The State of AI represents a partnership between the Financial Times & MIT Technology Review exploring how AI is transforming global power dynamics. Each Monday over the next six weeks, contributors from both outlets will discuss a facet of the generative AI movement that is altering international influence.
In this exchange, the FT’s technology columnist and Innovation Editor John Thornhill along with MIT Technology Review’s Caiwei Chen explore the contest for technological dominance between Silicon Valley and Beijing.
John Thornhill writes:
From an outside perspective, it appears inevitable that China will rise as the AI superpower of the 21st century.
In the West, our first reaction typically emphasizes America’s considerable advantage in semiconductor skills, cutting-edge AI research, and extensive investments in data centers. Warren Buffett famously cautioned: “Never bet against America.” He is indeed correct that for over two hundred years, no other “incubator for unleashing human potential” has equaled the US.
However, today, China possesses the capability, intent, and chance to engage in what could be termed technological takeover. When it comes to mobilizing the collective resources necessary to maximize AI development and deployment, it might be just as reckless to bet against them.
The statistics illustrate the trends. In terms of AI publications and patents, China is at the forefront. By 2023, China made up 22.6% of all citations, surpassing Europe at 20.9% and the US at 13%, as indicated by Stanford University’s Artificial Intelligence Index Report 2025. Furthermore, in 2023, China held 69.7% of all AI patents. Admittedly, the US continues to dominate the top 100 most cited publications (50 against 34 in 2023), but its dominance is gradually waning.
Likewise, while the US outperforms China in top-tier AI research talent, the gap is closing. A report from the US Council of Economic Advisers indicated that 59% of the world’s leading AI researchers were located in the US in 2019, compared to 11% in China. Yet by 2022, those numbers shifted to 42% and 28%, respectively.
The tightening of restrictions for foreign H-1B visa holders during the Trump administration might lead an increasing number of Chinese AI researchers in the US to return to their homeland. The talent distribution could further tilt in favor of China.
Concerning the technology itself, institutions in the US developed 40 of the globe’s most prominent AI models in 2024, with only 15 from China. Nevertheless, Chinese researchers have effectively learned to achieve more with fewer resources, and their leading large language models—including the open-source DeepSeek-V3 and Alibaba’s Qwen 2.5-Max—outperform the best US alternatives regarding algorithmic efficiency.
China is poised to excel in the future by leveraging these open-source models. The latest study from Air Street Capital indicates that China has now surpassed the US in monthly AI model downloads. In the domains of AI-enabled fintech, e-commerce, and logistics, China already outpaces the US.
Perhaps the most fascinating—and potentially the most fruitful—applications of AI are yet to emerge in hardware, especially in drones and industrial robotics. As research progresses towards embodied AI, China’s edge in advanced manufacturing will become apparent.
Dan Wang, the technology analyst and author of Breakneck, has accurately underlined the capabilities of China’s engineering state in cultivating manufacturing process expertise—even while he has illustrated the detrimental impacts of applying that engineering mindset in social contexts. “Technologically, China has been becoming stronger and economically more vibrant in a multitude of ways,” he shared with me. “However, repression is very real. And it is intensifying in numerous ways.”
I’d love to hear your perspective, Caiwei, regarding the advantages and disadvantages of China’s AI vision. To what degree will China’s orchestrated social control impede its technological aspirations?
Caiwei Chen responds:
Hello, John!
You’re correct that the US maintains a significant edge in cutting-edge research and infrastructure. However, “winning” in AI can encompass various interpretations. Jeffrey Ding, in his book Technology and the Rise of Great Powers, presents a surprising argument: For a general-purpose technology like AI, lasting advantages often hinge on how broadly and deeply technologies integrate into society. China is well positioned to excel in that competition (although “murder” might be an exaggerated term!).
Chips remain China’s most substantial obstacle. Export limitations have constrained access to premier GPUs, compelling buyers into gray markets and causing labs to recycle or repair banned Nvidia units. Even as national chip initiatives expand, the performance disparity at the very top is still evident.
Nevertheless, these same limitations have pushed Chinese firms toward an alternative strategy: pooling computing resources, enhancing efficiency, and releasing open-weight models. For instance, DeepSeek-V3’s training process utilized only 2.6 million GPU-hours—significantly less than that of its US counterparts. Yet Alibaba’s Qwen models have emerged as some of the most downloaded open-weights globally, and companies like Zhipu and MiniMax are developing competitive multimodal and video models.
China’s industrial strategy allows new models to transition quickly from lab to practical application. Local governments and major corporations are already deploying reasoning models in sectors like administration, logistics, and finance.
Education also presents a competitive edge. Leading Chinese universities are integrating AI literacy programs into their curricula, instilling skills before job market demands arise. The Ministry of Education has also unveiled plans to incorporate AI training for children across all school levels. While I’m not certain if the term “engineering state” accurately reflects China’s engagement with new technologies, decades of infrastructure development and centralized coordination have created a remarkably effective system for driving large-scale adoption, frequently encountering significantly less social resistance than seen elsewhere. The extensive use, of course, facilitates faster iterative advancements.
Meanwhile, Stanford HAI’s 2025 AI Index discovered that respondents in China exhibited the highest optimism globally regarding AI’s future—far more so than populations in the US or the UK. This is notable, especially considering that China’s economy has decelerated since the pandemic for the first time in over 20 years. Many within government and industry are now viewing AI as a crucial catalyst. While optimism can serve as a powerful motivator, whether it can withstand slower growth remains an open question.
While social control is indeed part of the equation, a different ambition is forming. The new generation of Chinese AI founders exhibits an unprecedented global outlook, effortlessly navigating between Silicon Valley events and pitch meetings in Dubai. Many possess fluency in English and are attuned to the dynamics of international venture capital. Having observed the previous generation grapple with the weight of a Chinese identity, they are now founding companies that are purposefully transnational from the outset.
While the US may still lead in rapidity and innovation, China has the potential to define how AI integrates into everyday life, both domestically and internationally. Speed is essential, but it is not synonymous with supremacy.
John Thornhill replies:
You’re absolutely correct, Caiwei, that speed doesn’t equate to supremacy (and “murder” might indeed be an exaggerated term). Additionally, you accurately emphasize China’s advantages in open-weight models versus the US’s inclination toward proprietary models. This struggle is not merely a clash of economic approaches between two nations but also a contest of contrasting methodologies for the application of technology.
Even OpenAI’s CEO, Sam Altman, acknowledged earlier this year: “We have found ourselves on the wrong side of history and must devise an alternate open-source strategy.” This evolving subplot will be fascinating to observe. Who will emerge victorious in this regard?
Further reading on the US-China rivalry
Extensive discussions have arisen regarding how individuals may incorporate generative AI into their everyday lives. This report from the FT’s visual storytelling team delves into the reality
From the Chinese perspective, FT journalists query how long Nvidia can sustain its lead over its Chinese competitors
In real-world applications, toys and companion devices represent a novel yet emerging use of AI that is gaining momentum in China—and is likewise making its way to the US. This MIT Technology Review article explored this topic.
The previously rapid data center expansion in China has encountered obstacles, and as restrictions and AI demands evolve, this MIT Technology Review piece provided a ground-level examination of how various stakeholders are adapting.
