AI, Wages and Demand: Who Will Absorb the Productivity Revolution?

The defining economic risk of artificial intelligence (AI) may not be mass unemployment alone. It may be a widening gap between what AI-enabled economies can produce and the purchasing power available to absorb that output.

According to a new conceptual review titled "Artificial Intelligence, Labour Income and Effective Demand: A Theoretical Framework for the Distributional Absorption Threshold of AI-Induced Productivity," and published in the journal Encyclopedia by Narcis Eduard Mitu of the University of Craiova, AI transition will not be judged only by how much technology can produce. It will also be judged by whether wages, prices, investment, public spending and redistribution allow societies to consume and benefit from that productivity.

The author introduces the idea of a Distributional Absorption Threshold: a point at which AI-related productivity grows persistently faster than broadly distributed real purchasing power and household consumption. The concept does not suggest that AI is inherently harmful, that mass unemployment is inevitable or that any economy has already crossed such a threshold. Instead, it asks a more difficult question: what happens when the gains from AI are concentrated so narrowly that demand cannot keep pace with production?

Productivity Is Not the Same as Prosperity

The current AI debate often treats productivity as the final measure of success. If a worker completes a report in half the time, a customer-service team handles more queries, or a company cuts operating costs, the assumption is that the economy has gained. At the task level, that may be true. The review points to evidence that generative AI can reduce completion time and improve output in professional writing, customer support and knowledge-intensive consulting. Yet such results are uneven; they depend on the task, worker experience, organisational design and whether the activity falls within AI's effective capabilities.

Moving from task-level efficiency to economy-wide growth is far more complicated. Firms need skills, data infrastructure, management capacity and redesigned workflows. Adoption does not necessarily mean effective use, and effective use does not guarantee a proportional rise in national productivity.

Even when productivity does rise, the economy still confronts a second stage: economic realisation. AI may allow businesses to produce more goods and services with fewer inputs. However, additional output must be purchased by households, firms, governments or foreign buyers. Productive capacity and market absorption are connected, but they are not identical.

The paper's key framework thus shifts attention from what AI makes technically possible to what the economy can financially sustain. Productivity gains may become fully realised when they feed into higher wages, lower consumer prices, increased investment, stronger public revenues or rising external demand. If they mainly increase retained earnings, capital income or the wealth of high-saving groups, the effect on immediate consumption may be weaker.

The key issue is not only how much value AI creates, but where that value goes and how it returns to the economy. Productivity can rise inside firms while household purchasing power stagnates. Company balance sheets may improve even as aggregate demand becomes less secure. The possibility is particularly important in economies already experiencing high inequality, weak wage growth or declining labour shares. In such settings, AI could strengthen productive capacity without delivering equally broad improvements in living standards.

The Real Risk Is Income Decoupling, Not Just Job Loss

Public anxiety about AI is dominated by one question: how many jobs will disappear? The concern is legitimate, but it may be too narrow. Occupations are collections of tasks, and AI rarely affects every task in the same way. It may automate some activities, augment others and create new responsibilities in areas such as model oversight, data governance, verification and human-machine coordination. Exposure to AI does not automatically mean displacement.

Employment numbers alone may also conceal a deeper distributional shift. A worker can remain employed while experiencing slower wage growth, fewer paid hours, higher work intensity, weaker bargaining power or a smaller share of the value produced. AI can improve worker productivity without guaranteeing that the resulting gain appears in wages.

The critical divide is not simply between employment and unemployment. It is between productivity growth and income transmission. In a favourable scenario, AI complements workers, raises their output and supports higher real wages, stronger employment income or better job quality. Consumers may also benefit when lower costs translate into lower prices.

In a more critical scenario, firms capture most of the gain. Profits rise, labour's share weakens and income becomes concentrated among owners, platforms and highly skilled employees. Employment may remain relatively stable, but broad purchasing power can still fall behind production.

The framework is especially useful because it connects this distributional outcome to demand. Lower-income and liquidity-constrained households tend to spend more of each additional unit of income, while wealthier households generally save a greater share. Where productivity gains go therefore affects how strongly they return to the economy as consumption.

An additional dollar paid as wages or transferred to a financially constrained household is more likely to support near-term spending than the same amount retained as a capital gain by a high-wealth investor. That does not make profits economically undesirable. Profits can finance investment, innovation and expansion. But the demand effect depends on whether that money is actually reinvested, taxed, redistributed or spent.

The paper's argument is more sophisticated than a simple wages-versus-profits dispute. The issue is whether income circulates with enough breadth and speed to support the output that higher productivity makes possible.

When Efficiency Gains Stop Circulating

The Distributional Absorption Threshold is the point at which AI-induced productivity growth begins to outpace broadly distributed real purchasing power on a persistent basis. In simple terms, the framework asks whether: AI-related productivity growth exceeds the growth of broadly distributed real purchasing power.

This is not a fixed numerical threshold and not a universal trigger for crisis. A temporary gap may be harmless. Strong investment, public expenditure, redistribution, exports or lower prices may absorb additional output even when wages grow slowly. The concern arises when the gap persists and those compensating channels remain too weak.

AI adoption may raise productivity, but the wider economic outcome depends on how those gains move through the economy. They can be transmitted through labour income, lower prices, profits, transfers, investment and external demand. The strength and distribution of these channels ultimately determine whether household purchasing power and consumption grow enough to absorb the additional output.

In a favourable scenario, productivity gains support higher real wages, stronger employment income, lower consumer prices, greater investment or increased public revenues. Purchasing power expands, demand remains resilient and the economy is better able to convert AI-enabled efficiency into realised growth.

In a more critical scenario, wage growth remains weak, profits become concentrated, working hours decline or disposable income fails to keep pace with productivity. Unless investment, redistribution, public expenditure or exports compensate for that shortfall, household demand may weaken and part of the economy's expanded productive capacity may go underused.

This does not imply that factories will suddenly produce mountains of unsold goods. Absorption tension could appear in subtler ways: weaker business sales, slower investment, excess capacity, pressure for more consumer credit, growing dependence on exports or repeated demands for fiscal stimulus.

Credit may temporarily mask the problem. Households can sustain consumption by borrowing even when income growth is weak. But debt-financed demand is not equivalent to broadly shared purchasing power, particularly when higher interest costs eventually force households to retrench.

The review proposes two preliminary indicators for future research. One compares productivity growth with real labour-income growth. The other compares standardised productivity growth with household consumption growth. Yet the author stresses that neither measure proves AI causation. Productivity is shaped by many forces, and consumption may be influenced by credit, wealth, demographics, inflation and fiscal policy.

The proposed threshold is a theoretical lens, not an observed law. The paper does not present new data, identify a country that has crossed the threshold or establish that AI is already creating a demand shortfall. It provides a framework for recognising the risk before policymakers mistake supply-side efficiency for fully realised prosperity.

The Policy Battle Is Over Who Captures the AI Dividend

Governments often approach AI through computing infrastructure, research funding, digital skills and business adoption. Those investments matter, but they address only the production side. An effective AI strategy must also examine how productivity gains reach households and return to demand.

Wage-setting institutions can influence whether workers share in productivity gains. Training and mobility policies can help people move into tasks that complement AI. Social insurance and tax-benefit systems can stabilise disposable income when labour-market adjustment is disruptive.

Competition policy matters because concentrated digital markets may allow dominant companies to retain efficiency gains rather than pass them to consumers through lower prices. Where AI infrastructure, data and models are controlled by a small number of firms, productivity may rise alongside profit concentration.

Public investment provides another route. Governments can transform part of the AI dividend into infrastructure, education, health systems and climate resilience, simultaneously supporting demand and future productive capacity.

Future research will need to combine AI adoption and exposure measures with productivity, real wages, working hours, household disposable income, consumption, inequality, investment and fiscal policy. Sector-level and household-level studies will be especially important because national averages can hide large differences between industries, workers and income groups.