📈 economics

AI and the Productivity-Income Gap: A Theory of Demand Decoupling

This paper presents a theoretical model arguing that artificial intelligence could create a "Productivity-Income Gap" by reducing labor's share of income in a way that outpaces aggregate consumption growth due to differing marginal propensities to consume, while clarifying that current data merely illustrates this pattern without proving causation and emphasizing that historical institutions may eventually—but not necessarily quickly—restore the balance.

Original authors: Ahmad Alamgir Khan

Published 2026-09-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ahmad Alamgir Khan

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

For most of the modern economic era, a simple rule seemed to hold true: when a country got better at making things, its people got richer. If factories became more efficient or workers learned new skills, the extra value created didn't just vanish; it flowed into wages, allowing families to buy more goods and services. This cycle of production and purchasing power kept the engine of the economy running smoothly. However, a new question has emerged as artificial intelligence begins to reshape how work gets done. Could it be possible for a society to become vastly more productive while its people, on average, find themselves with less money to spend? This is the central puzzle facing economists today: if machines and software can produce more output with fewer human hours, who captures the value of that extra production, and does that new wealth end up in the hands of the people who need it to keep the economy moving?

A researcher at Honeywell, Ahmad Alamgir Khan, has built a theoretical model to explore this possibility, calling it the "Productivity-Income Gap." The theory does not argue that artificial intelligence will necessarily cause mass unemployment or that people will lose their jobs entirely. Instead, it suggests a more subtle problem could arise even if everyone remains employed. The core idea rests on how income is shared between workers and the owners of the machines or software they use. In a standard economy, when productivity rises, wages usually rise with it. But if artificial intelligence allows companies to produce significantly more while paying a smaller share of the total income to workers, the balance shifts. The money goes instead to the owners of the technology, who are typically wealthier.

Here is where the mechanism becomes critical. People with lower incomes tend to spend almost every extra dollar they earn on food, housing, and daily needs. Wealthier individuals, who receive the bulk of the income when labor's share shrinks, tend to save a larger portion of their money rather than spending it immediately. If the economy produces a huge amount of goods but the people with the most money choose to save rather than buy them, and the people who would buy them have less money, the total demand for those goods can fall behind the ability to produce them. The factory can make more, but the customers cannot afford to buy it all. This creates a gap where the economy's capacity to produce outpaces the public's ability to purchase.

The paper does not claim that this gap has already fully arrived or that artificial intelligence is the sole cause of current economic trends. Instead, the author uses data from the first quarter of 2026 in the United States as a hypothetical illustration of what this scenario would look like if it were happening. In that specific snapshot, labor productivity—the amount of goods and services produced per hour—rose by 2.8 percent. At the same time, the real hourly pay for workers rose by only 0.6 percent. Furthermore, the share of total business income going to workers dropped to 53.7 percent, the lowest level recorded since 1947. These numbers fit the pattern the theory describes: output is growing much faster than wages, and the share of income going to workers is shrinking.

However, the author is careful to note that these numbers alone do not prove that artificial intelligence caused the shift. Other factors could explain the same pattern, such as changes in energy prices, shifts in which industries are growing, or the effects of interest rate policies. The data serves as a map of what the theoretical condition looks like in the real world, not as a verdict on what caused it. The model suggests that for this gap to become a lasting problem, two things must happen simultaneously: artificial intelligence must reduce the share of income going to workers, and the people who receive that income (the wealthy owners) must spend less of it than the workers would have.

The research also looks at how the economy might fix itself. History shows that when production and purchasing power have pulled apart before, institutions like unions, tax policies, and social safety nets have eventually stepped in to narrow the divide. This process, however, often takes decades and significant political effort. The paper suggests that new job creation could also close the gap, but only if those new jobs pay enough to restore the spending power of the workforce. Until such adjustments happen, the economy risks a situation where it is capable of producing more than it can sell, leading to a slowdown in growth not because of a lack of technology, but because of a lack of buyers.

Ultimately, the paper offers a conditional warning rather than a prediction. It does not say the future is doomed, but it specifies the exact conditions under which a problem could emerge. If artificial intelligence continues to boost production while concentrating income among those who save rather than spend, and if no new jobs or policies arise to redistribute that wealth, the economy could face a persistent shortfall in demand. The fundamental question raised is not whether we can produce more, but whether the gains from that production will be shared widely enough to sustain the demand needed to keep the system running. The answer depends less on the technology itself and more on how society chooses to distribute the wealth it creates.

Technical Summary: AI and the Productivity-Income Gap: A Theory of Demand Decoupling

Problem Statement
The paper addresses a specific macroeconomic risk associated with the adoption of Artificial Intelligence (AI): the potential for a "Productivity-Income Gap." This is defined as a scenario where AI-driven productivity growth ( $g_A$ ) outpaces the growth of aggregate demand ( $g_{AD}$ ), even in an economy operating at full employment. Unlike traditional narratives focusing on technological unemployment (a lack of jobs), this framework posits a "distributional transmission failure." The core problem is whether the gains from increased production are distributed in a way that sustains the purchasing power necessary to absorb that production. Specifically, the paper investigates the conditions under which AI reduces labor's share of income ( $\lambda$ ) while the marginal propensity to consume (MPC) out of labor income ( $c_L$ ) exceeds the MPC out of capital income ( $c_K$ ).

Methodology
The paper employs a theoretical macroeconomic model grounded in functional income distribution and standard consumption functions, rather than causal empirical testing of AI's effects.

Theoretical Framework: The model extends the standard aggregate demand equation ( $AD = C + I + G + NX$ ) by explicitly decomposing consumption ( $C$ ) into labor income ( $Y_L$ ) and capital income ( $Y_K$ ).
- $C = c_L Y_L + c_K Y_K$
- Labor share is defined as $\lambda = Y_L / Y$ .
- The model derives the growth rate of consumption ( $g_C$ ) as a function of output growth ( $g_Y$ ) and the growth rate of the labor share ( $g_\lambda$ ).
- The critical derivation (Eq. 5a) shows that if $g_\lambda < 0$ (labor share falls) and $c_L > c_K$ , then $g_C < g_Y$ .
Formal Condition: The "Productivity-Income Gap" is formally defined as the state where $g_A > g_Y \geq g_{AD}$ . This condition binds only when two simultaneous criteria are met:
- AI reduces the labor share of income ( $\lambda$ ).
- The marginal propensity to consume out of labor income is strictly greater than that of capital income ( $c_L > c_K$ ).
Empirical Approach: The paper utilizes U.S. data from the first quarter of 2026 (Q1 2026) strictly as an illustrative case study to demonstrate what the theoretical condition looks like in real data. The author explicitly disclaims that this single quarter of data establishes causality. The analysis acknowledges multiple confounding factors (monetary tightening, energy price shocks, post-pandemic sectoral shifts, and measurement lags) that could explain the observed divergence between productivity and compensation.

Key Results and Propositions
The paper formalizes five propositions derived from the model:

AI and Labor Dependence: AI increases productivity while reducing marginal dependence on labor, ceteris paribus.
The Gap Condition: A Productivity-Income Gap exists when $g_A > g_Y \geq g_{AD}$ . This emerges specifically when $g_\lambda < 0$ and $c_L > c_K$ .
Mediation: Income distribution mediates the link between AI-driven productivity and aggregate demand.
Concentration Risk: The probability of the gap opening increases with the concentration of AI-generated gains, as greater concentration implies a sharper decline in $\lambda$ for a given productivity increase.
Institutional Mitigation: Institutions that raise $\lambda$ (e.g., collective bargaining, profit-sharing) or broaden capital ownership (e.g., pensions, employee stock ownership) reduce the probability of the gap persisting.

Illustrative Evidence (Q1 2026)
The paper presents Q1 2026 U.S. data as a "stress test" of the theoretical condition:

Labor Share: Fell to 53.7%, the lowest since 1947.
Productivity vs. Compensation: Labor productivity grew 2.8% year-over-year, while real hourly compensation grew only 0.6%.
Corporate Profits: After-tax corporate profits reached 12.4% of GDP, and profit margins hit record highs.
Consumption Concentration: Private data suggests that nearly all growth in real retail spending since 2023 has come from households earning over $125,000, consistent with the assumption that high-income agents have a lower MPC.

The author emphasizes that while these numbers are consistent with the theoretical condition, they do not prove AI caused them. Alternative explanations, such as monetary policy tightening and the unwinding of energy shocks, remain plausible.

Significance and Claims
The paper's primary contribution is the formalization of a "demand-side constraint" on growth driven by distributional shifts rather than unemployment.

Distinction from Unemployment: The framework clarifies that a demand shortfall can occur even at full employment if income concentrates among agents (capital owners) who save rather than spend.
Distinction from Secular Stagnation: While related to secular stagnation literature, the proposed mechanism is distributional (who captures the gains) rather than demographic or tied to a falling natural rate of interest.
Policy Relevance: The model suggests that if the gap opens, market forces alone may not close it quickly. Historical precedents (e.g., the New Deal) suggest institutional adjustments (bargaining, taxation, welfare) are required to narrow the gap, but these processes are slow and politically contested.
Modesty of Claims: The paper explicitly avoids predicting that AI will cause a gap. Instead, it provides a precise framework to identify when the pressure for such a gap arises. It notes that new job creation, wage bargaining, and market adaptation could close the gap before it becomes economically significant. The model assumes a "wage-led" regime for the U.S. (where demand is driven by wages), a classification that remains an assumption subject to empirical debate.

Limitations and Omissions
The author acknowledges several channels omitted from the static model that could offset the gap:

Monetary Policy: Central bank rate cuts could stimulate demand, though inflation constraints limit this tool.
Investment Offset: Higher profits might spur investment ( $I$ ), but current AI-related capital expenditure (approx. 0.8% of GDP) appears insufficient to fully offset consumption shortfalls compared to historical tech cycles.
Wealth Effects and Debt: The model excludes wealth effects from rising asset prices and debt-financed consumption, which could temporarily sustain spending despite falling labor shares.
Price Adjustments: The model assumes price pass-through is incomplete; if AI significantly lowers prices, real incomes could rise even if nominal wages lag.

In conclusion, the paper does not claim AI has already decoupled productivity from demand but offers a rigorous theoretical lens to monitor the specific distributional conditions ( $g_\lambda < 0$ and $c_L > c_K$ ) under which such a decoupling becomes a macroeconomic risk.

Technical Summary: AI and the Productivity-Income Gap: A Theory of Demand Decoupling

More like this