As the United States marks the 250th anniversary of its founding, it confronts a new world order dominated by its relationship with China. In this wide-ranging series, we examine the pressure points and possibilities in those ties, from hard tech to soft power. Here, Vincent Chow looks at how China challenges core American assumptions about innovation and technology, and the historical stakes of their competition in artificial intelligence.
In 1969, the renowned British sinologist Joseph Needham posed a series of questions about China so influential that they became known as the “Needham Question”.
At its core, Needham was trying to reconcile China’s past glories with what he saw in the China of the mid-20th century.
Namely, why did the 18th-century Industrial Revolution happen in Europe and not China, a country that had led the world in technological innovation up until the 15th century, and what were the reasons for technology and science in China stagnating for centuries thereafter?
Like Needham, Erik Baark – professor emeritus at the Hong Kong University of Science and Technology – also spent much of his decades-long career researching Chinese innovation and technology. But at a guest lecture at the Hong Kong Baptist University last month, Baark proposed a fundamentally different question – an inversion of the original.
“The central question is now this,” said Baark, author of Innovation and China’s Global Emergence. “Why is China now able to contribute to global science and innovation? More than that, why does there even seem to be an alternative dynamism in Chinese innovation?”
For decades, the West questioned China’s ability to innovate and to invent truly original technologies. China can only copy Western technology or be reliant on fat government subsidies, the critics claimed.
But this dismissive attitude has seen a clear shift in recent years as China has established itself as a global leader in a range of cutting-edge technologies, from electric vehicles to humanoid robots.
“The biggest risk to AI is China getting ahead of us,” US Treasury Secretary Scott Bessent said at an event last week, underscoring the magnitude of the threat the Trump administration sees in China.
Chinese-American writer Kaiser Kuo called this “the great reckoning” in a viral essay in October. The term reflects a collective “recognition” in the West of China’s technological advances, even spawning a growing tourism industry where foreign investors and policymakers go on factory tours in China to see first-hand just how advanced the country has become.
“This reckoning touches all of humanity,” Kuo wrote. “[B]ut it falls ... hardest on the United States, where assumptions about exceptionalism and hierarchy are most exposed and most fiercely denied.”
In other words, it is the US today that is trying to reconcile the past and the present – with one particular technology at the centre of its worries.
Revolving revolutions
For years now, artificial intelligence has been compared with the Industrial Revolution in its expected impact and importance.
German economist Klaus Schwab called AI the “fourth” industrial revolution in a 2015 book, while the release of ChatGPT by US company OpenAI in late 2022 has even spurred claims that AI has already ushered in the fifth such revolution.
Successive US administrations have endorsed such claims. Jake Sullivan, National Security Adviser during the Biden administration, said that “AI will define the future”.
More recently, the Trump administration compared AI’s importance with the Industrial Revolution, in that it “presents a profound economic inflection point with the potential to significantly increase the GDP of countries that embrace it”.
For the US, the civilisational stakes of its AI competition with China have become undeniable. For the first time since its founding 250 years ago, the US today is undergoing simultaneous technological, political and geopolitical revolutions, Kyle Chan, a fellow at the Brookings Institution, said.
Back then, the world was dominated by European empires such as Britain and France. To break away from the British Crown, American revolutionaries launched their independence campaign in 1765, just as the first stirrings of the Industrial Revolution were emerging in Great Britain with the rise of steam engines and other innovations that would transform the world from human-driven economies to ones powered by machines.
The US once again finds itself in the midst of overlapping revolutions, confronting technological, political and geopolitical revolutions all at the same time with the rise of AI, authoritarianism at home and abroad, and China.
“The US made the most of the Industrial Revolution, which transformed it into the world’s only superpower,” Chan said. “This time, in the age of AI, it is unclear whether the US or China will make the most of this new technological wave.”
What is innovation?
Given the high stakes, both the US and China have obsessed over their AI capabilities in relation to each other, with the strength of their respective frontier foundational models drawing the bulk of the attention.
The gap was most apparent in late 2022 upon the release of ChatGPT. The development caught China by surprise, with leading Chinese experts estimating at the time that the gap was as large as three years.
But China has caught up rapidly since then. The first breakthrough was Hangzhou start-up DeepSeek’s release of R1 in January 2025, which stunned the world due to its low reported research and development cost. Almost overnight, the gap was no longer years but months.
Another major moment occurred last month with the release of GLM-5.2 by Beijing-based Zhipu AI, known as Z.ai internationally. The model showed powerful capabilities around the same time as the Trump administration curbed the roll-out of frontier US models for national security reasons, underscoring to many the alternative path that Chinese models represent.
According to research institute Epoch AI, Chinese models were more than a year behind their US counterparts as recently as May 2024. This has narrowed to as little as four months, it said. Nathan Lambert, author of the AI newsletter Interconnects, puts GLM-5.2 around seven months behind the US frontier.
China’s achievements in AI models reflect the country’s rapid advances in innovative capacity more broadly, Baark said. Last year, the country broke into the top 10 most innovative economies in the world on the World Intellectual Property Organisation’s Global Innovation Index, capping off a rapid ascent from 35th in 2013.
This came on the back of an explosion in China’s R&D spending.
A March report from the Organisation for Economic Co-operation and Development (OECD) found that China’s R&D investment in 2024 surpassed the US once purchasing power measures were factored in, soaring to 3.93 trillion yuan (US$551 billion) last year, according to official figures.
“China was very behind decades ago in innovation, but it has caught up very fast in the past few decades,” Dong Jielin, a former researcher at Tsinghua University’s China Institute for Science and Technology Policy and one of China’s leading experts on the history of technology and innovation, said.
For decades, Dong studied the innovation ecosystems of both the US and China through roles in academia and the private sector. But her rigorous approach meant that, while acknowledging China’s undeniable advances in innovation, she still believed China had yet to master the “zero-to-one” innovation.
In the AI community, a common term researchers use to refer to such innovations is “paradigms”, meaning a technical breakthrough that lays the groundwork for an entirely new field of inquiry and innovation.
Dong also emphasises this agenda-setting quality as central to the innovation that matters most to her.
Though Chinese models have narrowed the gap with their US counterparts, all the leading ones, including DeepSeek R1 and GLM-5.2, are still trained on advanced US semiconductor chips despite years of efforts by China to achieve self-reliance across the entire AI stack.
Both DeepSeek R1 and GLM-5.2 are also the type of AI models – large language models with strong coding ability – that were first pioneered by US start-ups OpenAI and Anthropic.
“Neither DeepSeek nor Zhipu really represent zero-to-one innovation because they improve upon US inventions through better engineering,” said Dong, the author of A Brief History of Human Technological Innovation. “Maybe China might eventually solve the problem of zero to one, but it’s not there yet.”
Different races
But not everyone agrees with this definition of innovation. For Baark, the HKUST emeritus professor, the challenge posed by China to the US is precisely its alternative approach to technological innovation.
In the realm of AI models, this is most apparent in two ways. First, while US models still lead in absolute capability, Chinese models lead in another key dimension – cost-efficiency.
Over the past year, third-party benchmark firm Artificial Analysis has consistently found that Chinese models are global leaders – beating their US counterparts – in maximising the amount of “intelligence” users can get per dollar spent. This increasingly matters as the dawn of agentic AI forces users and enterprises to seek ways to keep AI costs down.
Secondly, most frontier Chinese models are released as open-weight models, meaning that their weights – the parameters encoding their “intelligence” – are made freely available for developers and enterprises globally to use and customise for their own use cases.
These cost advantages have made Chinese models a prime target for US accusations that they are achieved through the “theft” of US models, with Anthropic and numerous US politicians all increasingly vocal in levelling the charge.
The claim is that Chinese models are entirely reliant on the “distillation” of US models, in which Chinese models are illicitly trained on US model outputs.
For Baark, the irony is that the US itself engaged in “industrial espionage” throughout much of its early years more than two centuries ago. Back then, it was the British who saw the strategic value of its new inventions and sought to make sure it would not fall into the hands of the young US, the rising power at the time.
Famously, the first US treasury secretary, Alexander Hamilton, coordinated an elaborate espionage campaign to steal and bring these technologies to the US.
“The US itself went through this catching-up process during the Industrial Revolution that China today is in,” Baark said.
“I wish they could therefore put some of their security worries aside and see the emergence of other centres of innovation as something they can actually benefit from.”
The cries of foul play also increasingly ring hollow in the Global South, where customers and firms are increasingly priced out of accessing frontier US models at premium prices.
Vince Iswara, CEO of DANA, one of Indonesia’s biggest digital wallet providers with more than 200 million users, said the company’s growing integration of Chinese open-weight models had led to a 50 per cent gain in company productivity while helping to keep R&D costs down.
Innovation under heavy cost and technological constraints is a daily reality in emerging markets. Iswara cites China’s digital payments revolution as typical of this model of innovation, where mass adoption of e-wallets was achieved through the proliferation of QR codes, in contrast to a Western payment model revolving around credit cards, which require expensive equipment such as card machines and ATMs.
Such constraints will not be unfamiliar to Chinese AI model developers. According to Epoch AI, the US has five times more AI supercomputers and three times better cutting-edge semiconductor chips than China.
Long-standing US export controls have restricted Chinese access to advanced US semiconductor chips from the likes of Nvidia, delivering a significant compute advantage for their US rivals.
But still, even with these difficulties, China has managed to remain competitive at the frontier of AI. For countries with even fewer resources, that in itself is a model of innovation.
“The China model is much more replicable for mass adoption [of new technologies] in emerging markets,” Iswara said. “There are a lot of things to learn from China.” -- SOUTH CHINA MORNING POST
