China’s AI Strategy: Encircling the Cities From the Countryside — by Di Dongsheng

"In the early stages of this competition, we should employ all available tools—state subsidies, financial support, diplomatic pressure—to encourage as many countries as possible to adopt Chinese AI."

Today’s edition opens with an introduction by my former colleague Rebecca Arcesati, Lead Analyst in the Science, Technology and Innovation programme at the Mercator Institute for China Studies (MERICS). In particular, her research examines China’s AI and data strategies, EU–China innovation relations and US–China tech competition. Much of her work can be found here. I’m delighted to have her insights included in this issue. — Thomas

There is much to unpack in Di Dongsheng’s essay. The author alternates cautious optimism about the development trajectory of China’s AI industry, which he sees as more resilient compared to that of US AI companies, with some gloomier views of the fate of socioeconomic systems and humanity at large in a world disrupted by AI technologies.

In the United States, a messianic pursuit of superintelligence has come to be the dominant paradigm – a costly and risky bet with unclear returns. The Chinese leadership so far seems to favour a more pragmatic approach that emphasises AI adoption in specific use cases. Di believes that America’s “AI bubble” will burst. He seems less preoccupied with Xi Jinping’s no less messianic belief that AI can fix all the country’s problems: China too may have its own AI bubble – though its economy is designed to tolerate inefficiency and declining profits.

Certainly, China will not escape AI’s disruptive effect on social structures. Informed by Marxist and neo-Marxist thoughts, Di is especially concerned about unemployment risks, foreseeing deepening inequalities and social conflicts. To manage this transition, he believes countries will need agile, resilient and adaptive institutions that strike a balance between bottom-up and top-down regulation. This type of thinking in part reflects the evolving debate around AI risks among China’s most influential technical experts and policy advisers.

Globally, third countries may become the decisive battlegrounds in a cut-throat competition between American and Chinese AI. Di urges China to partake in this land and data grab to gain commercial and geopolitical influence, leveraging open-source technology and low prices. This is already starting to happen, and one wonders whether the US government’s recent strategic shift from tightening compute export controls to favouring the global diffusion of the American AI stack might, inadvertently, help the very Chinese competition Washington seeks to contain.

While the United States is still ahead in computational power – a key ingredient in AI competition – China has an energy advantage bolstered by its grip on critical raw materials, something the author notes. AI runs on chips, and minerals are essential for making chips and to power energy-hungry data centres. With its recent restrictions on mineral exports, Beijing may have inched closer to the author’s envisaged outcome: A world dependent on cheap Chinese AI exports.

— Rebecca Arcesati

Key Points

1. The US pursues AI like a quest for divine enlightenment—seeking “the singularity”, the moment when machines transcend human capabilities—while fearing China might achieve this breakthrough first.

2. China, by contrast, follows a pragmatic “AI plus” path, prioritising specialised, application-oriented systems that deliver tangible returns and real-world efficiency, rather than speculative breakthroughs towards AGI.

3. Unlike America’s closed-source approach, China’s open-source model draws international developers and the Global South into its orbit—a tactic that echoes Mao’s “encircling the cities from the countryside”, building dominance through inclusive, peripheral consolidation.

4. Today’s AI boom resembles the dot-com bubble of the late 1990s, fuelled by speculation rather than sustainable profitability. Beyond Nvidia’s shovel-seller economics, most firms show weak commercial fundamentals and dangerously inflated valuations.

5. Artificial intelligence is propelling humanity towards Marx’s “communist stage”, where abundance replaces scarcity—yet, it also risks deepening inequality by dividing society into AI controllers and the controlled.

---

---

6. Employment could become a privilege rather than a right, while emotional intelligence may emerge as humanity’s most valuable skill. Social order could hinge on rewarding those who challenge power, intensifying social conflict.

7. Institutional flexibility and resilience will thus define which great powers thrive in the AI age, as nations must adapt rapidly to both technological potential and the shocks it brings.

8. Ultimately, supremacy in artificial intelligence will hinge on two important factors: the ability to capture and process the data of billions, and the capacity to attract and retain talent.

9. Should the US end up pursuing the renewable energy path advocated by the Democrats, its computing power could end up constrained by China’s dominance in the renewable energy supply chain.

10. For China to prevail, it must globalise its AI technologies, offering affordable, open-source solutions to developing nations—replicating its past manufacturing revolution through a new AI “price revolution”.

    → Di Dongsheng: “In the early stages of this competition, we should employ all available tools—state subsidies, financial support, diplomatic pressure—to encourage as many countries as possible to adopt Chinese AI technologies […] This should be approached with a land-grabbing [跑马圈地] mentality: securing market space in the intermediate countries at the lowest possible cost, even at a loss if necessary.“

Name: Di Dongsheng (翟东升)
Year of birth: 1976 (age: 48/49)
Position: Deputy Dean and Professor, School of International Studies, Renmin University of China (RUC) (2017-now); Dean of the Institute of Regional and Country Studies, RUC; Deputy Director and Secretary-General, Centre for Foreign Strategy Research, RUC (2011-now)
Other: Frequent exchanges with officials at China’s National Development and Reform Commission (NDRC), Ministry of Foreign Affairs, Ministry of Industry and Information Technology, International Liaison Department etc.
Research focus: Global political economy of money and finance; US political economy; Chinese foreign policy
Education: BA, MA and PhD Renmin University of China (1994-2004)
Experience abroad (as a visiting scholar or lecturer): Sciences Po Paris, Durham University, Vrije Universiteit Brussel, Georgetown University

SINO–US STRATEGIC COMPETITION IN THE AGE OF ARTIFICIAL INTELLIGENCE
Di Dongsheng (翟东升)
Published in RUC’s Overseas Monthly Report (域外月报), Sept 2025 edition
Translated by Paddy Stephens
(Illustration by OpenAI’s DALL·E 3)

If any event in the economic history of mankind deserves the label of revolution, it is the Industrial Revolution—that, at least, is how some foreign economic historians have evaluated that era. Yet two centuries after that great transformation began, such an assessment is likely to require major revision. The disruption [冲击] artificial intelligence brings to the global economy—and indeed to human civilisation itself [整个人类文明形态]—is at least no less profound than that of the Industrial Revolution.

In late July 2025, Shanghai hosted the World Artificial Intelligence Conference, at which [China’s] Global AI Governance Action Plan was released. In a parallel development [two weeks earlier], the White House issued Winning the Race: America’s AI Action Plan. This 23-page document contains more than 90 policy recommendations, focusing on three strategic priorities: “accelerating AI innovation”, “building American AI infrastructure”, and “leading in international AI diplomacy and security”. In fact, as early as 2015, a consensus had already formed within the US strategic community that the key arenas of future Sino-American competition would be artificial intelligence and 5G. Ten years on, the US has clearly fallen behind in the 5G race. In the field of artificial intelligence, however, the current balance of competition still slightly favours the US, with China close behind. The rest of the world trails far behind them.

How, during the Fifteenth Five-Year Plan [Note: 2026-2030], will the AI era influence global development and reshape great-power competition? This article will explore several perspectives below.

I. Divergent Paths of AI Development in China and the US

There is a striking contrast between the US and China in terms of their development paths and areas of emphasis, reminiscent of the distinction in martial arts training between internal cultivation [内功] and external cultivation [外功].

The US’ approach to developing artificial intelligence is much like the pursuit of immortality and enlightenment in the Daoist tradition: the US believes that general-purpose LLMs will have a sudden moment of enlightenment [顿悟], thus one day becoming god-like [修炼成仙]. Many in the US think that AI development should involve reaching the “singularity” [“奇点”], a point of breakthrough. Once they have achieved this “enlightenment”, large AI models could “transcend the moral realm” [超凡入圣]—attaining a level of mastery in every specialised field that matches or even surpasses the world’s top experts. Moreover, by integrating knowledge across different disciplines, such a superintelligent entity could follow its own curve of accelerated evolution, ultimately leading to a “dimension reduction strike” [Note: 降维打击, a more advanced civilisation crushing a less advanced one through technological superiority, a popular concept from Liu Cixin’s “Three Body Problem” trilogy]. As a result, Americans both fear that China might be the first to achieve AGI and pin their hopes of winning the strategic competition with China on breaking through to the “singularity” before their rival.

By contrast, the Chinese—whose non-belief in the divine lies deep in their bones [从骨子里就不信“神”的]—have always possessed a pragmatic spirit [秉持务实精神] and are [now] pursuing the development pathway of “AI plus”. This is a bottom-up approach, consisting of incremental gains that add up [积少成多]. Like mastering the eighteen traditional weapons [十八般兵器] [of ancient China], it is a path of hard, disciplined effort [硬功苦功]. China has a vast number of AI-related enterprises, each developing artificial intelligence for specific vertical application scenarios. These companies’ cash flows do not primarily depend on [convincing] the capital markets through storytelling or successive rounds of financing; from the outset, they focus on achieving profitability, generating revenue directly from paying customers. For instance, AI is applied to improve existing processes and tools in roads, mines, construction sites, ports, workshops, hospitals, hotels and even battlefields, thereby enhancing efficiency or controlling costs. Chinese AI is not an all-knowing, all-powerful [Daoist] immortal [神仙], but rather an experienced and skilled craftsman [老师傅], continually refining his craft within different specialised domains.

Not long ago, I was invited to visit a cutting-edge AI and robotics firm. This unlisted company, named MegaRobo Technologies, had already achieved a remarkably high volume of sales and impressive quarterly profits. Its clients are mostly major global firms in pharmaceuticals, semiconductors and new energy areas. The company empowers various experiments and manufacturing processes through AI and automation, significantly accelerating their research, development and production cycles.

For example, protein evolution in the wild takes tens of millions of years to complete. Under the current biopharmaceutical research framework, a team of around ten PhDs might still require more than a decade to achieve comparable results. With the company’s AI technology, however, intelligent laboratories can keep iterating and autonomously optimising 24 hours a day, multiplying experimental efficiency and reducing costs—thereby freeing PhDs to focus on more meaningful, higher-value research. Similarly, in fields with high technological barriers to entry—like semiconductor manufacturing—where it once took researchers years of repeated experimentation to build up the [required] experience and know-how, it is now possible with the help of AI to achieve major breakthroughs in a much shorter time.

What impressed me most, however, was a small café tucked into a corner of their factory. Two dexterous robotic arms prepared coffee for guests, emulating the techniques of world latte art champions. They could produce a variety of beverages efficiently and tailored to customer preferences. For children who don’t like coffee, the system could even make customised milk tea. This AI café is capable of producing a thousand drinks a day, while the role of the [human] “manager” is essentially that of a logistics worker, whose only task is to ensure timely replenishment of supplies. The robotic café serves not only as a relaxation space for employees but also as a testing ground for innovation. Reportedly, a newly opened Luckin Coffee store in New York is making brilliant use of the AI café’s remarkable innovations.

Of course, some Chinese companies are also following their American counterparts into the field of general-purpose artificial intelligence, with DeepSeek a case in point. But here’s the core difference between the two approaches: US companies generally adopt a closed-source model, with even those that originally championed open-source now shifting away from it. Meanwhile, Chinese companies uniformly choose an open-source path.

If I understand correctly, China’s approach to AGI means, objectively, that it is continuously thwarting the monopolistic ambitions of its American peers. By encouraging the rest of the world to “hitch a free ride with China” [免费“搭中国的便车”], it brings more countries and developers onto the side of Chinese large models. Within the global AI development process, it is a strategy [布局] akin to “encircling the cities from the countryside” [Note: 农村包围城市 - Mao Zedong’s strategy during the Chinese Civil War of gaining strength in the periphery before challenging the centre]. When combined with the development of AI in vertical application domains described earlier, it is clear that this constitutes a highly competitive “combination punch” [组合拳] strategy.

II. The Rise and Fall of Bubbles in Artificial Intelligence

When new innovations emerge, bubbles are a common occurrence. According to the Gartner Hype Cycle, important new technologies follow a certain trajectory between their early emergence and widespread adoption: there is generally an initial hype that then cools, and is subsequently followed by steady development. At first, large amounts of capital and talent flow into the new industry, which bubbles with excitement and new ideas [充满希望和想象]. However, early cash flows and profitability are far from sufficient to justify the feverish valuations, so a bubble inevitably forms. Under certain circumstances, this leads to a period of sharp valuation corrections and large-scale capital withdrawal. Only after the initial hype has subsided [繁华落尽] can the gradual and substantive process of growth truly begin, bringing with it tangible results.

Which stage of the Gartner Hype Cycle are the AI booms in China and the US at currently? How far are we from a bubble bursting? The chart here is Gartner’s own assessment, provided for reference only.

The current AI bubble reminds me of the Dot-com bubble from 1999 to 2001. Back then, the Internet bubble also spread from California and Wall Street in the US to [the tech hub of] Zhongguancun in Beijing. It is said that some US$7 billion of “crazy investment” was [sloshing around] looking for projects that could “tell a story” [讲故事] on the capital markets, and web page clicks could be converted into money in the market.

At the time, as a graduate student, I persuaded a few engineering friends from Tsinghua University to plan a company that would sell “effective clicks” to Internet companies. We even pitched our idea to some entrepreneurs and venture capital funds, and ended up enjoying a few nice meals as a result. Not long ago, while moving house, I came across my business card box from around 2000—and one of the cards turned out to be Jack Ma’s.

At that time, the Internet was likewise a groundbreaking technological and economic innovation. Its diffusion brought the US economy enormous advantages as well as financial benefits. The entire world was paying high prices for American Internet hardware, software and services. During those years, US economic growth was not debt-driven but genuinely technology-driven. The Clinton administration even achieved an annual fiscal surplus. Of course, the animal spirits of capital are prone to excessive exuberance. In 2001, the US witnessed the bursting of the Internet bubble, with the NASDAQ index [eventually] plummeting by more than 70 per cent from its peak.

Today’s tech bubble is concentrated mainly in the fields of artificial intelligence and digital currencies. [In both], the associated investment frenzies have received substantial political support.

The right-wing tech bros, [科技右翼势力] represented by Silicon Valley’s Peter Thiel, have overpowered Wall Street and the Federal Reserve, becoming the main financiers and sources of senior officials for the Trump administration. On the US West Coast, presidential backing is being used to outweigh the financial capital of Wall Street on the East Coast, with the recent stablecoin legislation serving as a signal of this shift.

The current frenzy may already be approaching the stage [the Dot-com bubble] reached in 1999: aside from Nvidia, which “sells shovels” [to the AI companies in this goldrush], most well-known AI companies still show little real prospect of profitability in artificial intelligence. Some firms have discovered that using AI to write code results in lower overall efficiency than traditional programmers’ manual work. Recently, many US companies have preferred investing borrowed money in digital currencies rather than developing their core businesses. [Meanwhile,] Nvidia has formed a financial closed loop with its major clients based on debt, valuations and orders—all phenomena typical of a period of inflated bubbles.

The US tech sector hopes for a singularity—a sudden moment of enlightenment where AI becomes god-like—that may never arrive. The real trajectory of AI capability growth may be a curve that gets perpetually closer to that goal, but never actually reaches it. The dream of superintelligent AI may, in reality, do little but provide a dose of copium [一剂安慰剂] to the US tech sector as it faces brutal competition from Chinese peers who are [in the process of] overtaking them in all sectors [全面赶超]. Should the market consensus one day recognise this point, it might be the moment when the current bubble collapses.

Of course, not all bubbles burst immediately. Given that US monetary policy is about to enter a looser period [of lower interest rates], this current bubble in artificial intelligence and digital currencies may remain for some time. My judgement is that during the 15th Five-Year Plan period, the speculative mania surrounding AI and the associated digital currency frenzy will reach [a state of] hysteria [癫狂], after which there will be a period of correction when the bubble bursts.

If this assessment is broadly correct, then investors in AI and digital currencies should begin reducing their positions at this high point, shifting instead to holding solid gold, silver and RMB cash. In a few years, when the tide has receded and the true picture emerges [水落石出], it will not be too late to invest at a reasonable price in those projects that have withstood the harsh winter [寒冬考验].

III. Artificial Intelligence and the Evolution of Institutions

Institutions are constantly evolving. Marxism teaches that the relations of production and productive forces must mutually adapt. Artificial intelligence has greatly increased productive capacity, yet existing relations of production and systems of distribution struggle to support such a leap in productivity. Whether in the US, China or other parts of the world, institutions everywhere face pressure to restructure themselves.

The advance of artificial intelligence is driving overall productivity towards the communist stage that classical Marxist theorists envisioned. The long-standing scarcity and inflation that have characterised human economic history are being replaced by abundance and falling prices. Meanwhile, labour is being supplanted by capital and technology. As a result, humanity itself is being divided into two categories: those who control artificial intelligence, and those who are controlled by it. The former enjoy long lives, high intelligence and exist in a state of rationality, health and authenticity; the latter are poor, ignorant, sickly and with short life spans, dwelling within an illusion carefully designed for them. Whether a nation’s system is genuinely socialist, nakedly capitalist, or a sham form of socialism is evident from one criterion alone: how much the two groups diverge and how separated are their lives.

The advent of industrialisation brought about a massive reset in employment: many traditional occupations disappeared, yet industrial machinery demanded greater numbers of well-trained and knowledgeable operators. For instance, horse-drawn carriages were replaced by automobiles and coachmen gave way to drivers, with the proliferation of cars creating far more driving jobs than there had ever been for coachmen—while also spawning new roles such as mechanics and highway workers. Artificial intelligence, however, differs from earlier waves of industrialisation in that it destroys jobs rather than displacing them. Across the humanities, social sciences, natural sciences and engineering, numerous positions associated with traditional professions will be replaced by AI and robots. Yet few new roles will emerge; even manufacturing and maintenance of robots will be done by robots.

To return to the earlier discussion on Sino-US competition in AI: in the US, general AI systems capable of writing essays, composing poetry, painting, making films and writing code are primarily replacing the work of arts and humanities graduates (indeed, programming may be seen as a form of writing in the language of computing). In contrast, although China’s AI can also compose poems, create art and draft texts, its greater impact lies in displacing the work of science and engineering graduates. Recently, a common refrain in public discourse is that studying humanities makes it difficult for one to find a job, so one ought to pursue science or engineering instead. From my understanding of the AI industry, this view could prove to be rather superficial.

My strong hunch [猜想] is that the global higher education bubble is about to burst. In the future, no matter which subject one studies at university, finding a job will become increasingly difficult, because the premium once conferred by accumulated knowledge will largely be offset by artificial intelligence. [General] ability [能力] will become more valuable than knowledge. If you have learnt the ability to interact effectively with people or the skill of providing emotional value [情绪价值] to clients, you might still manage to live with some dignity.

Therefore, in the age of artificial intelligence, employment is neither a basic right of survival [生存的基本权利] nor a fundamental duty of adulthood, but rather a privilege [特权]. Employment provides individuals with a sense of purpose and agency in social affairs. Genuine job opportunities, however, are becoming increasingly scarce. For the majority who cannot obtain such employment, their chief contribution to society lies in consumption. And for many people, their principal contribution will be their capacity to offer emotional value to others [为他人提供情绪价值]—to endow ordinary things with a sense of meaning. For the majority of people, so-called innovation will consist in their creative consumption of others’ products and the inventive ways in which they spend or enjoy their own lives [玩出新花样来].

I expect that the social order of the AI era will be characterised more by conflict than harmony. Classical Marxist writers envisaged that, once material abundance had been achieved, human society would distribute resources according to need. Yet in my view, this may underestimate the darker side of human nature. When most people’s labour is no longer required, while human nature itself remains unchanged, the likely outcome will be “distribution according to disruption” [按闹分配]—those who most alarm and unsettle the ruling class will receive greater government subsidies, benefits and privileges. In future, this may well become the strategic calculation [博弈策略] different groups adopt within society.

Therefore, in the AI era, competition between countries — especially among the great powers—may increasingly depend on whose institutions are more flexible and resilient. It may also depend on who can adapt to both the potential and the disruption brought by technology, as well as its shocks, through rapid institutional iteration and innovation.

As this new era dawns, an age-old debate reemerges before us, about whether the bottom-up logic of free expression and competition or the top-down instinct for effective control and regulation is more suitable. Alternatively, perhaps somewhere between the two, a balance might yet be found.

IV. How to Win the Sino-American AI Competition

The AI competition between China and the US has several different dimensions, including computational power, algorithms and data.

At present, the main focus of competition lies in computing power, which depends on high-end chips and the power supply. In previous years, the US sought to exploit its dominant position in the semiconductor supply chain to choke our development [卡我们的脖子]. Yet due to the improper design of its sanctions, Washington not only failed to throttle China, but instead spurred the creation of an entire alternative domestic supply chain. Since 2017, the general trend has been one of US strategic offensives in the chip sector forcing China to scale back [中国不得不战略收缩]. Recently, however, as indicated by China’s official rejection of Nvidia’s “scaled-down” [Note: 阉割版, literally “castrated version”] chips, the two sides have reached a strategic stalemate and China’s home-grown alternative supply chain has begun to take shape.

With a strong talent base and a new-model national mobilisation system [新型举国体制], we will probably move into a strategic counterattack phase in the chip sector [在芯片领域转入战略反攻阶段] within the next five years. Meanwhile, America’s development of computing power also faces a major bottleneck—constraints on its energy supply. As illustrated in this chart [depicting US net annual energy generation], US electricity generation growth has been sluggish for years. Democrats and Republicans are divided on how to expand America’s power capacity: Donald Trump’s slogan is “drill baby drill”, advocating aggressive exploitation of shale oil and gas resources; the Democrats, by contrast, call for large-scale development of solar, wind and battery storage. Should the US follow the Democratic path, its computing power could end up constrained by China’s dominance in the renewable energy supply chain.

Algorithms are fundamentally about human talent, and competition in this area is, in essence, competition for talent. [Our] relative weakness in computational power has spurred us to leverage our ethnicity’s innate talent [种族天赋] for mathematics, using algorithmic innovation to make up the difference—most notably exemplified by DeepSeek’s comeback during the 2025 Spring Festival. Photographs of winners of mathematics Olympiads alongside core AI research teams hint at a principle that would be somewhat politically incorrect in the US: fate has allotted different specialisations to different ethnic groups. AI algorithms are for Chinese people what sprinting and basketball are for Black people—a natural talent assigned to our ethnic group, which can earn astonishing market valuations. A series of recent mergers and acquisitions illustrates that, in this booming capital market, top AI developers can command staggering valuations, already exceeding the transfer fees of top NBA players.

A potential challenge here lies in the outflow of China’s maths and AI prodigies to the US—a factor that could ultimately decide the outcome of Sino–US AI competition. The vast disparity in income levels between the two countries means that top Chinese talent is easily drawn to America, undermining our own competitiveness. In the words of an American strategist: “In the 21st century, America will rely primarily on Chinese Americans to defeat the Chinese” [Note: we were unable to find a public source for this quote]. Hence, for some time to come, in the arena of algorithmic competition, the US will have fewer but higher-quality talents, while China will have a much longer “bench” [替补板凳], somewhat less elite yet with far greater numbers.

Ultimately, the decisive factors may be data and cash flow. Whoever’s AI incorporates the data of the more than six billion people outside China and the US, and can generate cash flow from as many of them as possible, will evolve and iterate more rapidly. Consequently, control over the intermediate space of the global market will be the key to victory in the long-term technological competition between the two powers.

Our planning must keep this strategic, global perspective in mind. For example, in the early stages of this competition, we should employ all available tools—state subsidies, financial support, diplomatic pressure—to encourage as many countries as possible to adopt Chinese AI technologies, including autonomous vehicles, domestic service robots, fully integrated smart homes, educational assistance systems, intelligent diagnostics and surgical systems, and even military operational systems. This should be approached with a land-grabbing [跑马圈地] mentality: securing market space in the intermediate countries at the lowest possible cost, even at a loss if necessary. We should continually encourage Chinese LLM developers to participate in open-source and free initiatives worldwide, offering comprehensive solutions that are high-performing yet affordable.

Just as China’s manufacturing sector delivered a price revolution in industrial goods for the world over the past thirty years, in the next thirty years, we must deliver a similar price revolution in the field of artificial intelligence.

READ MORE