The rapid advancement of artificial intelligence is set to profoundly reshape the global economy, yet its ultimate impact on inflation remains a subject of considerable debate. On one hand, the surge in investment across computing power, chips, and data centers could exert upward pressure on prices. On the other, potential productivity gains and labor substitution may create sustained disinflationary forces. To clarify this issue, we synthesize historical experience with theoretical analysis, categorizing the mechanisms through which technological revolutions affect inflation into four types: supply effects, expectation effects, substitution effects, and bottleneck effects. We then assess the relative strength and timing of these mechanisms. Compared to previous technological revolutions, the AI revolution is characterized by faster diffusion, higher-level substitution, and greater divergence in expectations, which could reshape labor markets and generate stronger disinflationary pressure on the demand side. Therefore, even if inflation rises temporarily, the medium-to-long-term disinflationary pressure may be more potent than in past eras. China, already in a low-inflation environment, may face particularly acute challenges. In response, fiscal policy could focus on income distribution, employment support, and social security, while monetary policy should be precise, forward-looking, and relatively accommodative, avoiding misinterpreting productivity gains as an inflation risk.

Lesson one: past technological revolutions have been accompanied by disinflation

Say's Law, often summarized as "supply creates its own demand," provides a natural starting point for analyzing how technological revolutions affect inflation. The idea is that production not only creates goods but also generates income and purchasing power, which can be directed toward consumption or saved and invested, preventing permanent losses in aggregate demand. Extending this logic to technological revolutions, if new supply correspondingly creates new demand, productivity gains would primarily manifest as higher output and welfare without systematically altering inflation. However, historical evidence from the last two technological revolutions shows that while new supply did create new demand, the expansion of supply and demand was not synchronized, leading to noticeable disinflation. In the late 19th century, the second industrial revolution brought rapid productivity growth and supply expansion, which, combined with the constraints of the gold standard, contributed to a roughly 30% cumulative decline in the US price level. Similarly, during the information revolution in the late 20th century, strong demand coexisted with significant supply-side expansion, leading to a marked decline in core inflation. The US CPI, which hovered around 5-6% in 1990, fell to approximately 1.5% in 1997-1998 and dipped below 2% again in 2001-2002. The experiences of both revolutions indicate that new supply can create new demand, but there is no automatic mechanism guaranteeing that aggregate demand expands in sync with supply capacity. When supply capacity grows faster, technological progress results in disinflationary pressure.

The critique of Say's Law has historically centered on two problems: long-term income distribution and short-term effective demand. Marx highlighted that if income gains from productivity improvements accrue disproportionately to capital and high-income groups, consumption capacity may not keep pace with production capacity, leading to crises of overproduction. Keynes focused on the failure of effective demand in the short run, arguing that income is not automatically converted into consumption, and investment is subject to volatile expectations. Both issues are evident in the past technological revolutions. The second industrial revolution shifted labor from agriculture to industry and services, while the information revolution led to employment polarization. The top 1% income share rose from 10.4% in 1980 to 17.3% in 2000, suggesting that productivity gains were not proportionally reflected in broad-based labor income growth.

Four mechanisms linking technology to inflation

Building on this foundation, we can identify four distinct channels through which technological progress influences inflation. The supply effect operates as productivity gains increase potential output. If nominal demand fails to expand in tandem, a negative output gap emerges, exerting downward pressure on prices. With sticky wages, nominal wages lag productivity growth, lowering unit labor costs and furthering disinflation. This effect is strongest when productivity improves rapidly but wage and price adjustments are slow. The expectation effect can be inflationary if households and firms become overly optimistic about future technology gains. If they incorporate these expected gains into current decisions, they may increase consumption and investment today, creating a positive output gap. However, this effect is mitigated if consumers are pessimistic about their personal income prospects. The substitution effect arises when technology enhances capital's ability to replace labor, potentially leading to a declining labor income share. Since capital owners and high-income earners typically have a lower marginal propensity to consume than workers, this can reduce overall consumption and exert disinflationary pressure. The bottleneck effect occurs when rapid capital spending creates a concentrated demand for key inputs with inelastic supply, driving up their prices. In the presence of price stickiness, these localized price increases can translate into broader inflation. The relative strength and timing of these four effects determine the net inflationary or deflationary impact of a technological revolution.

AI in the short run: investment leads, productivity lags

Applying this framework to the AI revolution, we see that its diffusion is notably faster than previous technological eras. AI chatbots have reached near 50% adoption among US adults—a pace far quicker than personal computers or the internet in their early stages. At the task level, evidence of AI's cost-reducing and efficiency-enhancing power is robust. For instance, customer service agents resolve 14-15% more tickets per hour with AI assistance, with more significant improvements for less experienced workers. However, at the macroeconomic level, evidence that AI is boosting productivity remains limited. Even though measured labor productivity in the US has accelerated recently, a significant portion of this is due to capital deepening from AI investment itself, rather than a jump in total factor productivity (TFP). The "utilization-adjusted TFP" remains relatively low compared to the productivity boom seen during the internet revolution. This suggests the supply effect from AI has yet to be fully realized. This is not to say it won't materialize. Like other general-purpose technologies, AI's productivity benefits often follow a "J-curve," taking time to manifest as firms reorganize around the new technology. As a result, the current phase is characterized by a bottleneck effect, as rapid AI capital expenditure collides with inelastic supply of essential inputs like chips, memory, power, and data centers. This effect, while notable, appears contained—the Minneapolis Fed estimates AI-related supply constraints have raised core PCE by only about 0.4 percentage points so far.

The substitution effect: AI's reach into high-skill jobs

A defining feature of the AI revolution is the higher level at which it substitutes for labor. Previous computer-based technologies primarily impacted routine, codifiable tasks concentrated in middle-skill, middle-income jobs. However, generative AI's ability to handle language, code, and reasoning means high-wage, high-skill cognitive professions are now highly exposed. The correlation between AI exposure and wages has turned strongly positive, unlike the "inverted U" shape of the computer era. This does not yet translate into widespread job losses globally. Macro unemployment remains stable, and the overall employment levels in high AI-exposure occupations haven't diverged significantly from others. This apparent paradox can be explained by organizational adjustment lags and the fact that AI often substitutes for specific tasks within a job, not the entire occupation. In reality, there is a significant impact on hiring flows and entry-level positions. Companies are reducing new recruitment for junior roles in fields like software development, consulting, and law, as AI can handle some foundational tasks. Data indicates that firms adopting AI reduce hiring for entry-level positions by an average of 22%. This particularly hurts young, highly educated workers, potentially creating long-term scarring effects on their careers and incomes. This dynamic, combined with the fact that AI is capable of taking on complex, high-value tasks, exerts pressure on the labor income share. As capital substitutes for labor across a broader swath of the economy, and as market concentration favors "superstar firms," the share of national income accruing to labor is likely to decline. The speed of this substitution effect may outpace the creation of new jobs and tasks, a key difference from previous revolutions.

The AI effect: short-term inflation, long-term disinflation

Synthesizing these mechanisms, the net effect of AI on inflation is likely to be phase-dependent. In the short run, inflation pressures should dominate. We are in a period where investment leads and productivity follows. The bottleneck effect from surging demand for chips, power, and other inputs is the primary driver. Simultaneously, expectations are polarized: investors are optimistic about growth, while households are pessimistic about income, creating a muted demand-side effect that is less inflationary than a broad-based optimism. Over the medium-to-long term, the pressure should shift decisively toward disinflation. As firms fully reorganize and TFP gains materialize, the supply effect will strengthen. Early bottlenecks will ease as key input supply expands. The substitution effect will become more powerful as AI's impact on employment transitions from reducing hiring to affecting wages and labor income share, dampening overall consumption. Compared to past technological revolutions, the disinflationary pressure from AI is likely to be more potent. While the supply-side story is familiar, the demand dynamics are different. The AI revolution targets a higher tier of cognitive work, accelerating the impact on wages and income distribution. This could lead to a situation where income and consumption lag behind the expansion in productive capacity, creating a more persistent disinflationary environment.

China-specific challenges and policy responses

China possesses distinct strengths in realizing the benefits of AI's supply side. The nation's expansive manufacturing base, abundant power supply, high automation levels from its world-leading industrial robot installation, and low-cost domestic AI models position it well to take advantage of productivity gains. Its giant user base and rapid product iteration also facilitate diffusion. However, China's ability to convert these productivity gains into domestic demand is weaker. A large share of its workforce is engaged in cognitive tasks like customer service and data processing that are highly exposed to AI, creating a direct risk to employment and income. China's unemployment insurance system has room for improvement, which could amplify precautionary saving and reduce consumption if job losses occur. Current household income and employment expectations are already weak, creating a situation where even high optimism about AI technology does not translate into confidence about personal income. This mismatch—rapid supply-side gains against a lagging demand-side response—could amplify disinflationary pressures in an economy that is already experiencing low inflation.

Given this outlook, policy should focus on managing the transition. Fiscal policy can play a key role by ensuring the tax and subsidy system does not inadvertently favor capital over labor. It should also bolster targeted transfer payments and active labor market policies, focusing on retraining workers for human-AI complementary roles. Increased public investment in sectors less susceptible to automation, such as care and education, can absorb displaced workers and support incomes. Monetary policy faces the critical challenge of distinguishing between an overheating economy and a rise in potential growth. In China's context, the greater risk is not that productivity gains cause overheating, but that AI-driven disinflationary forces reinforce the existing low-inflation environment, potentially leading to a self-fulfilling deflationary spiral. Therefore, a more supportive monetary policy stance should be considered to stabilize aggregate demand and prevent low-inflation expectations from becoming entrenched. Finally, macroprudential policy is necessary to manage the risks of excessive leverage in the AI investment boom. With high valuations and debt-financed investment in tech sectors, a shortfall in expected returns could trigger a financial accelerator mechanism, amplifying economic downturns. The aim of policy should not be to stifle innovation but to prevent the cycle of "real technology-boosted investment-excessive leverage-asset price decline-credit crunch-further investment decline."