DeepSeek: How China's 'AI heroes' overcame US curbs to stun Silicon Valley

When ChatGPT sparked a storm in the global artificial intelligence field, an inevitable question arose: Does this spell trouble for China, the United States' biggest tech competitor?

Two years later, a new AI model from China has changed the question: Can the U.S. stop China's innovation? There was a time when Beijing seemed to be fumbling in response to ChatGPT, which is unavailable in China. Baidu, the search engine giant, launched its chatbot "Ernie Bot," which was met with user dissatisfaction and ridicule. Subsequently, Tencent and ByteDance launched similar products, but they were seen as imitations of ChatGPT and of inferior quality.

The United States had been confident in its lead and hoped to maintain this advantage. Consequently, the Biden administration increased restrictions, banning the export of advanced chips and technology to China. This is why DeepSeek's release shocked Silicon Valley and the world. The company claims that its powerful model is much cheaper to develop than the billions of dollars that U.S. companies have spent on AI.

So, how did this little-known company—whose founder has been hailed as an "AI hero" on Chinese social media—achieve this? It was undoubtedly a blow when the U.S. banned leading global chip manufacturers like Nvidia from selling advanced technology to China. These chips are crucial for building powerful AI models that can perform a range of human tasks, from answering basic queries to solving complex mathematical problems. DeepSeek founder Liang Wenfeng stated in a local media interview that the chip ban was their "main challenge."

According to MIT Technology Review, DeepSeek acquired a "large number" of Nvidia A100 chips—estimated to be between 10,000 and 50,000—before the ban. Leading Western AI models are estimated to have used 16,000 specialized chips. But DeepSeek says it trained its AI model using 2,000 such chips and thousands of lower-grade chips, making its product cheaper. Some, including U.S. tech billionaire Elon Musk, have questioned this claim, arguing that the company could not disclose how many advanced chips it actually used under the restrictions.

But experts say Washington's ban has presented both challenges and opportunities for China's AI industry. Zhang Marina, an associate professor at the University of Technology Sydney, said it "forced Chinese companies like DeepSeek to innovate" to do more with fewer resources. "While these restrictions have presented challenges, they have also spurred creativity and resilience, which aligns with China's broader policy goal of achieving technological independence." The world's second-largest economy has invested heavily in large tech sectors—from batteries powering electric vehicles and solar panels to artificial intelligence.

Turning China into a technological powerhouse has long been President Xi Jinping's ambition, so Washington's restrictions also represent a challenge for Beijing. According to Gregory C. Allen, an AI expert at the Center for Strategic and International Studies, the release of DeepSeek's new model on January 20, the day Donald Trump was sworn in as U.S. president, was deliberate. "The timing and the way it was released—this is exactly what the Chinese government wants everyone to think—that export controls don't work, and the U.S. is not the global leader in AI," said Mr. Allen, who was formerly the director of strategy and policy at the U.S. Department of Defense's Joint Artificial Intelligence Center.

In recent years, the Chinese government has been cultivating AI talent, offering scholarships and research grants, and encouraging collaboration between universities and industry. Ms. Zhang said that the National Deep Learning Engineering Laboratory and other state-supported initiatives have helped train thousands of AI experts. Moreover, China has a large pool of talented engineers to recruit from.

Take DeepSeek's team as an example—Chinese media reports that the team consists of fewer than 140 people, most of whom are "homegrown talent" from China's top universities, as the internet proudly proclaims. Ms. Zhang said that Western observers have missed "the emergence of a new generation of entrepreneurs who prioritize basic research and long-term technological advancement over quick profits." China's top universities are creating a "rapidly growing pool of AI talent," and even managers tend to be under 35. "They grew up during China's rapid technological rise, and they are deeply driven by innovation and self-reliance," she added.

Deepseek's founder, Liang Wenfeng, is one such example—the 40-year-old studied AI at the prestigious Zhejiang University. In an article by tech media 36Kr, people familiar with him said he was "more of a geek than a boss." Chinese media calls him a "technology idealist"—he insists on keeping DeepSeek an open-source platform. Indeed, experts also believe that the thriving open-source culture enables young startups to pool resources and progress faster. Unlike larger Chinese tech companies, DeepSeek prioritizes research, which allows for more experimentation, according to experts and people who have worked at the company.

"The top 50 talents in this field may not be in China, but we can cultivate such people here," Mr. Liang said in an interview with 36Kr. But experts wonder how far DeepSeek can go. Ms. Zhang said that "new U.S. restrictions may limit access to U.S. user data, which may affect how Chinese models like DeepSeek go global." Others say the U.S. still has a huge advantage, for example, in Mr. Allen's words, "they have a massive amount of computing resources"—and it's unclear how DeepSeek will continue to use advanced chips to continuously improve its model.

But for now, DeepSeek is enjoying its moment in the spotlight, as most people in China had not heard of it until this weekend. His sudden fame has made Mr. Liang a sensation on Chinese social media, where he is hailed as one of "three AI heroes" from Guangdong province in southern Hong Kong. The other two are Yang Zhilin, a renowned expert from Tsinghua University, and He Kaiming, who teaches at MIT in the U.S. DeepSeek has delighted the Chinese internet ahead of the Lunar New Year, China's biggest holiday. This is good news for a struggling economy and a tech sector bracing for further tariffs and a possible sale of TikTok's U.S. operations.

One of the most popular Weibo comments read, "DeepSeek tells us that only true ability can stand the test of time." Another wrote, "This is the best New Year's gift. Wishing our motherland prosperity and strength." Wei Sun, chief AI analyst at Counterpoint Research, described the Chinese reaction as a "mixture of shock and excitement, especially in the open-source community." Zhou Fiona, a tech worker in the southern city of Shenzhen, said her social media feed was "suddenly flooded with DeepSeek-related posts yesterday."

"People are calling it 'China's glory' and saying it has shocked Silicon Valley, so I downloaded it to see how good it is." She asked it to give her "four pillars of destiny," or Bazi—like a personalized horoscope based on date and time of birth. But to her disappointment, DeepSeek got it wrong. While she got a detailed explanation of its "thought process," it wasn't the "four pillars" of her actual Bazi. She said she would still try it again at work, as it might be more useful for those kinds of tasks.