Hong Kong China’s artificial intelligence lab has emerged as the most recent front in the U.S.-China rivalry, casting uncertainty on how much and how long the United States will dominate the development of the strategically important technology.
After releasing AI models that experts claim match or surpass top American models at a tenth of the price, the Chinese startup DeepSeekhas caused a stir. U.S. investors were alarmed by the previously unknown company’s explosive growth on Monday, which caused the largest decline in the history of the American market, wiping away about $600 billion in the market value of the chipmaker Nvidia.
DeepSeek is being hailed in China as a representation of the nation’s AI breakthroughs in the face of U.S. export restrictions that Beijing claims are intended to stifle its technical progress.
According to Feng Ji, creator and CEO of Game Science, the firm behind the popular video game Black Myth: Wukong, this astounding innovation has come from a company that is entirely Chinese.
In a post on the Chinese social media site Weibo, he said that DeepSeek might be a scientific and technological triumph at the national level.
China would definitely win the U.S.-China AI race, according to Zhou Hongyi, co-founder of the Chinese cybersecurity company Qihoo 360.
“DeepSeek will undoubtedly be a key member of China’s Avengers team if we are to counter America’s AI tech dominance,” he stated in a Weibo video.
How, therefore, could a little-known firm turn into a major force in AI?
Founded in 2023 by Liang Wenfeng, the creator of the AI-driven trading strategy hedge fund High-Flyer, DeepSeek is headquartered in Hangzhou, a Chinese innovation powerhouse.
According to Liang, High-Flyer was a source of some of DeepSeek’s initial staff members and an investor, however it’s unclear how much it contributed.
He told the Chinese media site 36Kr in May 2023 that the company was seeking people who were curious and down to earth, adding that our core technical roles are typically filled by recent graduates or those with only a year or two of work experience.
The Chinese social media site WeChat posted in July that High-Flyer had purchased 10,000 of Nvidia’s high-performance A100 graphics processor chips by 2022. Later that year, the United States restricted sales of such chips to China.
Liang informed 36Kr that he had no specific business objective in mind when he purchased the chips, primarily out of curiosity regarding the limits of AI capabilities.
A week after DeepSeek released an open-source language model named R1 on January 20, which was widely praised in what venture capitalist Marc Andreessen dubbed AI’s Sputnik moment, the stock market meltdown occurred.
Earlier in the week, DeepSeek released a technical study that further alarmed investors and the U.S. IT industry by claiming that the model was built in just two months and cost less than $6 million, in contrast to the billions spent by top U.S. tech companies.
DeepSeek’s newly published software shot to the top of the Apple software Store on Monday due to the spike in interest. Later, the business announced that it was temporarily restricting user registrations because of widespread malicious attacks on its services, according to CNBC.
The 1989 Tiananmen Square crackdown, the self-ruled democratic island of Taiwan, which Beijing claims is part of its territory, and the use of Chinese government language were among the topics around which R1 seemed to be censored, as new users quickly pointed out. However, if the model is utilized on servers located outside of mainland China rather than via the DeepSeek website, it seems to function without these limitations.
A request for comment on Tuesday was not answered by DeepSeek.
According to an impartial rating of AI analysis, the R1 model is currently ranked second only behind California-based OpenAI’s o1. Additionally, it outperforms top models from Google, Meta, and Anthropic, a California company.
Liang, who is reportedly around 40 years old, has maintained a low profile in China, where the IT sector has been subject to a crackdown in recent years due to the ruling Chinese Communist Party’s fears that its largest corporations and executives may be growing too powerful.
However, he made an appearance on state television last week during a high-profile meeting with China’s No. 2 official, Premier Li Qiang, who asked Liang and other specialists in science, technology, education, and other areas to provide their thoughts on a draft government work report. According to official media, Li urged technical innovation during the conference to boost the economy.
With the aim of dominating the world in AI technology by 2030, China has made AI a national priority. Concerned about the possible military uses, the United States has taken steps to restrict China’s access to American technology, including additional limitations on AI chips that Joe Biden imposed in the latter days of his presidency.
The PresidentDonald Trump stated on Monday that American tech businesses should take note of DeepSeek’s ascent.
A request for comment was forwarded to the appropriate department by the Chinese Foreign Ministry on Tuesday.
R1 has also gained interest because, in contrast to OpenAI’s o1, it is open-source and free to use, allowing anybody to examine and replicate its development process.
According to Kevin Xu, the U.S.-based founder of Interconnected Capital, a hedge fund that invests in artificial intelligence, the majority of Chinese engineers are eager for their open-source projects to be used by foreign companies, particularly those in Silicon Valley, in part because no one in the West respects what they do because everything in China is stolen or created by cheating.
Additionally, they know that Chinese companies have been using a lot of open-source technology for free in order to progress. “But they want to create their own, contribute, and prove that their tech is good enough to be taken for free by foreign firms,” Xu, the former senior director for global expansion at GitHub, stated in a post on X on Monday. “Some engineering pride, some nationalism.”
A number of Chinese AI startups, including DeepSeek, claim to be able to train models for a tenth of the cost. According to Kai-Fu Lee, the creator of 01.AI, his company had to innovate because of U.S. limits on chip exports, which cost $3 million to train its model.
At a seminar in Saudi Arabia in October, Lee, a former president of Google China, stated that Chinese companies can observe the breakthroughs, create some of their own, and then perform better engineering.
It is not necessary to spend a billion dollars to train a terrific model when you do outstanding, precise engineering.