HONG KONG — An artificial intelligence lab in China has become the latest front in the U.S.-China rivalry, raising doubts as to how much — and for how much longer — the United States is in the lead in developing the strategically key technology.

The Chinese startup DeepSeek has made waves after releasing AI models that experts say match or outperform leading American models at a fraction of the cost. The meteoric rise of the previously little-known company spooked U.S. investors Monday, wiping out almost $600 billion in the market value of American chipmaker Nvidia in the biggest drop in the nation’s market history.

In China, DeepSeek is being heralded as a symbol of the country’s AI advancements in the face of U.S. export restrictions that Beijing says are aimed at suppressing its technological development.

“This jaw-dropping breakthrough has come from a purely Chinese company,” said Feng Ji, founder and chief executive of Game Science, the developer behind the hit video game Black Myth: Wukong.

“DeepSeek may be a national-level technological and scientific achievement,” he wrote in a post on the Chinese social media platform Weibo.

Zhou Hongyi, co-founder of the Chinese cybersecurity firm Qihoo 360, said China would “undoubtedly come out on top” in the U.S.-China AI race.

“If we are to counter America’s AI tech dominance, DeepSeek will definitely be a key member of China’s ‘Avengers team,’” he said in a video on Weibo.

So how did a little-known startup become a global AI sensation?

Based in the Chinese tech hub of Hangzhou, DeepSeek was founded in 2023 by Liang Wenfeng, who is also the founder of a hedge fund called High-Flyer that uses AI-driven trading strategies.

Liang has said High-Flyer was one of DeepSeek’s investors, though it’s unclear how much it contributed, as well as a source of some of its first employees.

“Our core technical roles are mostly filled by fresh graduates or those with just a year or two of work experience,” he told the Chinese media outlet 36Kr in May 2023, adding that the company was looking for “down-to-earth individuals with curiosity.”

By 2022, High-Flyer had acquired 10,000 of Nvidia’s high-performance A100 graphics processor chips, according to a post that July on the Chinese social media platform WeChat. The U.S. imposed restrictions on sales of those chips to China later that year.

Liang told 36Kr that he acquired the chips mostly because of “curiosity about the boundaries of AI capabilities” and that he had no particular commercial goal in mind.

The stock market rout took place a week after DeepSeek’s Jan. 20 release of an open-source language model called R1, which met with widespread acclaim in what venture capitalist Marc Andreessen called “AI’s Sputnik moment.”

Investors and the U.S. tech industry were further unnerved later in the week when DeepSeek published a technical report that said the model took only two months and less than $6 million to build, compared with the billions spent by leading U.S. tech companies.

The surge in interest sent DeepSeek’s recently released app to the top of Apple’s App Store on Monday. The company later said that it was temporarily limiting user registrations “due to large-scale malicious attacks” on its services, CNBC reported.

New users were quick to note that R1 appeared subject to censorship around topics deemed sensitive in China, avoiding answering questions about the self-ruled democratic island of Taiwan, which Beijing claims is part of its territory, or the 1989 Tiananmen Square crackdown or echoing Chinese government language. The model appears to operate without such restrictions, however, if it is used not through the DeepSeek website but on servers that host it outside mainland China.

DeepSeek did not respond to a request for comment Tuesday.

The R1 model is now second only to California-based OpenAI’s o1 in the artificial analysis quality index, an independent AI analysis ranking. It also beats leading models by Google, Meta and the California-based Anthropic.

Liang, who according to the China’s media is about 40, has kept a relatively low profile in the country, where there has been a crackdown on the tech industry in recent years amid concerns by the ruling Chinese Communist Party that its biggest companies and executives might be getting too powerful.

But he appeared on state television last week during a high-profile meeting with Premier Li Qiang, China’s No. 2 official, who invited Liang and other experts from technology, education, science and other fields to share their opinions for a draft government work report. At the meeting, Li called for “technological innovation” to foster the economy, according to state media reports.

China has made AI a national priority, with the goal of becoming the global leader in its technology by 2030. The U.S., concerned about the potential military applications, has moved to limit China’s access to American technology, including new restrictions on AI chips issued by Joe Biden in the final days of his presidency.

President Donald Trump said Monday that DeepSeek’s rise “should be a wake-up call” for U.S. tech companies.

The Chinese Foreign Ministry referred a request for comment Tuesday to the relevant department.

R1 has also drawn attention because, unlike OpenAI’s o1, it is free to use and open-source, meaning anyone can study and copy how it was made.

Most Chinese engineers are eager for their open-source projects to be used by foreign companies, especially 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,” said Kevin Xu, the U.S.-based founder of Interconnected Capital, a hedge fund that invests in AI.

“They are also aware that Chinese firms have been taking for free lots of open-source tech to advance,” Xu, the former senior director for global expansion at GitHub, said Monday in a post on X, “but they want to create their own, contribute, and prove that their tech is good enough to be taken for free by foreign firms — some nationalism, some engineering pride.”

DeepSeek is not the only Chinese AI startup that says it can train models for a fraction of the price. Kai-Fu Lee, founder of 01.AI, has said that it cost $3 million to train its model and that U.S. restrictions on chip exports forced his company to innovate.

“The Chinese companies are able to watch the innovations, make some themselves and then do better engineering,” Lee, a former president of Google China, said at a conference in Saudi Arabia in October. 

“When you do excellent, detailed engineering, it is not the case you have to spend a billion dollars to train a great model.”