The recent synchronized surge in AI-related stocks across both Chinese and US markets has solidified this sector as the primary investment theme for the second quarter of 2026. A notable characteristic of this rally is the significant outperformance of upstream hardware companies compared to their downstream application counterparts.

This rapid expansion in capital expenditure, coupled with rising valuations, naturally raises questions about a potential bubble. The central question is whether downstream applications can achieve sufficient productivity gains to justify these substantial upfront investments and generate sustainable long-term returns.

The primary mechanism through which artificial intelligence enhances efficiency is task automation, which involves substituting human labor to reduce corporate costs. This creates an unavoidable tension: the same process driving potential economic gains is also the source of significant anxiety over technological unemployment. Recent developments have seen AI evolve from simple conversational tools to more autonomous "agents," reinforcing optimistic market sentiment while simultaneously intensifying public fears about being replaced.

A critical issue is whether an AI bubble and widespread job losses are intrinsically linked. If AI spending is focused only on expanding capacity and inflating valuations without delivering significant labor substitution or efficiency gains, a bubble could form without causing severe employment disruption. However, if AI deeply integrates into production processes, rapidly displaces a large workforce, and supports capital returns, then the market might not be in a bubble, but society would face a substantial unemployment shock. This creates a difficult policy dilemma where achieving high returns on capital and avoiding job losses appear to be competing objectives.

Artificial intelligence is not solely a tool for replacement; it can also potentially enhance human capabilities and create new opportunities. This can happen through the development of entirely new industries and business models, which in turn can boost economic growth and overall employment. The relative strength of AI's labor-displacing versus its employment-creating effects will be a key factor for economists and investors to understand, as it directly impacts assessments of AI asset valuations.

Evaluating Machine Substitution of Human Labor Through Factor Endowment

Concerns about machines making human labor obsolete are not new. Since the Industrial Revolution, economists have grappled with this issue, from Ricardo's work on labor-saving technology to Marx's critique of machinery's impact under capitalism. Keynes highlighted the concept of "technological unemployment," which has now become a common topic in mainstream discussions as AI evolves.

The effect of machinery on employment can be analyzed from both supply and demand perspectives. On the supply side, machines can either substitute for labor, reducing the number of jobs, or complement it, where increased investment in machinery leads to more jobs. On the demand side, efficiency gains from technology are expected to lower production costs, which in turn can boost incomes and consumer demand, potentially creating more jobs even if the labor intensity of each unit of output falls. If demand expands enough, overall wage income can be maintained or grow despite efficiency improvements.

The fundamental difference between AI and previous technological advancements is its capacity for "intelligence," specifically its ability to perform cognitive tasks. However, the underlying principles governing machine-human interaction remain similar. The impact of AI is exerted in three key areas: the replacement of human workers, the augmentation of human capabilities that lead to new types of employment, and a general rise in total factor productivity. This last effect can increase societal incomes and fuel demand for labor across the economy, a crucial element often missing from simpler forecasts of AI's impact.

Standard economic models emphasize that factor endowments drive specialization. Developed economies, with abundant capital, tend to produce capital-intensive goods and see capital replace labor. In contrast, developing economies with large labor forces use labor to substitute for costly capital. The relative price of labor to capital is a key variable. If the cost of labor rises relative to the price of automation equipment, businesses are incentivized to substitute. For instance, while comparing these costs is straightforward, it is more challenging to apply to AI because its cognitive outputs are not easily converted into monetary terms.

Analysis of the industrial robotics market offers some useful parallels for understanding cognitive automation. In both China and the US, the price of industrial robots has been steadily declining relative to average manufacturing wages, creating a clear economic incentive for automation. However, the employment effects depend on why these price changes are happening. If a shrinking labor supply (as seen in aging populations) pushes up labor costs, the subsequent robotization is demand-driven and is unlikely to create unemployment. Conversely, if technological progress itself lowers the cost of robots, the automation it spurs is supply-driven and does displace workers. The current wave of AI-driven automation coincides with demographic shifts that are reducing labor supply, making it potentially well-timed to mitigate labor shortages, particularly in aging economies.

The international comparison reveals different incentives. Labor costs are considerably lower in China than in developed nations. As robots are tradable goods, their prices tend to converge internationally. Consequently, the economic rationale for replacing labor with robots is stronger in high-wage economies like the US than in China. This logic extends to AI, suggesting that the pace and intensity of cognitive displacement could be more profound in the US. However, China's status as the world's largest producer and user of industrial robots counters this simple cost-based logic. In 2025, China produced 773,000 industrial robots, a 28% annual increase, and in 2024, it accounted for 54% of global installations. Perhaps more tellingly, China's robot density in manufacturing surpassed that of the US in 2020 and the gap continues to widen.

The Role of Economies of Scale in Leveraging AI's Potential

While the price of robots may be uniform globally, economies of scale and local network effects offer significant competitive advantages. China’s massive installed base of robots creates a vast ecosystem that lowers the total cost of ownership. Spare parts are cheaper, service networks are denser, and a wealth of application data improves predictive maintenance, reducing downtime and operational costs. These factors make automation a rational economic decision for Chinese firms, even with lower labor wages, by giving them a significant cost advantage over their US counterparts.

In the AI era, the economies of scale of the manufacturing sector combine with the network effects of the digital economy, setting the stage for embodied intelligence and humanoid robots. The "creative destruction" described by Schumpeter is visible here, as AI simultaneously creates and destroys jobs. The benefits of AI, however, are not limited to substitution. AI is a powerful tool for augmenting human capabilities, often by lowering the skill barriers to entry. Historically, technological progress has created employment opportunities for less-skilled workers. Before the Industrial Revolution, crafts like spinning and weaving required long apprenticeships; technological advances later opened these jobs to ordinary workers.

The gig economy is a modern example of this augmentation. Ride-hailing drivers, delivery couriers, and live-stream sellers require far less prior training than the taxi drivers of the past who needed deep knowledge of street layouts. This is enabled by a platform’s network effect, where a single worker can serve a huge customer base. AI is set to accelerate this trend toward more flexible employment by simplifying cognitive tasks and lowering the costs of collaboration, reinforcing the shift from traditional fixed employment to a more gig-based structure. This could lead to a rise in the "one-person company," where individuals leverage cloud services and AI tools to operate independent businesses.

These changes are likely to lead to a structural differentiation in the labor value. As AI lowers the cost of information retrieval and integration, it erodes the job security of the mid-skilled "cognitive middle class" that often relied on an information advantage. Conversely, we may see an increased premium on high-level cognitive and creative skills like conceptualization, resource orchestration, and innovation, while hands-on, practical skill sets may also retain their value.

The Macroeconomic Impact of Artificial Intelligence

There is a stark contrast between technology researchers and economists regarding the potential for AI to boost economic growth. While technologists are often exuberant, economists generally adopt a more conservative outlook. Predicting GWP growth rates above 20% per year is a common, but highly unrealistic, forecast from the most optimistic tech visionaries. Economists typically rely on historical analogs and attempt general equilibrium analysis, which considers the entire system's *interactions*, including price changes and resource reallocations, and tends to highlight real-world frictions that limit growth. In contrast, technologists often use a partial equilibrium framework that overlooks profitability constraints that are key to sustained investment.

Using a task-based growth accounting model, 2024 Nobel laureate Acemoglu predicts AI will only lift annual US productivity growth by 0.07 percentage points over the next decade. 2025 Nobel laureate Aghion's similar task-based estimates are 0.68 percentage points, while his historical extrapolations from the power and IT revolutions are between 0.8 and 1.3 percentage points. In its 2024 "AI Economics" report, CICC Research employed a "meta-task" model and estimated AI could raise China's annual GDP growth by 0.8 percentage points. Meta-tasks are abstracted, fundamental skill units that can be applied across various jobs, which makes them a better fit for modeling how AI agents actually function. However, quantifying these effects proves difficult. It's easiest to measure the substitution effect of standard tasks, which have hard wage and employment data. Modeling other AI effects, such as human-machine complementarity or the creation of entirely new tasks, is empirically difficult due to a lack of established parameters, leading many studies to over-simplify and focus only on AI's job-displacing effects.

The growth impact of AI can be viewed through three possible channels: labor augmentation, labor substitution, and technological innovation, each with distinct constraints. If AI primarily augments labor, its potential to drive growth is fundamentally limited by the size of the labor force. This is problematic for economies like China's, which faces a shrinking working-age population. If AI primarily substitutes for labor, it can effectively expand the effective labor supply and offset demographic decline, but only if the displaced workers are quickly re-employed. If not, idle labor weakens the overall economic boost. If AI primarily accelerates technological innovation, the growth constraint becomes the economy's overall innovation efficiency, making scale and network effects advantageous for large economies. The question then rests on the speed of AI's diffusion.

Historically, general-purpose technologies take a long time to fully penetrate the economy. The adoption of AI may be faster, but it still faces significant barriers. First, there's scale diseconomies. The training of AI models shows diminishing returns, a constraint that better algorithms can only partially relieve. The physical bottleneck of matching AI with real-world operations also raises questions about the economic viability of embodied intelligence. Unlike traditional internet services, the variable costs for AI model inference cannot approach zero. Second, there are input constraints like a well-organized supply of high-quality data, and the availability of electricity. While China is well-positioned to leverage cost-effective clean energy, trade protectionism could limit the global spillover of this advantage. Third, institutional governance and geopolitics can create anti-scale effects. Highly regulated sectors like medicine and finance impose strict safety and liability rules that slow down AI adoption, while geopolitical competition can hamper cross-border scaling. Fourth, negative externalities are a concern. AI facilitates the rapid spread of deep-fakes and undermines trust in information. It also poses new challenges for intellectual property rights, potentially disadvantaging *fundamental* research in favor of incremental optimization. Perhaps most alarmingly, the potential use of AI in autonomous drone and robot warfare presents a serious global threat. A governance gap emerges as nationalistic competitive pressures encourage nations and developers to prioritize strategic advantage over international safeguards, a problem that history shows is often only addressed in response to a crisis.

Towards a Socially Beneficial Technology: Strengthening the Safety Net and Improving Income Distribution

Artificial intelligence is a general-purpose technology with immense potential to boost growth and welfare. However, its path is a systemic one, shaped by economic, ethical, and governance forces, not just technological progress. A purely supply-side focus on efficiency misses crucial demand-side dynamics. If AI drives mass unemployment, it will create a demand gap that undermines the very economic growth AI was meant to deliver.

History offers some reassurances: long-run technological progress *has* not led to permanent mass unemployment, thanks to self-correcting market mechanisms and public policy. Productivity gains do not occur uniformly across all sectors. As productivity rises in one sector, it releases workers who are then absorbed by other sectors that are less productive but have unmet demand, as consumer needs are diverse and satiable. This adaptation process explains why AI's "bottlenecks" are a double-edged sword. They limit the speed of efficiency gains, but they also prevent rapid mass unemployment.

Wage flexibility provides another market stabilizer. As machines displace workers and put downward pressure on wages, the relative cost of robots to labor increases, reducing the incentive for further substitution. However, if rigid labor markets or social policies prevent necessary wage adjustments, the resulting imbalance will translate into higher unemployment rather than price stability. A concentration of income due to labor replacement can lead to weak aggregate demand, reducing the overall demand for labor and hurting capital returns, as capital and labor depend on each other.

The core challenge of AI ultimately converges on the issue of income distribution, which is fundamentally a political economy issue. The reason industrial revolutions have not led to permanent mass unemployment is linked to the evolution of fiscal systems and the development of robust social safety nets, which have regulated distribution and narrowed inequalities. For China, the priority is twofold: to strengthen the social security system, particularly for rural residents in areas like pension, healthcare, and education; and to build a governance structure for the gig economy that addresses legal definitions, the protection of workers' rights, career development paths, and the ethics of algorithmic management.