According to Erik Baark, professor emeritus at the Hong Kong University of Science and Technology, the rise of China as a technological peer rival to the US in recent decades is so deeply unsettling for the US because it challenges longstanding assumptions surrounding US exceptionalism.
Last month, at a guest lecture at the Hong Kong Baptist University, Baark proposed a modern-day equivalent of the 20th century Needham Question. While the original question proposed by the British sinologist focused on why Chinese technology and innovation had stagnated for centuries, Baark’s version suggested that the defining question of today is precisely an inversion of the original:
Why is China now able to contribute to global technological innovation and even represent an “alternative dynamism” to the innovation model espoused by the US?
There is no better evidence of this than in the realm of frontier AI models – the area most obsessed over by successive US administrations, which have introduced elaborate export control measures covering advanced semiconductor chips and chipmaking equipment precisely to extend and entrench its lead over China in frontier AI.
Yet China has proven itself difficult to defeat. Nearly a year and a half after DeepSeek’s first breakthrough cut the gap between the two countries from years to months almost overnight, another model, this time from Beijing-based Zhipu AI, has once again sparked discussions of another “DeepSeek moment”.
That’s because GLM-5.2, released in mid-June, is not just the best Chinese model to date but also the first Chinese model to rank in the top three globally on major industry benchmarks, edging out US giant Google and on par with leading models from the likes of OpenAI and Anthropic.
Praise of the model has also accumulated as developers and industry peers began to try the model themselves. Matt Velloso, a former vice-president at Meta Platforms and Google DeepMind, said on X that “things are not going to be the same” after concluding that GLM-5.2 was the “first open model that passes the bar as a daily driver”. AI expert Nathan Lambert wrote on his influential newsletter Interconnects that GLM-5.2 was truly a “step change” for the industry.
But even setting aside the benchmark scores, GLM-5.2 is indicative of China’s “alternative dynamism” that Baark highlighted. That’s because the model, like so many frontier Chinese models before it, continues to be much cheaper than its US counterparts and, crucially, open-sourced, meaning that developers and users globally can deploy it locally and customise it for their own use cases.
This matters increasingly for the Global South, where firms are being priced out of accessing frontier US models with premium prices. Vince Iswara, CEO of DANA, one of Indonesia’s biggest digital wallet providers, said China’s model of innovation was more attractive for emerging markets because it is carried out under serious cost and resource constraints.
Chinese AI model developers will no doubt agree, with estimates that the US has five times more AI supercomputers and three times better cutting-edge semiconductor chips than China. Still, China continues to be competitive in frontier AI, and there are signs that its innovative capacity is even bearing fruit in breaking free of US chokeholds on advanced chips.
On Tuesday, Chinese food delivery giant Meituan announced that it had developed the country’s largest AI model trained entirely on domestic chips. The claim, if true, raises serious questions about the merits of the ongoing US strategy of constraining China’s AI industry by starving it of advanced US chips, as China seems to be innovating its way out of such restrictions to remain at the forefront of global AI development while also furthering its own self-reliance strategy.
For countries with even fewer resources operating under similar constraints, that in itself is already a resounding model of success. #🇨🇳 ChinA.I. 🤖🧠🦾🤖
https://www.scmp.com/tech..
