On Monday, the global financial landscape faced a jolt as the U.S. stock market plunged significantly, largely due to the actions of a little-known Chinese tech company known as DeepSeek.
It’s crucial to distinguish between DeepSeek and "deepfake." While deepfake technology employs advanced AI to manipulate faces in videos or voices in audio, DeepSeek is an innovative startup located in the city of Hangzhou (known for its natural beauty), China, dedicated to AI research.
Just a short time ago, many tech experts and geopolitical analysts were confident that the United States held a commanding lead over China in the AI race. This belief was fueled by the dominance of U.S.-based companies like Nvidia and OpenAI, which spearhead AI advancements globally. Furthermore, the Biden administration has actively sought to curb China's AI progress by limiting the export of advanced computer chips critical for AI model development. This ongoing rivalry underlines the importance of vigilance in safeguarding U.S. technological leadership.
DeepSeek's recent unveiling of its R1 AI model has caused significant excitement in the U.S. tech sector. AI experts have praised R1 as one of the world's leading AI models, placing it on par with OpenAI's o1 reasoning model—a remarkable achievement for DeepSeek. OpenAI, the pioneering American tech company behind ChatGPT, a key player in the AI revolution, now faces a strong competitor in DeepSeek's R1.
Not only does DeepSeek's R1 model match the performance of its rivals, but it also does so at a fraction of the cost. The Wall Street Journal (WSJ) reported that DeepSeek claimed training one of its latest models cost approximately $5.6 million, compared to the $100 million to $1 billion range cited last year by Dario Amodei, the CEO of AI developer Anthropic. This cost-effectiveness highlights DeepSeek's innovative approach and its potential to disrupt the AI industry.
DeepSeek's novel approach to AI development has truly been groundbreaking. An AI expert I follow on X.com, Morgan Brown, noted that one reason DeepSeek can achieve lower costs is due to their thinking process—"They thought everything from the ground up. Traditional AI is like writing every number with 32 decimal places. DeepSeek was like, 'What if we use 8? It is still accurate enough!'" This out-of-the-box thinking resulted in a 75% reduction in memory usage, demonstrating the untapped potential of AI technology.
Furthermore, DeepSeek stated that R1 achieves its performance by utilizing less advanced chips from Nvidia, owing to U.S. government export controls on the most advanced chips from the company. As a result, Nvidia's stock experienced a significant decline on Monday, as anxious investors worried that demand for Nvidia's most advanced chips—which also have the highest profit margins—would drop if companies realized they could develop high-performance AI models with cheaper, less advanced chips.
I question DeepSeek’s assertion that it doesn't rely on the most advanced chips. There exists a robust underground network that successfully smuggles restricted Nvidia chips into China. As reported by the WSJ last July, more than 70 Chinese distributors openly market what they claim to be Nvidia's restricted chips online. The consistency of this supply is remarkable, with many sellers taking preorders and promising delivery in just a few weeks.
DeepSeek's R1 is disruptive not only because of its accessibility but also due to its free and open-source model. Users are empowered to access, use, and modify the source code at no cost. While they do pay a modest fee to connect their applications to DeepSeek, the overall low barrier to entry is significant. This disruption was clearly reflected in Monday’s stock market selloff, which affected nearly all major U.S. tech companies, from Meta to Microsoft. Investors are rightly concerned about how DeepSeek's model could challenge the established dominance of major American tech firms in the AI sector, from chip manufacturing to infrastructure, allowing for rapid and cost-effective development of new AI applications by users and businesses alike.
There is no doubt that DeepSeek is a remarkable technological advancement that will alter the competitive landscape between China and the U.S. However, despite its sophistication, the model has critical shortcomings. One of the most pressing concerns is data security and privacy, as it openly states that it will collect sensitive information such as users' keystroke patterns and rhythms.
Moreover, the issue of censorship cannot be ignored. DeepSeek has conceded that its programming and knowledge base are tailored to comply with China’s laws and regulations, as well as promote socialist core values. For instance, when a user inquired about Taiwan's president, they were met with the predictable rhetoric of the Chinese Communist Party, which maintains that "Taiwan is an inalienable part of China."
On Monday, I tweeted, “The U.S. stock market is oversold today. Investors should have the conviction that the country upholds free speech will win the tech race against the regime enforces censorship.” I didn't just express my opinion; I backed it up by purchasing several shares of Nvidia stock.
Comentários