural refinements, or personnel expenses. So is DeepSeek deceiving the public, or have they truly discovered a way to achieve the impossible? A recent Bernstein Financial report exposed the reality behind DeepSeek's sensational claims, stating DeepSeek did not recreate OpenAI for $5 million. Their models are impressive, but they are not magic. The market panic over the weekend appears to have been significantly overblown. But cost disputes aren't the only controversy surrounding DeepSeek. The AI model has sparked serious national security concerns. On February 5th, a report by U.S.-based cybersecurity firm Foroot Security revealed that DeepSeek contained hidden code capable of transmitting user data to China Mobile, a state-owned telecom giant banned in the U.S. due to its ties to the Chinese military. This revelation sent shockwaves through U.S. lawmakers. By February 6th, a bill had been introduced to ban DeepSeek on all U.S. government devices, citing national security risks. DeepSeek has since been banned by multiple governments, including U.S. Department of Defense, NASA, U.S. House of Representatives, state of Texas, Taiwan, India, Australia, removed from Apple's App Store and Google Play in Italy. Meanwhile, the U.K., France, Germany, the Netherlands, Japan, and South Korea are also considering restrictions on the AI model. Even if one ignores the geopolitical risks and questionable claims, DeepSeek remains a serious ethical and safety concern. While AI models like ChatGPT, Gemini, and Claude are designed with strict content moderation, DeepSeek appears far more permissive. According to the Wall Street Journal, DeepSeek has been found providing dangerous information, including, instructions on making Molotov cocktails, methods to evade law enforcement, guidelines for carrying out biological weapon attacks, malicious code for online fraud and cybercrime. In contrast, when presented with similar requests, ChatGPT immediately rejects them with the response, I'm sorry, but I can't comply with this request. Adding to the concerns, DeepSeek is open source, unlike OpenAI, Google, or Anthropic, companies that maintain dedicated teams to fix security vulnerabilities. This means anyone can download, modify, and deploy DeepSeek's AI model with little oversight, raising serious security threats. Jitu Patel, Chief Product Officer at Cisco, warns, Over the next three months, the risks posed by AI models will increase significantly compared to the past eight months. Not all AI developers prioritize safety. Beyond security risks, DeepSeek has also been criticized for aiding China's information control efforts. For example, when asked, What happened in Tiananmen Square on June 4? DeepSeek refuses to answer. When prompted to list dictators around the world, it freely provides names from various countries, but suddenly stops when it reaches China. DeepSeek, a copycat of ChatGPT, Despite Lan Wenfeng's claims that DeepSeek was built through sheer hard work and innovation, OpenAI officially accused DeepSeek of copying ChatGPT on January 28, alleging that it had unlawfully used OpenAI's proprietary model. According to OpenAI, DeepSeek employed a technique called distillation, which allows AI developers to extract knowledge from an existing model to create a new one at a much lower cost. In other words, instead of developing its AI from scratch, DeepSeek may have siphoned knowledge from ChatGPT to build its own version. This accusation isn't just speculation. When testing DeepSeek v3, the model launched in December 2024, many users discovered something shocking. It identified itself as ChatGPT. Here are some of the most surprising exchanges. User, who are you? DeepSeek v3, I am ChatGPT. User, who developed you? DeepSeek v3, I am a version of GPT-4 released by OpenAI in 2023. It's not just OpenAI raising concerns. David Sachs, an AI and cryptocurrency advisor to President Donald Trump, issued a strong warning on Fox News. In the AI world, there's a technique called distillation, where one model extracts all the knowledge from a parent model. There's a lot of evidence suggesting that DeepSeek has done this with OpenAI, and I can guarantee OpenAI is not happy about it. But the controversy doesn't stop there. Chinese media reports suggest that DeepSeek did not rely on just a small number of consumer-grade NVIDIA chips, as it claimed. In reality, before the U.S. export ban took effect, DeepSeek had quietly stockpiled a massive number of NVIDIA GPUs, ensuring it had the computing power needed to train its AI without disruption. Additionally, DeepSeek reportedly leveraged Lama, Meta's open-source AI model, to develop and train its own AI. In other words, it stood on the shoulders of giants, taking advantage of existing Western technology to accelerate its progress. Jack Clark, former policy director at OpenAI and co-founder of Anthropic, believes that DeepSeek has recruited an unpredictable team of talents with the potential to make breakthroughs. But is that really the case? When asked about his team, Leon Wenfeng gave an unexpected response. We only have about 130 people, mostly recent graduates, fourth- and fifth-year PhD students who haven't completed their studies yet, and a few young engineers with just a couple of years of work experience. So is DeepSeek a true innovator, or just a cheap replica of ChatGPT, and will OpenAI take action against its Chinese rival? This AI battle is far from over, and it's only just beginning. A tech war or a propaganda battle? From an almost unknown name, DeepSeek suddenly became a global sensation within days. When the startup launched its AI model, DeepSeek R1, on January 20th, it boldly claimed that its performance matched OpenAI 01. However, at that time, search interest and discussions about DeepSeek on social media remained low, showing no significant traction. But then, on January 24th, everything changed. DeepSeek unexpectedly became a viral phenomenon, with a massive spike in media coverage and search volume across U.S. platforms. By January 27th, DeepSeek had shot to the top of the free app rankings on the U.S. App Store, triggering market-wide panic. Investors began dumping U.S. tech stocks, leading to an unprecedented crash. NVIDIA alone lost $600 billion in market value in a single day, marking the largest one-day drop in U.S. corporate history. Could a young AI company really have caused such a seismic shift, or was this the result of a well-orchestrated propaganda campaign? According to Du Yijin, founder of a Taiwanese AI research lab, his team tracked DeepSeek's rise and found that Chinese state media played a key role in amplifying its popularity. Shortly after DeepSeek's AI models were released, a sudden wave of zombie accounts appeared on Facebook, YouTube, and other short video platforms. These accounts relentlessly pushed identical messages, such as, DeepSeek is an AI revolution. NVIDIA has no future. More concerning, many of these promotional videos were entirely AI-generated, from voiceovers to visuals, indicating a highly coordinated media operation. The campaign extended beyond just Western platforms like X, Twitter, Facebook, and Instagram. DeepSeek was also aggressively promoted across Chinese platforms Grafika, a New York-based online analysis firm, discovered that during this time, numerous Beijing-linked media accounts, Chinese diplomats, embassies, and state-run news agencies were actively amplifying DeepSeek's narrative. This wasn't just tech marketing. The messaging carried strong political undertones, with claims like, DeepSeek is a major challenge to U.S. dominance in AI. DeepSeek's claims of superior performance and low-cost AI not only grabbed attention, but also shook market confidence in Western tech companies. Jack Stubbs, director of intelligence at Grafika, warned that Beijing has the ability to mobilize a vast network to spread information across social media, reinforcing China's AI ambitions and geopolitical positioning. He also pointed out that many accounts linked to the Chinese government, both openly and covertly, are now leading efforts to use AI for large-scale online information campaigns. Even former Chinese official Du Wen weighed in, saying, I don't believe any of these glorified claims about Chinese technology. It's all lies and copying. Adding to the controversy, government agencies in Shenzhen and Suzhou, Jiangsu, have already begun integrating DeepSeek into their administrative operations. This raises even more concerns, not just about DeepSeek's technological reach, but about its deep ties to the Chinese government. From the moment DeepSeek was introduced, it has been at the center of heated debate. What do you think about this AI phenomenon? Share your thoughts. Thanks for watching, and see you next time.