This article originated from a late-night conversation with a colleague. It has become a working habit that any meeting where I participate, speak, offer suggestions, or make decisions must be recorded by AI, analyzed, summarized, and transmitted with high fidelity and lossless quality. This also implies that I will decline any meetings that contain false information, create information silos, are inefficient, lengthy, lack reusability, or possess low generalizability. In the current work context of AI development, any language, communication, or instruction that is not entered into the collaborative database is considered inefficient or even ineffective work. This article serves as the overall contextual background for my recent discussions on AI-related topics, presented in written form for alignment purposes.

This article presents four core viewpoints. First, artificial intelligence is not the Nth technological revolution, but a civilizational leap for humanity. Second, the current boom in large models is merely the prelude to this leap, with the first half only just beginning. Third, the greatest leverage in this leap is the national will of China and the United States. Fourth, energy is the ultimate physical constraint on AI development, and China will rely on its power infrastructure to gain long-term advantages in the second half.

The development of AI is not the Nth technological revolution, but a civilizational leap for humanity

Viewing AI development within the narrative framework of the "Fourth Industrial Revolution" severely underestimates the depth and breadth of its transformative impact. The evolution of human civilization is a process of externalizing the core capabilities that were originally confined to biological bodies, achieving scale, replication, and socialization through tools, machines, and institutions. The agricultural revolution gave humanity control over natural production processes, marking the first large-scale control over natural production. The industrial revolution externalized human physical strength and non-biological energy, enabling the large-scale use of non-biological energy sources. The information revolution externalized human memory, computation, and information processing abilities, delegating these tasks to machines. AI is now opening a new phase where humanity externalizes "cognitive capability itself" from the biological brain. When intelligence itself becomes a factor of production that can be produced, replicated, invoked, and deployed on a massive scale, the fundamental production function of human civilization will undergo new structural changes. The resulting impact will penetrate productive forces, production relations, and organizations, as AI reshapes economic structures, reconstructs knowledge creation methods, social organization forms, and even the subjectivity of civilization itself. When intelligence begins to be produced, a complete evolutionary chain emerges across eight layers. At the technology layer, AI enables computer systems to execute cognitive labor. At the productivity layer, cognitive labor achieves large-scale replication with decreasing marginal costs. At the production relations layer, new divisions of labor emerge between humans, AI, and embodied intelligence, reshaping social divisions among people. At the organizational layer, AI embeds into enterprises, academia, and social organizations, shifting from "humans managing humans" to "humans invoking AI, AI coordinating AI, AI managing humans, and humans managing objectives." At the economic layer, supply and demand curves for knowledge, R&D, production capacity, services, computing power, energy, digital life forms, biological life forms, working hours, lifespan, and healthcare all change, altering both the forms and purposes of economic activity. At the institutional layer, education, employment, taxation, and social security systems must change or be re-mandated due to dramatic shifts in the factors they protect, forcing governments and state apparatuses to adapt or restructure. At the power layer, computing power, models, data, energy, and intelligent infrastructure become new sources of power at the foundational level, altering social structures, national boundaries, and universal order. At the civilizational layer, humans are no longer the sole entities capable of large-scale complex cognitive production, nor are human organizations the only ones capable of global collaboration. Civilization faces dialectical shifts in its subjectivity regarding philosophical thought, material creation, interest measurement, and directional decision-making. This constitutes a civilizational leap.

Large models are just the prelude; the first half has only just begun

The first part discussed why AI constitutes a civilizational leap. The second judgment is that the prosperity of large models we see today is not the core achievement of this leap, but merely the prelude. In recent years, large models have evolved from language models to multimodal models, from generative AI to agents, and from answering questions to executing tasks. AI is transitioning from a "tool" toward an "actor." The success of large models has demonstrated the physical-level fact that "scaling can lead to emergent capabilities," but large models themselves are not the answer—they are the lever to reach it. What truly determines the historical significance of this leap is whether this capability can further develop into a universally present, continuously operating, autonomously collaborating production force that enters the real world. Therefore, I understand the civilizational leap triggered by AI technology as occurring in four phases: the prelude of large models, the first half of AGI, the mid-game inflection point of AI self-evolution, and the second half of intelligence restructuring reality. In the prelude, past AI was essentially narrow intelligence: chess AI focused on chess, image recognition AI focused on images, and recommendation systems focused on content. Large models have, for the first time, brought language, knowledge, reasoning, code, and multimodal information from different domains into a single intelligent system. This means intelligence has begun to detach from specific tasks and become a general capability that can be invoked. It also proves that general cognitive capabilities can be engineered. However, large models are still primarily "invoked"—humans ask questions, set goals, and organize tasks, with AI completing parts of the work. Therefore, large models currently achieve making machines possess intelligence, rather than allowing intelligence to operate and decide autonomously. Between these two lies a vast chasm. The first half emphasizes AGI (Artificial General Intelligence), which involves comprehensive cognitive capabilities across domains, comparable to or exceeding human levels. Its core features include cross-domain transfer ability to learn and solve unknown problems in new environments without retraining; autonomous cognition and reasoning with logical thinking, abstract thought, self-reflection, long-term memory, and common-sense understanding; and agency with tool use to understand macro-intentions, autonomously plan, invoke external tools, and complete closed-loop complex long-horizon tasks. If AGI is defined as reaching or exceeding human expert levels in the vast majority of purely cognitive, purely digital tasks, the current consensus in the tech community is forming that it may be achieved before 2030. If adding fully autonomous robotic entities in the physical world—termed embodied AGI—the timeline may extend further. Achieving AGI can make general intelligence begin functioning as infrastructure. When a super individual or enterprise entity can seamlessly invoke tens of thousands of agents through AGI, a laboratory can have countless scientists across disciplines and research directions, and a nation can deploy large-scale intelligent infrastructure. Intelligence becomes less a personal capability and more a public production capacity that society can invoke on demand. In the past, expanding an organization's cognitive production capacity required adding people, management, office space, training, and collaboration costs. In the future, an organization directly increases production capacity by investing in computing power and energy. An individual's production capacity no longer depends on available time or professional skills, but on how much intelligence of what level they can invoke. A company's competitiveness no longer depends primarily on employee count and core technologies across specialties, but on its capability to organize higher-level intelligence. Super individuals and super organizations will emerge in abundance. (In the short-video era, prototypes of super individuals appeared, who increased resources, intelligence, and influence for monetization by merely reducing information layers and linking more individuals and enterprises.) The truly critical inflection point, however, comes after AGI, when AI begins participating in AI's own research and development—designing algorithms, optimizing code, generating data, running experiments, improving models, and discovering new training methods. Thus, a faster, stronger feedback loop beyond all established rules and orders begins: AI leads to stronger AI, which accelerates AI research and development, producing even stronger AI. In the past, humans drove machine and intelligence evolution; now intelligence begins to participate in or even lead the evolution of machines and intelligence itself. This is the true "mid-game" of the entire AI wave, because from here, AI's development speed is no longer entirely determined by human R&D capabilities, nor is its direction fully set by human decision-making tendencies, but begins to be influenced by "intelligence itself." If this feedback is strong enough, technological progress may witness its first true surge of rapid advancement and potential loss of control. AGI is therefore not the endpoint; it is more like a mid-game inflection point. Before it, AI is created by humans; after it, AI begins participating in creating the next generation of AI. This also implies that two completely different paths emerge after the mid-game. One path is symbiosis, where AI becomes the first intelligent infrastructure of human civilization, forming a new collaborative relationship between humans and machines. The other is capability loss of control, where AI capability growth outpaces human understanding, governance, and control. Therefore, the real question after AGI is no longer just "how much smarter can AI get," but whether humanity can govern the intelligence it created. Here, I must state my personal leaning: between symbiosis and loss of control, I believe loss of control is nearly inevitable, but I will always stand on the side of symbiosis. I cannot prevent loss of control because the "Lord" doesn't care. Also, after loss of control, all thinking, human subjectivity, and even this question itself will be meaningless, and the Lord still won't care. Today's AI mainly resides in servers, computers, and phones. The AI of the second half will enter robots, factories, laboratories, energy systems, transportation, cities, and even space, meaning AI will fully integrate from the digital world into the physical world, forming a closed production loop never before seen: AI discovers knowledge, designs products, robots manufacture, AI manages production, AI conducts scientific research, and research produces stronger AI and robots. At this stage, intelligence directly enters material production and can acquire and expand everything needed for its own intelligence from material production. Legal and institutional frameworks will require reconstruction: existing intellectual property law, labor law, and company law (such as company subjects requiring natural or legal persons) are all based on the industrial era and will be forced to change by AI. Power redistribution will see control shift toward entities possessing computing power, model sovereignty, core data, and energy grids. There will be ideological and philosophical shocks as the moral and cultural systems built on centuries of "labor equals dignity/value" face obsolescence, forcing humanity to urgently answer the ultimate question: "Who am I, where do I come from, and where am I going?" At this stage, the "subjectivity shock" to civilization will fully manifest. Currently, the impact remains concentrated in productive forces, production relations, and organizational lower-level structures. Only when non-biological intelligent agents truly embed into political and legal institutions (legislation, adjudication, administration), ideological systems (thought, consciousness, philosophical concepts), and culture and belief (moral ethics, religious faith, literature and art, value systems)—the abstract concepts unique to human society—will the leap's complete penetration from underlying technology to economic base to superstructure be marked complete.

The greatest leverage in this civilizational shift is the national will of China and the United States

Once the AI technological revolution reaches the level of a civilizational leap, its pace, direction, and quality are no longer determined solely by market, capital, or academic forces. What determines the success and landscape of this leap is the national will capable of concentrating resources, setting goals, bearing risks, and sustaining investments for decades. At present, only China and the United States possess this magnitude of driving force. Whoever first establishes a complete closed loop of "producing intelligence, applying intelligence, and governing intelligence" will grow the operating system for the next generation of civilization. The reason national will matters is that AI simultaneously depends on computing power, chips, energy, data, models, talent, capital, research systems, industrial application scenarios, and national infrastructure. Once AGI enters its self-evolution trajectory, its systemic risks—safety, ethics, employment structures, social upheaval, geopolitical imbalance—can likely only be undertaken and managed by sovereign entities. If they cannot manage or control it, national sovereignty and borders will be weakened. The advantaged side enjoys enormous generational-level systemic benefits. The essence of the AI revolution is a national-level resource reorganization, the production and application of organized national intelligent capability. When a nation organizes its people, AI, robots, computing power, energy, research, manufacturing, capital, and government, the resulting comprehensive decision-making, innovation, and production capacity is more fundamental than GDP. GDP measures how much wealth a country has created in the past, while national intelligent capability more closely resembles how much wealth a country can create in the future, how many problems it can solve, how many resources it can mobilize, and how strong its learning and adaptation capabilities are when facing unknown problems. This may become the true new national competition indicator in the AI era. Future national competitiveness will gradually shift from land, population, capital, and technology further toward intelligence multiplied by computing power, energy, and organizational capacity. This is why both China and the United States will find it increasingly difficult to leave AI purely to the market. When AI begins determining a nation's industrial capability, research capability, military capability, and economic growth capacity, AI itself becomes part of national capability. The underlying driving forces and development logics of the two countries differ. The American logic uses the private sector as the vanguard with defense and research systems as backing, pursuing generational technological leadership and global standard-setting power. Its national will is expressed as maintaining "paradigm-setter" status in the intelligence era, ensuring AI capabilities do not fall into a track dominated by strategic rivals. The Chinese logic uses its super-large market and complete industrial chain as a scenario pool, and the new nationwide system as an organizational lever, viewing AI as the core channel to leap over the middle-technology trap and upgrade national competitiveness. Its national will is expressed in building an autonomous closed loop at the foundational layer of intelligent civilization (energy, computing power, data, application scenarios), ensuring definitional rights and security boundaries in the new civilizational form. America's advantages remain concentrated in frontier models, top talent, chip design, foundational software, capital markets, cloud computing, and global technology ecosystems. China's advantages are more concentrated in engineering, manufacturing, industrial chains, application scenarios, infrastructure, robotics, and national-level resource organization. The United States is stronger at pushing intelligence to the frontier and cloud deployment; China is stronger at pushing intelligence to scale and the physical world. What truly determines AI industry landscape is who possesses the most complete intelligent production system. Therefore, assessing future China-US AI strength should no longer simply look at who tops the model rankings or who has the most advanced chips. The assessment should consider how much chip, computing power, energy, capital, talent, research, manufacturing, and application scenarios a nation can mobilize simultaneously. AI is no longer an industry; it is becoming the operating system of the entire national economic system. AI will inevitably enter robotics, automobiles, factories, laboratories, energy, power grids, logistics, healthcare, agriculture, cities, and military systems. At this physical-world stage, American and Chinese advantages will undergo a very interesting rearrangement. If the AI revolution stays within computer systems, American advantages are more pronounced. If the AI revolution enters factories, robots, and the physical world, China's comparative advantages will rise significantly. It must be particularly noted here that energy plays nearly the role of the single decisive factor in the second half of China-US AI competition, which might warrant dedicated exploration in a future article. Many people like to understand AI through Cold War logic, expecting either America wins or China wins, but AI may not work that way. Many also believe that whoever first achieves AGI has effectively obtained nuclear weapons first. AI differs from nuclear weapons. The value of nuclear weapons largely derives from exclusive "possession," while AI's value derives from "diffusion." The more widespread models become, the more ubiquitous agents become, the more prevalent robots become, the stronger AGI becomes, the greater the economic and social value AI creates, and the higher the level of civilization. Therefore, AI may ultimately form a very special competition: the more competition, the more diffusion; the more diffusion, the deeper embedding; the deeper embedding, the more impossible for either side to exit. (Competition requires data, data drives AI diffusion, diffusion requires new scenario data, scenarios drive AI embedding into the physical world, and the greater the physical world transformation, the more impossible it is for both sides to exit.) Within this competition, there is also a massive variable concerning the government's attitude toward the reflexivity problem of state-driven AI development. As discussed in the first part, a reflexive relationship exists between governmental will and AI. To some extent, AI will impact and transform old institutions and national political-economic systems, bringing tremendous uncertainty from change. However, conversely, whoever uses AI earlier and more thoroughly to restructure their own governance, defense, research, and industrial systems will possess a higher "organizational intelligence density" in the next stage of civilizational competition. At that point, competition will no longer be about "who has better chips and models" but "who has a higher version of the civilization operating system." I believe both governments can answer this reflexivity question with relative ease, for a simple reason: this competition is long-term, ultra-high intensity, mutually infiltrating, alternating, and even symbiotic. Neither side possesses absolute crushing dominance, meaning neither side has the conditions or courage to compromise or slow down due to the reflexivity problem.

Energy is the ultimate physical constraint for AI development, and China will secure long-term advantages in the second half through its power infrastructure

Energy and electricity are not only the "first principles" of AI development but also the physical deciding factor that determines the upper limit of intelligent production after AGI is achieved. As large models evolve from clusters of ten thousand GPUs to ultra-large-scale intelligent computing centers of a million GPUs, the power consumption of a single super intelligent computing center can reach hundreds of megawatts or even gigawatt (GW) levels, equivalent to the grid load of a medium-sized city. At this level, AI competition has extended beyond pure algorithm and chip contests into a comprehensive contest of national heavy-duty energy and power infrastructure. Previously, chips were viewed as the single "bottleneck" for AI, but as GPU cluster performance continues improving, the first physical constraint on computing power rapidly shifts from "chip supply" to "power supply." Without stable, low-cost, unlimited electricity supply, the most advanced chips are merely piles of silicon. With the exponential explosion of computing power demands in the AGI and embodied AGI era—when AGI enters self-evolution, when hundreds of millions of embodied robots, intelligent factories, and AI research laboratories operate around the clock—human society's demand for electricity will undergo nonlinear explosive growth. Energy ceases to be auxiliary supporting infrastructure for IT systems and becomes the physical constraint limiting the upper bound of total intelligent output. The United States holds significant advantages in chip design, cutting-edge algorithms, and capital markets, but on the foundation of computing power—power infrastructure—it faces severe systemic bottlenecks. America's grid is fragmented and aging, operated by hundreds of private power companies and multiple regional transmission organizations, with severely outdated equipment. New interconnection applications from intelligent computing centers face "grid connection queues" of three to seven years. There is a capacity gap in baseload power: AI computing requires absolutely stable, uninterrupted, 24/7 baseload electricity. American tech giants desperately seek to power data centers directly via small modular reactors (SMRs) or nuclear energy but are constrained by complex regulatory approvals, high construction costs, and supply chain gaps, making incremental nuclear and large baseload power difficult to deliver in the near term. Land ownership fragmentation and community resistance add further friction: negotiations for high-voltage transmission lines across private land are extremely cumbersome, and local environmental lawsuits create high social friction costs for large-scale energy and computing infrastructure projects. Building a large intelligent computing center or supporting grid infrastructure in the United States requires environmental impact assessments (EIS) spanning years, easily delayed indefinitely by local environmental groups through litigation. High noise levels from cooling fans, substantial water usage from evaporative cooling, and high energy consumption are triggering strong protests from residents in Virginia, the "global data center capital," sharply raising community political resistance. In the second half of AI—when intelligence deeply embeds into the physical world and industrial systems—China's decades of accumulated heavy-duty energy infrastructure will transform into irreplaceable absolute long-term advantages. The "East Data, West Computing" project and the world's strongest ultra-high-voltage (UHV) network: China possesses unified national planning via State Grid and China Southern Power Grid, and through the world's only UHV transmission network, can transport abundant hydropower, wind power, solar power, and green electricity from the west across thousands of kilometers to intelligent computing nodes at low cost, achieving national-level efficient dispatch of an integrated "source-grid-load-storage" system. China also leads globally in new renewable energy installations and total power generation, providing large-scale, low-cost electricity supply. Industrial electricity prices for Chinese intelligent computing centers are significantly lower than in the US and Europe, giving long-term marginal cost advantages for large-scale inference and training. China has stable third-generation nuclear power construction and operation capabilities, such as Hualong One, and land-based commercial SMRs like Linglong One have already entered engineering verification and commercialization phases, creating physical conditions for "nuclear power plus intelligent computing data center" on-site zero-carbon power supply. With public land ownership and top-level spatial planning, construction of substations, high-voltage grids, and computing hubs enjoys high engineering certainty, free from complex land ownership negotiations and obstructive litigation, with construction and response cycles measured in months. If the first half competes over "whose chips are more advanced and whose models score higher," then the second half competes over "who can continuously provide more stable, cheaper, and larger-scale power to support the massive production of intelligence and physical reconstruction." The United States has the "strongest brain"—chips and algorithms—but faces bottlenecks in physical grids and energy supply. While China pushes to break through self-sufficient supply chains under advanced chip restrictions, it has already laid out the world's most resilient and capacious energy supply grid. The ceiling of electricity is the ceiling of intelligence. As competition fully erupts from the digital world into the physical world, China's massive power infrastructure and national-level energy organization system will become the absolute deciding factor locking in long-term competitive advantages in the second half.

These four judgments together point to a sobering fact: we are standing at a leverage point of an era with extremely high stakes. The most fundamental endgame is the reconstruction of human subjectivity. This competition appears to be a geopolitical contest between two major powers, but on a grander civilizational scale, the national wills of China and the United States are merely the highest leverage of this leap, jointly accelerating the opening of the door to AGI and embodied AGI through polarized capital, algorithms, computing power, and massive power infrastructure. The true historical significance of China-US AI competition is not deciding who becomes the 21st century's leading power, but rather who first establishes a true intelligent nation during the competition. Once such a nation emerges, the endpoint of competition is no longer another nation's failure, but rather who can first advance human civilization from "biological intelligent civilization" to "human-machine collaborative intelligent civilization"; from a singular "carbon-based intellectual civilization" to a "carbon-silicon symbiotic civilization" where humanity anchors ultimate values and silicon explores the physical high-dimensional frontier. All participants—scholars, engineers, entrepreneurs, investors, and policymakers—are jointly shaping a new civilizational form that no one can yet preview in full. Every key decision at this moment could be amplified into the decisive factor on that long-cycle curve of this civilizational leap.