Artificial intelligence's role in reshaping the economy is no longer up for debate. What keeps policymakers, academics, and industry leaders up at night now is a more complex set of calculations. Can massive capital expenditure on computing power and models translate into genuine productivity gains at a macroeconomic level? Following a massive surge in supply, will we see prosperity or a prolonged period of deflation? As technology narrows the entry-level opportunities for young people, how can we avoid repeating history's "Engels' Pause"? And in the midst of increasingly complex global competition, how do we construct a new framework for international trade and rules in this digital era?
These were the central questions posed at the "New Paradigm of AI Economic Growth and Social Value Reconstruction" forum during the 2026 Inclusion·Outer Temple Conference. Held on September 9th, the forum brought together prominent voices including Liu Yuanchun, President of Shanghai University of Finance and Economics; Huang Yiping, Dean of the National School of Development at Peking University; Jiang Xiaojuan, a professor at the University of Chinese Academy of Social Sciences; Miao Yanliang, Chief Economist at CICC; Nobel laureate Thomas Sargent; and industry representatives. The discussion moved beyond simple demonstrations of technological capability, delving into growth measurement, inflationary volatility, job displacement, and the international division of labor to confront one fundamental question: how can AI transition from a fervent technological variable into a healthy, sustainable macroeconomic force?
01 The AI Economy Has Arrived, But the Growth Equation Remains Unsolved
"While the new form of the AI economy is established, the economics of AI hasn't truly begun," Liu Yuanchun stated, offering a sobering opening thought. Early stages of technological revolutions are often marked by intense speculation and capital bubbles. Mentions of AI on global earnings calls have skyrocketed, chip prices have soared, and infrastructure investment around computing centers has surged, even recently boosting global trade and exports. Yet, when looking beyond the micro-level noise to the macroeconomic fundamentals, a significant divergence of opinion is emerging.
Commercial institutions paint an encouraging picture of future productivity. Firms like McKinsey and Goldman Sachs project annual growth in total factor productivity (TFP) of up to 3%. However, within serious academic economics, conservative estimates, such as early models by MIT scholar Daron Acemoglu, suggested AI's average annual contribution to TFP over the next decade could be negligible, amounting to less than a few tenths of a percent. Why is it that everyone seems to be using AI at the micro level, yet the macro data has been slow to show results?
Huang Yiping cautioned that we are likely experiencing the classic "Solow Paradox" and the "J-curve effect" from the economics of technology. During the early phase of computer普及 in the late 20th century, productivity data also remained dormant for a long time. For a disruptive general-purpose technology (GPT) to integrate into real industrial systems requires lengthy organizational restructuring, process adaptation, workflow changes, and retraining. The initial phase not only fails to deliver an output boom but can also incur significant adjustment costs.
Sargent compared the current stage of AI development to the "Kepler stage" in astronomy. AI can now precisely fit patterns from astronomical amounts of data, but humanity has not fully understood or grasped the underlying economic and structural logic.
Addressing the concern of high capital investment and valuation bubbles potentially being unproductive, Liu Yuanchun reviewed the latest academic paradigms. He suggested that a bubble itself can act as a "coordination mechanism." If the bubble persists long enough to drive the accumulation of substantial general-purpose physical capital—such as computing networks, energy infrastructure, and digital assets—past a critical threshold, the economy could remain locked into a high-output equilibrium, even if financial valuations later correct sharply. The first consensus among economists is thus taking shape: the direction of technology is clear, but the pace, scale, and efficiency of translating it into real asset accumulation remain fraught with significant lags and uncertainty.
02 From Investing in AI to AI Making Money: The Crucial Shift to Creating New Demand
If the previous phase of the AI narrative was centered on capital expenditure, the current phase's lifeline is establishing a viable business model. Han Xinyi, CEO of Ant Group, pointed out that the AI economy cannot rely solely on capital investment, and AI applications cannot just focus on improving existing efficiencies. If companies treat AI merely as a tool for cost reduction and efficiency gains, making existing processes faster without creating any incremental supply or demand, the inevitable result will be severe supply-demand imbalances and a disconnect between investment and output. Enterprises must move beyond simple cost-cutting to genuine value creation.
Li Zhenhua, President of the Ant Group Research Institute, also highlighted the divergence within the industrial chain. The path to monetization is clear for upstream computing power and chips, but midstream large models and downstream applications still carry heavy commercial pressure. Whether AI's business narrative can hold up depends entirely on whether downstream applications can integrate deeply into real industries and create new value loops.
Zhao Bo from Peking University presented a logic for growth based on "expanding transaction boundaries." AI's core economic value lies in drastically lowering the barriers to acquiring professional skills and reducing the cost of innovation and experimentation. This means it's not about making original tasks 10% faster, but about making things that were previously too costly, too complex, or impossible to scale, productizable, purchasable, and tradable for the first time.
Shifting focus from efficiency to creating incremental value is the only path to navigate past the cliff of capital expenditure.
03 Answering the Musk Question: Deflationary Pressures and the Narrowing Entry Point for Youth Employment
As supply capacity explodes exponentially due to AI, a sharper macroeconomic paradox emerges: if machines produce faster and faster, and more and more, will people have sufficient effective income to consume? Huang Yiping characterized this potential macroeconomic imbalance as a long-term "strong supply, weak demand" scenario. If the benefits of technology flow disproportionately to capital owners while the return on labor remains suppressed, aggregate demand will fail to keep pace with supply expansion, breaking the traditional economic cycle.
From a financial and nominal variable perspective, Miao Yanliang, Chief Economist at CICC, offered a penetrating analysis. He looked back at history, noting that both the Second Industrial Revolution and the Information Technology Revolution were accompanied by prolonged periods of deflation. "Say's Law"—that supply creates its own demand—often fails during periods of rapid technological change because income inequality and differing consumption propensities can block the natural reconciliation of supply and demand.
Miao proposed four forces through which technological revolution impacts inflation:
Bottleneck and Expectation Effects (short-term inflationary): Prices of scarce elements, such as memory chips and high-end computing power, surge initially. However, these bottlenecks typically last only three to five years, and their peak is gradually passing.
Supply and Substitution Effects (long-term price suppression): AI technology diffuses extremely quickly and directly takes over high-level intellectual tasks, with the substitution effect of capital for labor being far greater than in the past. Since capital owners have a lower marginal propensity to consume, a declining share of labor compensation will suppress prices over the long term from the demand side.
"In the long run, the deflationary or even disinflationary pressure could be greater than in previous industrial revolutions," Miao cautioned. If this trend is confirmed, traditional monetary policy tools may face a liquidity trap, and macroeconomic regulation must shift towards deep fiscal policies, redistributive tools, and labor safety nets.
Behind the deflationary pressure lies a structural fracture in the labor market. Unlike previous IT revolutions that replaced middle-skill jobs while preserving high-skill and low-skill roles, AI exposure in this round is directly correlated with income level. High-income white-collar jobs in finance, law, consulting, and software development are the most exposed.
Miao noted that AI has not yet caused a major jump in the macro unemployment rate, primarily because its replacement of established jobs is a gradual "task decomposition". However, at the entry point of corporate recruitment, the career paths of young people are being severely squeezed. Demand for junior roles in investment banks, law firms, and software development has plummeted. This narrowing of the "career entrance" for young people disrupts the mechanism of human capital accumulation, potentially leading to long-term "scarring effects."
Huang Yiping drew a parallel to the "Engels' Pause" following the First Industrial Revolution, where output expanded exponentially but ordinary workers' wages stagnated for decades. Lu Ming and Xing Ziqiang replied to Musk's question, pointing out that AI will not easily eliminate all professions, but it will thoroughly reconstruct the core of jobs. Standardized, codifiable "functions" (机能) will be quickly taken over by models, while "human capabilities" (人能) such as empathy, organizational skills, and value judgment will be the key to survival for workers. The so-called "employment-first" approach is not about hindering technological evolution, but about designing institutions that enable more ordinary people to "do more" because of AI.
04 The New AI Economy's Final Test: Institutional Capacity and "Born Global" Dynamics
While individuals face the pressure to adapt their human capabilities, macro-industries and corporate structures face a different set of challenges related to rules and norms. Jiang Xiaojuan, a professor at the University of Chinese Academy of Social Sciences, argues that the AI economy is decisively ending the traditional path of globalization. The multinational model of the manufacturing era was a "sequential" process, moving from technology introduction, digestion, and absorption, to local validation and strengthening, before finally going abroad.
But in the AI era, technology and products are "born global." AI innovation is fundamentally reliant on algorithmic models and global digital networks, with extremely low marginal costs for model replication and distribution. From day one, it targets global developers and markets.
Jiang illustrated this paradigm shift using examples from Chinese industry:
The ecological symbiosis of open-source large models: Downloads and calls for Chinese open-source models on major overseas open-source communities have grown explosively, breaking the past industry norm where international giants offered "top-tier closed-source, low-tier open-source" models. High-performance open-source supply now provides a public good for the global tech community.
The reshaping of vertical division of labor in AI4S and "selling seedlings": In vertical fields like AI-driven drug discovery, Chinese companies have developed explosive capacity in target identification and molecular design. However, because the pace of downstream clinical digestion differs, overseas licensing deals for Chinese innovative drugs exceeded $100 billion in the first half of the year. This "selling seedlings" style of going global reflects how AI is restructuring the traditional global division of labor in biomedicine.
Leveraging massive industrial data deficiencies: China's vast and complex manufacturing sector has accumulated a huge amount of valuable industrial "error and defect data," allowing vertical industrial large models to undergo high-intensity domestic trial-and-error and then be rapidly exported to the world.
However, Jiang also stressed that being "born global" not only means exporting technology but also places high demands on domestic institutions, corporate awareness, and international competition strategy. "From their inception, companies must build high-standard global compliance systems," she stated, covering cross-border data flows, privacy compliance, algorithmic ethics, and intellectual property. Companies cannot rely on the past "domestic first, international later" approach based on luck. In terms of trade philosophy, China is shifting from seeking special and differential treatment as a developing country to establishing "horizontal competition and cooperation" as an equal with developed economies, learning to build mutually beneficial economic ecosystems using mature "two-handed" thinking.
Conclusion
From the arms race of capital expenditure to a rigorous examination of the macroeconomic ledger, this year's Outer Temple Conference has pressed a rational reset button on the fervent AI frenzy. What determines whether AI can become the next historic engine of economic growth is never solely the scale of computing clusters or model parameters, but rather a deeper systemic adaptation: Can micro-level algorithmic efficiency traverse the lag curve and transform into real-economy TFP? Can distribution systems and fiscal tools effectively manage deflationary pressure and broaden new value entry points for the younger generation? Can industry successfully navigate the challenges of compliance and cooperation to seize the "born global" dividend of the era?
As Han Xinyi and Huang Yiping noted at the forum's conclusion, AI's ultimate value cannot be self-justified by valuations on a balance sheet. It must ultimately be validated by genuine macroeconomic growth and measured by the tangible benefits experienced by a broader workforce.