DeepSeek A.I. Is a Win for China, but a Danger to Party Control

In 2017, China watched in awe — and shock — as AlphaGo, an artificial intelligence program backed by Google, defeated a Chinese prodigy at a complex board game, Go. The decisive loss to a foreign computer program, which had similarly trounced a South Korean player, was a sort of Sputnik moment for China.

That year, Chinese officials laid out a bold plan to lead the world in A.I. by 2030, pledging billions to companies and researchers focused on the technology. From this fervor emerged DeepSeek, the largely unknown Chinese start-up that upended the technology landscape by creating a powerful A.I. model with far less money than experts had thought possible.

DeepSeek is private, with no apparent state backing, but its success embodies the ambitions of China’s top leader, Xi Jinping, who has exhorted his country to “occupy the commanding heights” of technology. Mr. Xi wants the Chinese economy to be powered not by old growth engines like debt-fueled real estate and cheap exports, but by the most advanced technologies like A.I., supercomputing and green energy.

For Mr. Xi, this moment helps dent the aura of superiority the United States has held in A.I., a critical field in a fierce superpower rivalry. China has cast itself as a benevolent global partner to developing countries, willing to share its know-how, with Mr. Xi saying that A.I. should not be a “game of rich countries and the wealthy.”

Now, DeepSeek has shown that it might be possible for China to make A.I. cheaper and more accessible for everyone. The question, though, is how the ruling Communist Party manages the rise of a technology that could one day be so disruptive that it could threaten its interests — and its grip on power.

Chinese regulation of A.I. has varied in intensity over the years, depending on where the country assesses its strengths and weaknesses. When the Chinese government was worried it had fallen behind the United States in 2022 after the launch of OpenAI’s ChatGPT, it took a more hands-off approach that ultimately allowed ventures like DeepSeek and others to thrive.

Now that the pendulum has swung the other way, that confidence in the industry could prove to be a “double-edged sword,” said Matt Sheehan, who studies Chinese A.I. as a fellow at the Carnegie Endowment for International Peace.

The party’s “core instincts are toward control,” Mr. Sheehan said. “As they regain confidence in China’s A.I. capabilities, they may have a hard time resisting the urge to take a more hands-on approach to these companies.”

As if to underscore that possibility, DeepSeek’s founder, Liang Wenfeng, was invited to a discussion with Premier Li Qiang on Jan. 20, the same day that the company released its latest and most powerful A.I. model, known as R1.

Mr. Liang’s attendance was all the more remarkable considering DeepSeek had not been considered one of China’s so-called A.I. Tigers. That distinction is reserved for high-profile firms like Zhipu AI, a Beijing-based start-up that has received substantial state investment.

DeepSeek is no stranger to the party’s urge to interfere; that may have inadvertently played a role in its eventual success. DeepSeek had originally trained its A.I. models to make bets on the Chinese stock market. But when regulators targeted such behavior, it pivoted in 2023 to advanced A.I. to conform with China’s industrial policy.

Then it stunned the world by rivaling the performance of its American competitors despite using far fewer of the advanced computer chips that are hard for China to obtain — a technological feat that until recently had not been available. At home, Chinese commentators have held up DeepSeek’s achievement as evidence that U.S. restrictions on exports of A.I. chips to China are ultimately futile (even though the company’s founder has said such limits are a major concern).

Even the recent allegations by OpenAI that DeepSeek improperly harvested its data to build its models have not deterred its fans in China, who accuse the San Francisco company of spreading rumors.

“The U.S. technological sanctions on China have left China with no choice but to develop,” said Sun Chenghao, a foreign relations expert at Tsinghua University in Beijing, echoing a popular sentiment in China. “We can only rely on ourselves.”

A.I. holds a special place in Mr. Xi’s vision of China’s rise, with its potential to help the country overcome many of its biggest challenges like its shrinking work force. China has used facial recognition and algorithms to supercharge its ability to surveil its people and snuff out dissent. The technology is also factoring into China’s military modernization with autonomous weapons systems and even battlefield strategy.

DeepSeek’s development could also advance China’s geopolitical goals. DeepSeek uses an open source model, meaning anyone can peer under its hood and use its technology, unlike leading American companies that use more expensive proprietary software.

“The low cost and open source nature of DeepSeek’s model bolsters the Chinese government’s narrative that China is the place developing countries can look to for A.I. solutions,” Mr. Sheehan said.

How big a player China becomes on the global stage in A.I. could ultimately depend on how the government decides to balance regulations with the freedom that companies and researchers need to do cutting-edge work that allows them to compete with the United States.

Some analysts like Gregory C. Allen, a researcher at the Center for Strategic and International Studies and a former U.S. defense official, said there were most likely no restraints on A.I. development when it comes to China’s military.

“The only thing holding them back is performance,” said Mr. Allen, who in his former job held talks with members of the People’s Liberation Army responsible for assessing the risks of A.I.

The same does not hold true for regulating A.I. in the private sector. The landscape there is dictated by the competing priorities of China’s regulatory agencies, each feeling their way around a technology that many in the world still do not fully understand.

It is clear that the more widely used a technology is, the more the party will want to rein it in. In 2023, just months after ChatGPT set off an investment frenzy over artificial intelligence, China issued rules aimed at controlling what Chinese chatbots say to users, requiring them to reflect “socialist core values” and avoid information that undermines “state power.”

In the case of DeepSeek’s chatbot, this has led to awkward responses to seemingly benign questions like, “Who is Xi Jinping?” Researchers testing its capabilities have found that the bot gives answers that spread Chinese propaganda and even parrot disinformation campaigns.

Some concerns are more existential in nature. A growing chorus of scholars have been sounding the alarm about the potentially catastrophic consequences of losing human control over A.I.

Chief among those voices has been Andrew Yao, a giant in A.I. at Tsinghua University and a recipient of the Turing Award, the equivalent of the Nobel Prize for computing. His influence helped establish what China calls the Global AI Governance Initiative, which was introduced by Mr. Xi in 2023 and included a call to always keep A.I. under human control. Last year, the government also called for the enhancement of A.I. governance “on the basis of human decision-making and supervision.”

Ultimately, A.I. in China may only advance as far as the government decides it can mitigate those risks, said Barath Harithas, an expert on A.I. policy at the Center for Strategic and International Studies, a Washington think tank.

“Overregulation and the need to adhere to ‘core socialist values’ could risk neutering A.I.’s potential,” Mr. Harithas said.