China’s AI Path and the Needham Question: From 1 to 10, Not 0 to 1

"In the context of US-China rivalry, innovative resilience may matter more than the capacity for breakthrough innovations". — Huang Ping (黄平)

---

As both nations invest ever more significant resources into artificial intelligence, the divergent trajectories between China and the US have become a key issue—with many pointing to a philosophical—even quasi-religious—divide between the rapid adoption of practical AI applications in the former and the more abstract pursuit of AGI in the latter. Huang Ping, an up-and-coming associate professor at the Chinese University of Hong Kong, agrees with these analyses. He notes that despite state-funded investment drives in basic research, China’s innovation system remains relatively weak at original breakthroughs (moving from “0 to 1”) but excels at scaling and commercialising technologies (moving from “1 to 10”).

In developing this argument, Huang evokes an indigenous cultural tendency towards “practical application” over “knowledge for its own sake”, drawing on the well-known response of Wu Guosheng, a historian of science, to the “Needham Question”. This was the question posed by the polymath and sinologist Joseph Needham, of why modern science developed in the West and not China, when Chinese civilisation had previously been “more efficient than occidental [civilisation] in applying human natural knowledge to practical human needs”. Wu’s answer is that the concept of “science” and its connotations of the pure pursuit of truth are unique to the Western Christian cultural sphere, rather than universal values. From this perspective, the fact that China lacked “science” (科) but had “technique” (技)—the practical side of technology (科技)—is not at all surprising.

Huang argues that it is wrong-headed to see the relative absence of science in the Chinese tradition as a deficiency to be filled; this emphasis on “application” over “breakthroughs” carries its own strengths as well as weaknesses. Such a counter-intuitive response to Needham’s musings departs widely from that of other scholars, including Justin Yifu Lin, who blamed an overly formalised and hierarchical civil service examination system for holding China’s potential innovators back (a view developed in recent years by MIT professor Huang Yasheng).

In another respect, Huang Ping is clearly influenced by Lin’s economic theories—which we covered in a previous piece. Given China’s comparative disadvantage in “0 to 1” innovation, Huang believes that huge state investment in basic research is uneconomic. Instead, the government should harness the state’s demand capacity to shape the market conditions for unleashing China’s comparative advantage in scaling and commercialising AI applications.

— James Farquharson

Key Points

1. The stakes in the AI contest could not be higher: they concern the “underlying operational logic” of future modes of production, power relations and models of cognition.

2. AI competition is shaped by the differing political, developmental and value systems of China and the US—its outcome will determine whether the world is one of “universal values” or “plural civilisations”.

3. Yet whereas the US tends to frame the contest as requiring absolute victory, China’s aim should be to maintain approximate parity by keeping pace in the most urgent areas—rather than seeking to “crush one’s opponent”.

4. Because China lacks a tradition of “scholarship for its own sake and knowledge for its own sake”, its capabilities in “original innovation” inevitably lag behind those of the US.

5. However, China excels in commercial applications of AI and in scaling tools, partly as a result of its long-standing emphasis on “learning for practical application”.

SINIFICATION IS HIRING
Full- or part-time analyst, £35k–£55k FTE, Remote
Click here for more details
Please send your resume and cover letter by 5 October to:

6. This advantage in commercial applications, contrasted with US leadership in general-purpose models and chip architecture, is producing “performance and ecosystem differences within the same generation” of AI.

7. China does not need to prioritise achieving “0 to 1” breakthroughs; instead, it can build on its comparative advantage in scaling applications, drawing on its “national system of innovation”, sheer market size and complete industrial chains.

8. Rather than investing disproportionately in basic research, the government can leverage its affiliated enterprises and departments to generate demand, thereby incentivising businesses to pursue commercial improvements in AI applications.

9. The “entrepreneurial spirit” of local governments is key, and they should be empowered to support locally driven, commercialised AI development.

10. China should resist the urge towards “technological nationalism” and the pursuit of full autonomy in AI; since its strengths lie in commercialisation, open markets will be essential.

The Scholar

Name: Huang Ping (黄平)
Year of Birth: Not publicly disclosed (age: late 30s)
Position: Associate Professor, School of Public Policy, Chinese University of Hong Kong, Shenzhen; Deputy Director (Development), The Institute for International Affairs, Qianhai
Research Focus: Innovation Studies, Economic Geography, Urban Governance, Energy Transition
Education: BS Harbin Engineering University, China (2010); PhD Harbin Engineering University, China (2015)
Experience Abroad: Postdoc, The Bartlett School of Architecture, University College London, UK; Postdoc, The Fletcher School of Law and Diplomacy, Tufts University, US; Research Associate, The Urban Institute, University of Sheffield, UK

TECH OBSERVATION | HUANG PING: CHINA’S CHOICE OF TECHNOLOGICAL PATHWAYS UNDER THE SINO-US AI COMPETITION
Huang Ping (黄平）
Published by Greater Bay Area Review on 16 September 2025
Lightly edited machine translation
(Illustration by OpenAI’s DALL·E 3)

1. Beyond Technology, the AI Contest Is a Trial of National Strength

Since the release of ChatGPT in 2022, global artificial intelligence has entered an explosive phase. As the leading forces of technological innovation, China and the US have successively launched multiple large language models, such as Google’s Gemini in the US and DeepSeek in China. As iterative improvements in technology accelerate and the future industrial landscape becomes clearer, countries have gradually reached a consensus: AI will be the core of the next industrial revolution. Governments everywhere have introduced AI development strategies to secure advantageous positions in the emerging order. The AI race is no longer merely a competition among tech companies, but a contest between the national powers behind them.

Before DeepSeek’s emergence into the world [问世], mainstream international research and media largely concentrated on comparing the technologies of US tech giants, with few viewing China as capable of competing with the US. DeepSeek’s meteoric ascent [横空出世] has forced American circles to face up to [正视] China’s strength in AI. This has prompted more comprehensive observations and objective comparisons [between the two powers] spanning compute power, algorithms, data resources and practical applications. The focus has shifted from the technologies themselves to the industrial system and the broader national strength that underpins them.

AI is by no means an ordinary technology [绝非普通技术]. Unlike high-speed rail, 5G or nuclear power—technical fields that primarily reflect industrial capability—AI is regarded as the “operating system” of future human society [未来人类社会的“操作系统”]. It may reorganise modes of production, restructure power relations and even reshape human cognition. Its development spans frontier basic research, capital- and talent-intensive technological transformations, high-end manufacturing, scaled and high-quality application scenarios and market demand—as well as systematic [approaches] to state governance. Hence, this contest has already passed well beyond [the realm of] technology and industry; it is essentially a contest of national power.

The American elite have formed a consensus: winning the AI contest means seizing global leadership. The newly released America’s AI Action Plan [Note: A White House policy document released in July 2025] makes it clear that the US is not seeking leadership in the ordinary sense but rather indisputable global dominance [无可争辩的全球主导地位]. Its aim is to transform AI into a new-model “colonial” tool [新型“殖民”工具]. By controlling chips, models and application programming interfaces (APIs), it may compel other nations to act according to its rules [按其规则行事].

What the US truly fears is not the release of multiple large language models by China, but rather China leveraging AI to reconstruct the global economic order and [balance of] international discourse power [国际话语权]. Although China harbours no such intent, we must soberly recognise that the Sino–US AI competition is ultimately a contest of comprehensive national strength, and its outcome will profoundly shape the future global balance of power and trajectory of human civilisation.

2. Economics, Politics and Even Value Systems Are at Stake

In the global AI race thus far, the only two nations capable of fully competing in technological capacity and development potential are China and the US. These two nations embody starkly different development models, state systems and even value frameworks. Consequently, the outcome of this contest will shape the world’s future trajectory profoundly, with the winner potentially determining its “underlying operational logic” [“底层运行逻辑”].

First, the outcome of the AI contest concerns how the remodelling of the global economic domain will proceed. Mastery of core AI capabilities will grant the opportunity to be a rule-maker in global industrial chains. Historically, Britain during the Industrial Revolution relied on the steam engine to achieve global hegemony, while the United States leveraged the internet to seize the initiative in the information era. In the AI era, compute power, algorithms and data will form the new “troika of growth drivers” [Note: “三驾马车”, referring to the three components of GDP: consumption, investment and exports]. Whoever can seize these commanding heights first [占领这些制高点] may be able to set the “rules of the game” for the future digital world and establish a new economic and financial order.

Second, it will drive the restructuring of the global political order. The redistribution of economic power [through advances in AI] will inevitably reshape geopolitical dynamics. It will not only alter the competitive and cooperative relations between major powers but also affect relations between developed and developing nations. It could even transform the international governance system that, since the Second World War, has revolved around the United Nations, centred on the West and been directed by capital.

Finally, the deeper impact lies in the possibility that the AI struggle will reshape humanity’s dominant ideologies and even its value systems. Throughout history, whether under capitalism or socialism, technological revolutions have profoundly altered ideological and moral frameworks. The outcome of the AI competition may well redefine the core values of human society—and China and the US are likely to offer different answers. Will it be the might-makes-right ideology [强权主义] of the “hegemonic way” [霸道], or the inclusivity and collective rule [包容共治] of the “kingly way” [王道] [Note: a Confucian concept of virtuous rule adopted by the modern Tianxia school of international relations, associated with Zhao Tingyang]? Will it be the imposition of monolithic “universal values” [普世价值], or respect for “plural civilisations” [多元文明]?

Hence, this contest is not merely about which nation has stronger companies or industries, but about who will dominate the future global economy, how international political power will be distributed and what trajectory humanity’s fundamental value systems will ultimately follow.

3. Contrasting Development Models in the Sino–US AI Contest

The competition between China and the US in AI has far-reaching consequences and is essentially a struggle of national power—the significance of which requires no elaboration. The more pressing question at this stage is: how far has this competition advanced, and how did it reach this point? Accurately gauging this stage is crucial as AI is iterating at extraordinary speed [迭代极快]—not only much faster than traditional industrial technologies, but even outpacing internet technologies. Only by clearly understanding the stage where we currently stand can we identify the core areas for focusing effort and establish a hierarchy of priorities.

Overall, the Sino–US AI competition can be divided into three stages, each marked by a distinct technological trajectory and industrial focus. Together, they reflect the two countries’ contrasting approaches to development.

The first stage spanned roughly from 2012 to 2019. The year 2012 is widely regarded as the “inflection point of deep learning” [“深度学习拐点”], when the feasibility of large-scale deep neural networks and GPU-based training was established. On this foundation, DeepMind achieved breakthroughs in deep reinforcement learning, culminating in the release of AlphaGo in 2016. Thus, the US essentially laid the technological foundations of AI, seizing the critical “chokepoints” at the origin [握住了“卡脖子”的源头环节]. Meanwhile, NVIDIA released GPU products such as the P100, gradually assembling the early-stage ecosystem of “algorithms–chips–data” and establishing dual leadership in both research paradigms and foundational hardware.

China’s strategy during this period was to follow closely and accelerate implementation. By relying on NVIDIA chips and US open-source frameworks like TensorFlow [Note: an open-source machine-learning framework developed by Google], Baidu established its Deep Learning Research Institute, while Alibaba and Tencent set up AI labs. These quickly embedded technologies into vertical [application] scenarios [垂直场景] such as financial risk control and e-commerce recommendations, using these to drive technological development. One could say that at this stage, the US dominated basic research and the rule-setting of the [technological] ecosystem, while China primarily played catch-up.

The second stage ran from 2020 to 2023 and was marked by [the release of] OpenAI’s GPT-3 and the subsequent explosion of ChatGPT, the first large-scale application of large language models. Google launched PaLM and Microsoft integrated GPTs into Office and Azure, forming a closed loop of “large models–compute power–cloud services”, with business models and distribution channels maturing in tandem. On the hardware front, NVIDIA’s A100/H100 chips came to almost monopolise the high-end compute market with a share exceeding 80%, combining both technical and supply-side advantages.

During the same period, China developed general-purpose models such as Baidu’s ERNIE Bot and Alibaba’s Tongyi Qianwen (Qwen); at the same time, Huawei’s Pangu series targeted specialised fields such as industry and meteorology and its Ascend 910 chip entered commercial use. In many vertical [application] segments [Note: Specialised applications for specific industries and use-cases, as opposed to general-purpose models], particularly AI penetration in industrial [production], China demonstrated faster advancement.

Overall, the US maintained absolute leadership in both software and hardware, but the Sino–US technological pathways began to diverge [分化]. While the US emphasised stacking compute power [堆叠算力] and enhancing general-purpose capabilities [通用能力], China focused more on deep integration into industries, using the business context to drive model optimisation [用业务场景倒逼模型优化].

The third stage runs from 2024 to the present, and is marked by the release of DeepSeek-V2. DeepSeek-V2 matched—and in some areas even surpassed—the capabilities of top closed-source models, while significantly reducing inference costs through architectural innovation. This attracted global attention, especially from US tech circles. On the hardware front, Huawei’s Ascend 910B achieved about 90% of the performance of NVIDIA’s A100 in inference scenarios—at a lower cost. Humanoid robots such as UBTECH’s Walker X sprang forth in large numbers, entering commercial applications across various contexts. All the while, US general-purpose large models have maintained a clear overall lead through continued iterative improvements in the multimodal space, advancing towards real-time interaction across text, image and audio.

This is creating a landscape shift from [the US] “being a generation ahead” [“代际领先”] to [the US and China having] “performance and ecosystem differences within the same generation” [“同代际内的性能与生态差异”]. The US cannot fully suppress all of China’s breakthroughs in vertical applications [垂直应用], but nor can China easily surpass US advantages in general-purpose models and chip architecture in the short term. Thus, at the technical level, the situation is one of “stalemate” [“相持”]. In the near future, it is unlikely that any one side will achieve “comprehensive dominance” [全面压制] over the other; rather, a long tug-of-war [长期拉锯] will unfold within the same technological generation over performance, cost, ecosystems and application capacities.

4. The “Needham Question”: China’s Strengths and Weaknesses in Innovation

The Sino–US AI struggle has entered a stage of “stalemate” where the decisive factors are resilience and systemic capacity. AI competition has never been just a contest over a single technology, but rather a decisive battle of the overall national power underpinning the technology. Therefore, any assessment of China’s strengths and weaknesses [优势与短板] must be situated within this current stage and context. Each stage brings its own advantages and disadvantages, and observations at different levels [不同层面] yield distinct conclusions. At this juncture, what we need most is systematic judgement and careful weighing of trade-offs [across alternative pathways]. In the face of this great transformation, direction matters more than speed [方向比速度更重要].

Meanwhile, the fully-fledged contest between China and the US in trade, technology and politics has fuelled nationalist sentiment in both countries. Discussions around AI are tending increasingly towards the irrational, while technological nationalism [技术民族主义] is gaining ground. In this context, maintaining rationality is crucial. [One must] first acknowledge our shortcomings objectively, and then systematically identify our strengths on that basis. Overall, China’s [main] shortcoming lies in [achieving] original breakthroughs [原创突破] in core technology—[moving] “from 0 to 1”; our relative strength lies in scaling and commercialising technologies—[moving] “from 1 to 10”.

China’s shortcomings in basic research and original innovation relative to the US are largely undisputed [没有太大争议]. But as to why it lags, or whether these gaps can be closed, scholars hold differing views. Much like the diverse answers to the famous “Needham Question” [李约瑟之问], interpretations vary widely [Note: the influential question posed by biochemist, sinologist and historian of science Joseph Needham—namely, why did the modern science revolution take place in the West and not China, whose civilisation in previous centuries had been “much more efficient than occidental [civilisation] in applying human natural knowledge to practical human needs”?].

I tend to agree with the perspective of Professor Wu Guosheng, an expert in the philosophy of science and technology at Tsinghua University. In What is Science (《什么是科学?》), Wu argues: “Ancient China did not possess science in either the modern sense of mathematical-experimental science [数理试验科学], or in the Western sense of rational science [理性科学].” He also notes: “Chinese culture lacks the spirit of ‘scholarship for its own sake and knowledge for its own sake’ [中国文化中缺乏‘为学术而学术、为知识而知识’的精神]; instead, the tradition of learning for practical application [学以致用] is overwhelmingly strong [太过强大].”

Understandably, many are reluctant to accept or confront such an interpretation. But it is not a denial of the scientific achievements China has made since entering modernity; rather, it is an attempt to examine our traditions from an objective and rational perspective. As the Sino–US AI competition enters the current “stalemate” stage, such an awareness is especially important in understanding how we might accurately grasp our fundamental shortcomings [根本短板] in the AI contest.

So-called “original innovation” [原始创新] refers to “0-to-1” breakthroughs [突破]—creating something from nothing [从无到有]. Historically, such breakthroughs have mostly arisen from the West’s “scientific spirit” [“科学精神”], centred on rationality. Rationality here is first and foremost a value system, rather than [merely] a tool.

Since the Self-Strengthening Movement [洋务运动] [Note: a late nineteenth-century Qing dynasty initiative to adopt Western technology and, to a certain extent, institutions and thinking] and the establishment of the modern system of scientific disciplines, China has gradually formed an academic system similar to that of the West. The quantity and quality of Chinese scholars’ papers in international journals have also entered the global forefront. Yet empirically speaking, the motivation of many researchers to carry out research is not the pure pursuit of truth [对真理的纯粹追求], but rather the pursuit of “useful knowledge” [“有用之学”]. This is not to make a value judgement: seeking truth is not necessarily superior to seeking utility [求真未必天然高于求用]. Nonetheless, one must acknowledge that original knowledge generally arises from a sustained and disinterested pursuit of truth.

5. China’s Core Strengths are “Systemic Capacity” and Making Iterative Improvements

China’s longstanding advantage in technological development has always lain more in its ability to carry out research, development and iterative improvements within existing technological frameworks. On this point, there is broad consensus in the tech and academic worlds. Whether in high-speed rail, 5G communications, mobile payments or—in recent years—rapid advances in new energy and electric vehicles, China has demonstrated formidable capacity for “application innovation” [“应用创新”] and “engineering implementation” [“工程化落地”].

Yet in recent years, as China has notched up a series of breakthroughs in frontier areas such as chip design, operating systems, large language models and bioinformatics, some experts and scholars have begun developing a “blind” confidence [“盲目”自信] in China’s innovative ability. They argue that China has already levelled up technologically, striding into the “8-to-10” cutting-edge band—sufficient to contend with the US and even to create two mutually independent global tech systems. In their view, China’s technological specialism has leapt from “mid- to low-end iterative improvements” to “mid- to high-end leadership”.

But the reality may be more complex. Based on our team’s field research over the past two years on research institutions, leading enterprises in strategic emerging industries and innovative start-ups in the Greater Bay Area [Note: comprising the Guangdong Pearl River Delta, Hong Kong and Macau] and the Jiangsu–Zhejiang region, the scientific and industrial communities’ evaluation of China’s current technological level is far less rosy than public opinion would suggest.

The prevailing view is that China still sits at a “medium technology” level in most industrial sectors. Professor Zheng Yongnian also pointed this out explicitly in his book The Middle-Technology Trap, published last year: “In terms of China’s technological level, whether from the perspectives of supply chains, industrial chains or value chains, it sits broadly at the medium level” [Note: Zheng’s book is based on a 2023 article in which he argued that China’s technological advancement lay at around the “4 to 7” level, sparking much controversy and a lively debate on China’s approach to innovation]. This means that China still faces “chokepoints” [“卡脖子”问题] in many critical core technologies—especially in high-end chip manufacturing, industrial software and precision instruments—where a clear gap with the world-leading level remains.

Meanwhile, another extreme view is also spreading: some believe that China’s manufacturing advantage is being gradually supplanted by Southeast and South Asian countries, and that [consequently] China’s status as the “world’s factory” will be hard to sustain [难以为继]. This view rests on the observation that in recent years, Chinese companies have accelerated their expansion overseas [加速出海], with many labour-intensive industries relocating to Vietnam, India and elsewhere. Some mid- and low-end “Made in China” products have already been replaced by local production capacity. This appears to be replaying the historical trajectory of manufacturing shifting from Europe to the US, and then to East Asia. But this time, the centre of global manufacturing may not “relocate” so easily for three reasons:

First, no other country can match the completeness of China’s industrial chain. Today’s global consumption trends have shifted from standardised mass production to personalised and customised demand. Even traditional “mid- to low-end” products like clothing and footwear now require rapid response [to demand], small batches and multiple styles. This capability depends not only on cheap labour but also on efficient, flexible and comprehensive industrial-chain coordination, and China is currently the only country in the world possessing all the industrial categories classified by the United Nations. In the Greater Bay Area in particular, every segment—from raw materials and components to design, logistics and sales—can be completed within a radius of just 100 kilometres. This “complete industrial chain + rapid response” capacity is something Southeast Asian countries cannot easily replicate in the short term.

Second, structural advantages arise from ultra-large scale [超大规模]. China possesses the world’s most powerful manufacturing capacity, whose economies of scale give China an almost irreplaceable manufacturing edge in cost control and product value-for-money. Even though labour costs have risen in recent years, China’s manufacturing industry can still rely on its complete and efficient supply chain system and operational efficiency to maintain competitiveness. Such resilience—supported by “sheer scale” [“体量”]—is something smaller economies cannot match.

Third, the “systemic capacity” [体系能力] of the national system of innovation has largely taken shape. As the concept’s originator, British economist Chris Freeman, noted, innovation as a systemic process depends on the collaborative interaction of multiple actors—enterprises, universities, research institutes, government bodies and intermediary institutions [Note: Freeman is known for his work on “national systems of innovation”]. China’s “systemic capacity” for innovation is manifested not merely in effective industrial policy or in isolated technological breakthroughs, but in an integrated capability of virtuous interaction—in other words, an innovation ecosystem [创新生态]. Over recent decades, China has progressed from simple processing to imitative innovation, integrated innovation and indigenous innovation. Behind this lies the formation of the national system of innovation’s “systemic capacity”—an “institutional asset” [“制度性资产”] that is difficult to replicate.

In AI, these three advantages will be fully brought to bear. On the one hand—and most importantly—the essence of Sino–US AI competition is no longer simply a contest over technological breakthroughs, but one of comprehensive national strength. Particularly in the context of Sino–US rivalry, innovative resilience [创新的韧性] may matter more than the capacity for breakthrough innovations, and China’s national system of innovation provides precisely this innovative resilience. On the other hand, as AI moves from technological exploration to large-scale application, the key to victory is no longer merely “whose technology is more advanced”, but “whose technology is most usable, cheap and reliable”. China’s complete industrial chain and the ultra-large scale of its manufacturing is enabling rapid product roll-out and cost-effective commercialised scaling. Therefore, within the AI competition, China should stay the course and fully leverage its strengths in technological iterative improvement, industrial-chain coordination and market application.

6. What Should China Avoid Doing in the AI Race?

In the Sino–US rivalry, America’s goal is clearly to win. That, however, should not be China’s goal; instead, “not losing” [“不输”] would be substantively a win for China. With regard to weaknesses, we must acknowledge and understand these gaps rationally while keeping pace with the US’s technological rhythm and generational progression. [The aim should be to] avoid “being left behind” [被甩开], rather than indiscriminately striving for parity in every sector or blindly pursuing total catch-up [追赶] or overtaking [反超]. In terms of our strengths, resources should be concentrated to reinforce and amplify them, converting comparative advantages into stable core competitiveness and institutional influence. We can thus leverage our advantages to gain greater room for development. Integration and cooperation—rather than seeking to crush one’s opponent—should be China’s guiding principle in the Sino–US AI competition. Under this principle, China should be clear about what to avoid doing rashly and where to focus its efforts.

First, [we should] not overinvest in basic research. In recent years, anxiety over “chokepoints” has prompted policymakers to place a high priority on basic research. As a corollary, this has led to many applied projects siphoning off large sums of funding under the banner of basic research. The outputs [of these projects] often remain at the research paper level [留在论文层面] and far from industrial application, resulting in wasted resources. We must not allow ourselves to be taken hostage by “chokepoint” emotions [不应被“卡脖子”情绪绑架] because, in practice, it is the technologies—not basic science itself—that are being blocked.

Basic research and the knowledge behind it cannot be sealed off. Since our goal is “not losing”, there is no need to stake our bets on foundational breakthroughs in the short term. Instead, we should concentrate resources on catching up and iterative improvements in the most urgent technologies to ensure that we do not fall behind [不掉队]—and that the outcomes can be implemented and scaled.

Second, [we must not] allow AI to be hijacked by “technological nationalism”. Against the backdrop of fully fledged Sino–US competition, nationalist sentiment is rising in both countries, and AI now sits at the centre of this technological contest. From large models to chips and robots, everything is labelled by country of origin. The politicisation of technology harms the innovation ecosystem on either side and is especially detrimental to China. “American AI” is generally not systematically excluded by mainstream markets, whereas the “Chinese AI” label is often rejected by the West—and even by some developing countries—on the grounds of national security or ideology.

Therefore, Chinese AI must not be captured by “technological nationalism” in pursuit of absolute autonomy in technology and supply chains. On the contrary, China should adopt a unilaterally open [单边开放] posture in order to advance localised development [推进在地化发展], doing so through the means of joint R&D, open ecosystems and cross-border cooperation across technology, products, corporate governance, ownership structures, branding and operations [Note:The “unilaterally open” proposition comes from Huang’s colleague at CUHK, Zheng Yongnian].

7. Where Should China Focus Its Efforts?

First, the state mobilisation system [举国体制] should be used to offer AI application scenarios and markets. AI is a highly market-oriented racecourse, where R&D and commercialisation require sharp commercial judgement and rapid iterative improvements. On the supply side, the public sector should yield fully to market actors. Government should not overstep when directing industrial development [政府不越位主导产业发展]; nor should central and local state-owned enterprises (SOEs) compete with the people for profits [与民争利]. The market should be the selector of technological routes and business models. Only in this way can China’s AI development keep pace with the innovation speed of US tech giants.

Conversely, on the demand side, the public sector should fully leverage the institutional advantages of the state mobilisation system and its ultra-large market to create application scenarios and markets for mature and immature AI technologies and products. This includes not only the many public domains directly administered by government departments—such as healthcare, education, transport and government services—but also key industries dominated by central and local SOEs, including energy, communications, aviation and the military-industrial complex. Creating such a large volume of real scenarios, demand and feedback will provide the most direct and effective policy support for China’s AI development, helping firms accelerate iterative improvements in technology from prototypes to mass production and from pilots to scalability. This will thereby win China time and space [赢得时间和空间] in the Sino–US AI competition.

Second, [we must] fully unleash the “entrepreneurial spirit” [“企业家精神”] of local governments. Over more than forty years of reform and opening-up, a key source of China’s economic vitality beyond the central government’s top-level design and institutional reform has been local governments’ “entrepreneurial spirit”—that is, their proactive role in industrial and economic development, their willingness to experiment and their partnerships with local firms to open up markets. As BYD’s founder Wang Chuanfu put it, “Without Shenzhen, there would be no BYD.” Cities such as Hefei, Suzhou and Wuhan have likewise nurtured leading tech champions [科技龙头] and the vital role of local governments has been repeatedly proven in practice. Of course, unrestrained local governments can give rise to collusion and rent-seeking [权钱交易], but excessively shackled local governments lead to more serious problems such as governance inertia [懒政惰政]; the former is a problem within development, but the latter is the deadlier problem of non-development.

In the present and foreseeable future, when the Sino–US AI race is counted in days, development is the foremost priority. Therefore, we should fully unleash the “entrepreneurial spirit” of local governments, especially in regions where innovation factors [创新要素] are highly concentrated, such as the Greater Bay Area, the Yangtze River Delta, and the Beijing–Tianjin–Hebei region. [These areas] should be granted greater space for institutional innovation and trial-and-error. Only in this way can more breakthrough players [突破者] like DeepSeek emerge on the AI racecourse and foster “the next BYD” across various sub-racecourses.

READ MORE