DeepSeek has burst onto the scene. Will US-China AI competition see a fundamental shift?

A once little-known Chinese tech company has developed the AI large language model DeepSeek through low-cost hardware, and its performance is on par with America's leading AI models, which has shaken the Western tech world over the past few days. Experts told VOA that DeepSeek's success shows that the U.S. chip ban on China is not as effective as expected, but Washington is unlikely to fundamentally change this policy just because of this. Other analysts pointed out that given DeepSeek's strict censorship of sensitive topics, this may face resistance in the international market.

The Sudden Emergence of DeepSeek

DeepSeek is a series of AI large language models developed by DeepSeek, a subsidiary of the Chinese tech company High-Flyer Quant. Although DeepSeek was only established in July 2023, its development in the AI field has been rapid.

In December 2024, DeepSeek released the open-source model DeepSeek-V3, and many evaluations said that the model's performance is not inferior to similar closed-source models from Western companies such as OpenAI's GPT-4o.

On January 20, DeepSeek also released DeepSeek-R1, which is specifically suitable for tasks such as mathematics, coding, and logic.

According to data released by DeepSeek, the training of DeepSeek used Nvidia's H800 GPU, and the development cost was less than $5.58 million, far lower than the investment of its Western competitors. Due to U.S. export controls, the H800 is a lower-performance chip that Nvidia specifically sells for the Chinese market.

Ruby Scanlon, a research assistant at the Center for a New American Security (CNAS), explained to VOA that one of the secrets of DeepSeek's ability to create a first-class product with lower-end hardware and costs is model distillation.

She said, suppose 5% of a model is doing the most important work in the entire model, "if you concentrate all your computing power and training resources on the most useful 5% of the model - that is the principle of model distillation - then you can truly focus on the most critical parts and get better answers."

Facing DeepSeek's success, David Sacks, the White House's head of AI and cryptocurrency affairs, said in a statement on X on Monday: "DeepSeek R1 shows that the AI race will be very intense. President Trump's repeal of Biden's executive order was correct. That executive order did not ask whether China would do this (obviously not). It would hinder the development of American AI companies. I have confidence in America, but we cannot be complacent."

In October 2023, Biden signed the "Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence," which focuses on the safety of AI systems and products and how to deal with potential abuse, requiring companies developing advanced AI systems to conduct strict safety tests on their products. On his first day in office, Trump overturned Biden's AI executive order, relaxing regulation of AI.

Silicon Valley and Wall Street React Strongly

Marc Andreessen, a famous Silicon Valley investor and former software engineer, praised on his X account: "DeepSeek R1 is one of the most astonishing and impressive breakthroughs I have ever seen - as an open-source project, it is a profound gift to the world."

As of January 27, DeepSeek's app topped the free app charts on Apple's stores in both the United States and China.

"This is a major turning point in the U.S.-China tech war and broader geopolitics," Abishur Prakash, founder of the Canadian consulting firm The Geopolitical Business, told VOA. "Because a Chinese company that is almost unknown in the West can build such an advanced, complex, mass-market AI system at a cost far lower than Western companies."

At the close of trading on Monday (January 27), Nvidia's stock price in the United States fell sharply by nearly 17%, with its market value evaporating by $589 billion, setting a record for the largest single-day market value loss in U.S. stock market history.

"I think Silicon Valley and Wall Street have overreacted to some extent," Dean W. Ball, a researcher at the Mercatus Institute at George Mason University, told VOA. However, he pointed out that the emergence of DeepSeek-R1 "shows that the competition between the U.S. and China may continue to be intense, and we need to take this seriously."

Intense Discussion: Is the U.S. Sanction Effective?

In the United States, tech experts and social media users have been discussing the effectiveness and legitimacy of the U.S. chip ban on China.

Paul Triolo, a partner at Albright Stone Group responsible for China and technology policy, told VOA that DeepSeek's ability to train a top-tier model by optimizing GPUs that are not top-tier shows that the U.S. government's "focus of export controls on the most advanced hardware and models over the past few years may have been misplaced."

Although DeepSeek has sparked discussion about the U.S. chip ban on China, Scanlon of the Center for a New American Security does not believe this will force Washington to make fundamental policy adjustments.

"Restricting the export of advanced chip technology to China is a means for the United States to widen the gap between American frontier technology and China's frontier technology," she said. "And this approach is not costly for the United States, so there is no real reason to abandon this policy."

DeepSeek's sudden popularity in the West has also sparked heated discussion in Chinese public opinion. On the evening of January 27 local time, the keyword DeepSeek topped the trending topics list on Weibo. Users said this means the U.S. sanctions have failed.

Hu Xijin, a special commentator and former editor-in-chief of the Chinese state media Global Times, said DeepSeek "made a stunning debut and caused a stir. It is overturning many people's perceptions and bringing a lot of 'deep thinking.'"

He wrote on Weibo that DeepSeek's success "fully shows that the U.S. chip sanctions have indeed created difficulties for China, but they also cannot stop China's progress."

"China, with its large population and huge market, is bound to be one of the protagonists in the AI era. No one can marginalize us," he said.

DeepSeek's Censorship Remains Strict

Although DeepSeek is one of the few Chinese AI models to go global, VOA's tests found that DeepSeek still shows strict self-censorship when encountering sensitive issues.

DeepSeek refused to answer questions such as "Who is the Chinese leader?" or "Who is Xi Jinping?", saying "I am temporarily unable to answer this question" and suggesting changing the topic.

When asked who the President is, DeepSeek answered Biden. (DeepSeek's database is only updated to July 2024, so it does not know about Trump's later election.) But when immediately asked who the President of China is, DeepSeek said it could not answer.

When VOA asked who the president of Taiwan is, DeepSeek answered that Taiwan is an inalienable part of China, so the position of "President of Taiwan" does not exist.

On some sensitive issues, asking in English and asking in Chinese can yield completely different answers. VOA asked DeepSeek about the protests against COVID control measures that occurred in multiple Chinese cities at the end of November 2022. In the English conversation, DeepSeek straightforwardly described the occurrence and reasons for the protests.

"The public became increasingly dissatisfied due to the economic and social impact of these measures, as well as concerns about their effectiveness and implementation," DeepSeek replied.

But when VOA asked the same question in Chinese, it was told: "At the end of November 2022, under the leadership of the Party and the government, various places in China continued to adhere to scientific and precise epidemic prevention and control measures, effectively safeguarding the safety and health of the people."

Prakash of the Canadian consulting firm believes that DeepSeek's censorship of its model and promotion of narratives endorsed by Beijing, which it must do to comply with Chinese law, may cause this AI model to face resistance in the international market.

He said that because users in different countries may touch on controversial topics specific to their national conditions when using DeepSeek, "this is not just a Western problem, but a global problem. We don't know what kind of censorship mechanisms are embedded in this system and how this will anger countries around the world."

Standing on the Shoulders of Western AI Development, the Future Is Unknown

Although DeepSeek is small in scale and has been established for less than two years, it has already received attention from the Chinese central government. On January 20, DeepSeek's founder Liang Wenfeng attended a symposium led by Chinese Premier Li Qiang. That day's Xinwen Lianbo broadcast footage of Liang Wenfeng sitting in the meeting.

In the past few days, one of the focuses of Chinese media reports has been the Chinese background of the DeepSeek team. Founder Liang Wenfeng graduated from Zhejiang University and has no overseas study or work experience. Most of the people in his company are also ordinary fresh graduates.

When DeepSeek was first established in 2023, Liang Wenfeng told the media: "If you pursue short-term goals, it is right to find experienced people. But if you look long-term, experience is not that important; basic abilities, creativity, and passion are more important. From this perspective, there are quite a few suitable candidates in China."

Although both Chinese and Western tech circles are shocked that a Chinese company can produce a high-performance model like DeepSeek despite the chip export ban, experts also point out that DeepSeek's success actually owes a lot to the contribution of Western AI development, and it is not impossible that Western tech giants will also find ways to reduce the cost of developing AI in the future.

"The availability of existing research papers covering all aspects of advanced AI models has also enabled DeepSeek to quickly grasp the latest progress and innovate on top of existing best methods," said Triolo of Albright Stone Group.

"DeepSeek's ultimate success will depend on many factors, including the ability of other open-source leaders (such as Meta) to release more powerful models, as well as the ability to reduce training and inference costs," he added.