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

When ChatGPT stormed the world of artificial intelligence (AI), an inevitable question followed: did it spell trouble for China, America's biggest tech rival?

Two years on, a new AI model from China has flipped that question: can the US stop Chinese innovation?

For a while, Beijing seemed to fumble with its answer to ChatGPT, which is not available in China.

Unimpressed users mocked Ernie, the chatbot by search engine giant Baidu. Then came versions by tech firms Tencent and ByteDance, which were dismissed as followers of ChatGPT - but not as good.

Washington was confident that it was ahead and wanted to keep it that way. So the Biden administration ramped up restrictions banning the export of advanced chips and technology to China.

That's why DeepSeek's launch has astonished Silicon Valley and the world. The firm says its powerful model is far cheaper than the billions US firms have spent on AI.

So how did a little-known company - whose founder is being hailed on Chinese social media as an "AI hero" - pull this off?

The challenge

When the US barred the world's leading chip-makers such as Nvidia from selling advanced tech to China, it was certainly a blow.

Those chips are essential for building powerful AI models that can perform a range of human tasks, from answering basic queries to solving complex maths problems.

DeepSeek's founder Liang Wenfeng described the chip ban as their "main challenge" in interviews with local media.

Long before the ban, DeepSeek acquired a "substantial stockpile" of Nvidia A100 chips - estimates range from 10,000 to 50,000 - according to the MIT Technology Review.

Leading AI models in the West use an estimated 16,000 specialised chips. But DeepSeek says it trained its AI model using 2,000 such chips, and thousands of lower-grade chips - which is what makes its product cheaper.

Some, including US tech billionaire Elon Musk, have questioned this claim, arguing the company cannot reveal how many advanced chips it really used given the restrictions.

But experts say Washington's ban brought both challenges and opportunities to the Chinese AI industry.

It has "forced Chinese companies like DeepSeek to innovate" so they can do more with less, says Marina Zhang, an associate professor at the University of Technology Sydney.

DeepSeek's founder Liang Wenfung (R) at a recent government meeting

"While these restrictions pose challenges, they have also spurred creativity and resilience, aligning with China's broader policy goals of achieving technological independence."

The world's second-largest economy has invested heavily in big tech - from the batteries that power electric vehicles and solar panels, to AI.

Turning China into a tech superpower has long been President Xi Jinping's ambition, so Washington's restrictions were also a challenge that Beijing took on.

The release of DeepSeek's new model on 20 January, when Donald Trump was sworn in as US president, was deliberate, according to Gregory C Allen, an AI expert at the Center for Strategic and International Studies.

"The timing and the way it's being messaged - that's exactly what the Chinese government wants everybody to think - that export controls don't work and that America is not the global leader in AI," says Mr Allen, former director of strategy and policy at the US Department of Defense Joint Artificial Intelligence Center.

In recent years the Chinese government has nurtured AI talent, offering scholarships and research grants, and encouraging partnerships between universities and industry.

The National Engineering Laboratory for Deep Learning and other state-backed initiatives have helped train thousands of AI specialists, according to Ms Zhang.

And China had plenty of bright engineers to recruit.

The talent

Take DeepSeek's team for instance - Chinese media says it comprises fewer than 140 people, most of whom are what the internet has proudly declared as "home-grown talent" from elite Chinese universities.

Western observers missed the emergence of "a new generation of entrepreneurs who prioritise foundational research and long-term technological advancement over quick profits", Ms Zhang says.

China's top universities are creating a "rapidly growing AI talent pool" where even managers are often under the age of 35.

"Having grown up during China's rapid technological ascent, they are deeply motivated by a drive for self-reliance in innovation," she adds.

Deepseek's founder Liang Wenfeng is an example of this - the 40-year-old studied AI at the prestigious Zhejiang University. In an article on the tech outlet 36Kr, people familiar with him say he is "more like a geek rather than a boss".

And Chinese media describe him as a "technical idealist" - he insists on keeping DeepSeek as an open-source platform. In fact experts also believe a thriving open-source culture has allowed young start-ups to pool resources and advance faster.

Unlike bigger Chinese tech firms, DeepSeek prioritised research, which has allowed for more experimenting, according to experts and people who worked at the company.

"The Top 50 talents in this field might not be in China, but we can build people like that here," Mr Liang said in an interview with 36Kr.

But experts wonder how much further DeepSeek can go. Ms Zhang says that "new US restrictions may limit access to American user data, potentially impacting how Chinese models like DeepSeek can go global".

And others say the US still has a huge advantage, such as, in Mr Allen's words, "their enormous quantity of computing resources" - and it's also unclear how DeepSeek will continue using advanced chips to keep improving the model.

But for now, DeepSeek is enjoying its moment in the sun, given that most people in China had never heard of it until this weekend.

The new AI heroes

His sudden fame has seen Mr Liang become a sensation on China's social media, where he is being applauded as one of the "three AI heroes" from southern Guangdong province, which borders Hong Kong.

The other two are Zhilin Yang, a leading expert at Tsinghua University, and Kaiming He, who teaches at MIT in the US.

DeepSeek has delighted the Chinese internet ahead of Lunar New Year, the country's biggest holiday. It's good news for a beleaguered economy and a tech industry that is bracing for further tariffs and the possible sale of TikTok's US business.

"DeepSeek shows us that only if you have the real deal will you stand the test of time," a top-liked Weibo comment reads.

"This is the best new year gift. Wish our motherland prosperous and strong," another reads.

A "blend of shock and excitement, particularly within the open-source community," is how Wei Sun, principal AI analyst at Counterpoint Research, described the reaction in China.

DeepSeek's success has been cheered in China during its biggest holiday

Fiona Zhou, a tech worker in the southern city of Shenzhen, says her social media feed "was suddenly flooded with DeepSeek-related posts yesterday".

"People call it 'the glory of made-in-China', and say it shocked Silicon Valley, so I downloaded it to see how good it is."

She asked it for "four pillars of [her] destiny", or ba-zi - like a personalised horoscope that is based on the date and time of birth.

But to her disappointment, DeepSeek was wrong. While she was given a thorough explanation about its "thinking process", it was not the "four pillars" from her real ba-zi.

She says she will still give it another go at work, as it will probably be more useful for such tasks.