The Abundance Paradox

The promise of the Intelligence Economy rests on a compelling proposition: we will be able to produce more with less human effort. Capabilities once scarce and expensive could become widely available. A company might obtain software, accounting, analysis and coordination from systems that perform work previously distributed across specialised firms. The technical achievement would be considerable. Its meaning for prosperity is less straightforward.

If an expanding range of these activities is absorbed by relatively few providers, productive capacity may grow while wages, business income and independent ownership become less widely distributed. The economy would become better at supplying services while weakening some of the means through which people obtain them. How will the capacity to produce be connected to the capacity to buy?

This is what I mean by the abundance paradox. A technology capable of reducing the effort required to create wealth could weaken the economic arrangements through which people share in it. There is no logical contradiction here. Production and access depend on different conditions. The paradox arises when we expect progress in the first to guarantee progress in the second.

The tension does not make collapse inevitable. Income can reach people through new work, lower prices, ownership or public provision. Where these adjustments are insufficient, however, productive capacity may advance faster than the distribution of viable claims on its output. That could undermine profits and investment. It could also sustain a narrower prosperity in which commercial success coexists with widespread economic insecurity.

The question concerns the purposes of economic progress as much as its mechanisms. If greater capability leaves many people with less control over their lives, we need to examine the institutions through which capability becomes prosperity. That examination should begin while those institutions can still be shaped.

Production and the distribution of prosperity

Modern capitalism relies on more than private ownership and competition. It also relies on a circulation of income. Businesses pay workers and suppliers, distribute returns to owners and contribute taxes. Households spend, save and borrow. Governments purchase services and redistribute resources. Financial institutions connect present spending with expectations of future income. A market registers people's needs through the spending they can command. Increasing the ability to satisfy a need does not automatically give the person who has it the means to pay.

For most people, employment is their principal claim on this system. A wage gives them the means to buy goods and services produced by others. It also supports access to credit, retirement provision and, in some countries, healthcare. Employment has therefore carried a large part of the responsibility for turning economic production into broadly distributed purchasing power.

This arrangement has always been incomplete. Unpaid care sustains economic life without receiving a wage. Informal workers often lack protection. Inherited wealth gives some people substantial claims on output without requiring them to work. These exceptions show that the relationship between contribution and income already reflects institutional choices. Paid employment nevertheless remains central to the legitimacy and operation of the prevailing economic model.

Automation does not make income disappear by definition. If a firm reduces its wage bill, the saving may become lower prices, higher profits, new investment or higher pay for its remaining workers. Owners can spend their returns. Pension funds can distribute them. Governments can tax them. A serious argument about AI must account for all these channels.

The issue is whether they preserve sufficient demand and sufficiently broad participation. A household living largely from its salary will usually spend a substantial portion of an additional euro. A wealthy owner can save much more of it. Those savings can finance useful investment, but investment depends on expected returns and the availability of productive opportunities. There is no guarantee that every reduction in household spending will be matched, at the same time and place, by additional investment.

Nor does a falling share of income going to labour necessarily mean that workers become poorer. Their share can decline while their real incomes rise if the economy grows fast enough. We need to examine actual purchasing power, its distribution and the security of the income supporting it. A ratio alone cannot settle the argument.

A wage performs two functions that we often treat as inseparable. It pays for a contribution to production and gives its recipient a claim on what others produce. If technology reduces the need for that contribution, the second function still has to be fulfilled somehow. The difficult question is how far access to prosperity should depend on the continuing market demand for a person's labour.

Concentration and the ownership of opportunity

By the Intelligence Economy, I mean an economy in which machine intelligence becomes a widely available input into reasoning, decision making and the execution of work. Its significance comes partly from the range of activities it can enter. A machine designed for one industrial task has a relatively clear economic boundary. A system that can interpret instructions, operate software and coordinate other systems can move across the boundaries of many service industries.

A specialised firm may initially adopt AI to improve its work. Later, the provider of that AI may offer the service directly. An independent software product becomes a feature within a larger platform. An agency loses its intermediary role. A professional practice retains responsibility for an outcome while the platform increasingly controls the tools, distribution and customer access on which it depends.

This is a plausible commercial trajectory, not a claim that every profession can be absorbed. Responsibility, trust, local knowledge and legal accountability give many specialised providers durable value. Integration can also improve quality and remove expensive coordination. Preserving every existing intermediary would be an unreasonable objective.

The concern arises when integration creates control over the conditions under which everyone else can compete. A provider that supplies the model, hosts the business and mediates the customer's choices can influence prices and access across several markets. If a user's AI assistant also selects suppliers, ownership of that assistant can become a powerful commercial advantage. Providers may then have to pay for access to demand that they once reached independently.

The US Federal Trade Commission's 2025 study of major cloud and AI partnerships identified potential competition concerns involving access to computing resources, switching costs and access to sensitive information. It did not establish that concentration inevitably produces monopoly, but it showed why the relationships between these layers deserve scrutiny. FTC staff report ↗

There are countervailing forces. Models can become cheaper. Open models can support new entrants. Customers can combine suppliers. Stanford's 2025 AI Index documented a steep decline in the cost of obtaining a fixed level of model performance, illustrating how quickly one apparent barrier can weaken. Yet cheaper intelligence does not guarantee open distribution. Control can migrate towards customer relationships, proprietary workflows or infrastructure that remains costly to replace. Stanford AI Index ↗

We should also distinguish concentration of suppliers from concentration of ownership. Several competing providers may still distribute their returns to a narrow group of asset owners. Conversely, a concentrated infrastructure can distribute some of its returns widely through pensions or public investment funds. Competition and ownership address different parts of the problem.

There is a further cost when independent firms disappear. They have served as places where people acquire expertise, build professional reputations and eventually become owners themselves. If AI removes the early work through which experience is accumulated, the economy may weaken its future supply of judgement. Dependence on common providers also creates shared points of failure. An independent business is also a means of making decisions without another provider's permission. A cheaper service can improve efficiency while the loss of such independence narrows economic agency.

What the evidence allows us to say

The central argument would fail if we treated every task exposed to AI as a job destined to disappear. Jobs combine activities, and automating one can increase the value of the others. In its 2025 assessment, the International Labour Organization estimated that one in four workers globally was in an occupation with some exposure to generative AI. Only 3.3 per cent of employment fell into its highest exposure category. Its assessment favoured transformation of jobs over wholesale replacement. ILO exposure study ↗

There is also no fixed quantity of useful work. Lower costs can expand markets. Better tools can make small businesses viable. New products create new demands, while ageing populations and unmet needs in care, education and infrastructure leave substantial room for additional activity. Economic research distinguishes the displacement of existing tasks from the creation of new tasks in which people regain a productive role. Automation and New Tasks ↗

The unresolved question concerns the balance between these effects. New work must appear in sufficient volume, pay enough and be accessible to the people whose previous work has weakened. A displaced administrator does not become an experienced nurse or an infrastructure engineer because demand has increased in those occupations. Adjustment takes time, investment and often a change of location. An economy can create jobs overall while particular generations or regions suffer lasting losses.

The evidence so far supports caution in both directions. A study of more than 5,000 customer support agents found that access to an AI assistant raised issues resolved per hour by 14 per cent on average, with larger gains among less experienced workers. A Danish study, revised in March 2026, found substantial changes in tasks but no significant average effect on earnings or recorded hours over its early observation period. Generative AI at Work ↗ Danish labour-market study ↗

Meanwhile, Stanford researchers using US payroll data reported in August 2026 that employment among young workers in exposed occupations had fallen behind that of less exposed peers, principally through reduced hiring. They explicitly described the findings as early descriptive indicators rather than causal estimates, and found no widespread economy-wide displacement. Stanford employment study ↗

These results concern different settings and cannot be assembled into a single forecast. They show that productive assistance, limited aggregate effects and concentrated employment pressures can coexist. They also suggest that the transition may first appear as a missing opportunity: the junior position that is never opened, rather than the experienced employee who is dismissed.

Prices reveal another aspect of the paradox. A service can become more useful to society while generating less revenue for its providers. If AI makes something previously expensive freely available, the loss of a market can accompany an increase in human welfare. We must therefore distinguish the usefulness of what is produced from the income that can be earned by selling it. A decline in commercial value is not, by itself, a failure of economic progress.

However, a household buys a mixture of things. Housing, energy, food, healthcare and physical infrastructure do not become abundant merely because a digital task becomes cheap. AI may improve their production, but land, materials, investment and institutional constraints still matter. A person whose income falls sharply can be worse off even while enjoying almost free access to excellent digital services.

Abundance is always abundance of something. Affordable intelligence does not remove every material constraint or give everyone an equal claim on scarce resources. Its effect on prosperity depends on how the gains reach people and on the cost of the whole life they need to live.

The limits of individual economic rationality

Consider a transition in which displacement spreads across services, new employment absorbs people slowly, and ownership of the productive systems remains narrow. Households facing weaker income reduce expenditure. Businesses selling to them encounter weaker demand and respond by cutting costs, often through further automation. The original efficiency gain can then contribute to a second round of pressure on employment.

This feedback is possible, but several forces can interrupt it. Lower prices can support consumption. New investment can employ people and purchase equipment. Public spending can stabilise incomes. Rising returns to broadly held assets can reach households. Whether demand weakens depends on the relative strength and timing of these adjustments.

Debt can also sustain spending for a while. It becomes a fragile substitute when borrowers have no credible prospect of recovering their income. Research on indebted demand describes how a concentration of income and the accumulation of household debt can reinforce demand weakness under particular economic conditions. AI could intensify such a mechanism; that is a risk scenario, not an established finding about its current aggregate effects. Indebted Demand ↗

Selling abroad offers an outlet for individual countries and companies. It cannot provide a universal answer because the world as a whole has no external customer. Business purchases and government contracts can sustain demand too. Investment produces capital goods with real value, and public spending need not wait for an identical amount of tax revenue to arrive. But continued investment requires some expectation of useful future output, while public spending remains constrained by productive capacity, inflation and financing conditions.

The paradox becomes sharper when considered from the perspective of the firm. Reducing costs and capturing a larger share of expenditure are rational commercial objectives. An individual business can benefit from demand sustained by other employers while bearing the full cost of maintaining its own wage bill. Its incentives therefore need not support the level of household income on which firms collectively depend. Markets do not automatically reconcile these two levels of rationality. The institutions connecting productivity to household income help determine whether they remain compatible.

Still, the leading AI companies might remain profitable. They could capture a larger share of a slower-growing economy, serve affluent customers and governments, or charge for infrastructure on which other firms depend. A decline in mass purchasing power would not hit every provider equally. Some businesses could thrive while others fail.

An economy can remain commercially viable while becoming harder to justify to those who live within it. This possibility is more troubling than the expectation of an automatic correction. If the beneficiaries of concentration can prosper without restoring broad participation, society cannot rely on their eventual losses to resolve the problem. Profitability and social legitimacy are related, but neither guarantees the other.

There is also a different route to a collapse in expected value capture. Competition could make AI services so cheap that their providers cannot earn the returns anticipated when infrastructure was financed. Useful technology would survive while valuations and some business models failed. Such a correction would concern the monetisation of abundance, even if household demand remained healthy. It should not be confused with the demand crisis described above.

Four possible economic settlements

One future is a broad expansion of prosperity. AI raises productivity, prices fall, new firms enter and workers move into activities that become more valuable. Real household incomes rise, and gains reach people through both wages and ownership. This is economically plausible. It requires competition that remains open, institutions that support transitions and sufficient demand for new human work. The speed of technical progress alone tells us little about whether those conditions will hold.

A second future is durable concentration. A relatively small group controls important productive assets and access to markets. Many households depend on insecure work, modest transfers or asset income that is unequally distributed. Firms remain profitable through essential services, affluent demand and public contracts. The economy continues to function, but economic independence becomes harder to obtain. Political influence may then reinforce the distribution of ownership that produced the problem.

A third future involves a disruptive correction. Labour income weakens faster than replacement income develops, consumer markets disappoint and investment expectations are revised downwards. High debt or concentrated financial exposures amplify the adjustment. Some AI providers could be caught directly; others would experience it through their customers. This outcome requires several adverse conditions to coincide. Concentration by itself is insufficient to establish it.

A fourth future emerges through deliberate institutional change. Productivity gains support a combination of higher living standards, shorter working time, stronger public services and more widely distributed investment income. Employment remains valuable, but access to a decent life depends less exclusively on obtaining a full-time wage. This would demand choices about ownership, taxation and the allocation of resources that today's debate often postpones.

These futures can overlap. A country may enjoy productive expansion in one sector and entrenched dependency in another. A financial correction may precede wider diffusion. They describe possible economic settlements rather than forecasts to which we can assign precise probabilities. Their differences depend partly on technology, but also on how societies organise ownership, distribute income and negotiate the meaning of participation.

The global context makes the transition more demanding. The IMF's October 2026 outlook describes uneven growth, fiscal vulnerabilities and geopolitical risks alongside potential gains from AI investment and productivity. Governments are therefore approaching this transformation with competing demands on public resources. Countries benefiting from AI investment may have more room to adjust than those facing expensive imports and weaker fiscal positions. IMF World Economic Outlook ↗

The current investment boom also complicates the timing. Spending on data centres, equipment and energy infrastructure creates demand before the full effects of automation are known. The IMF has described this as a positive demand shock. That spending can support a successful transition. It can also postpone the moment when weaker demand elsewhere becomes visible. A buoyant construction phase does not establish that the eventual distribution of income will sustain the capacity being built. IMF Blog ↗

Policy will also depend on political power. Concentrated firms can finance research and infrastructure that fragmented markets might struggle to provide. They can also become difficult for governments to discipline when public services and national ambitions depend on them. A serious strategy must retain the productive advantages of scale while preserving the practical ability to challenge its exercise.

Developing economies and the terms of participation

For developing economies, the paradox concerns the terms on which they can enter the Intelligence Economy. Access to powerful tools can expand what their citizens are able to do, while control of the businesses and infrastructure through which income is earned remains elsewhere. The problem extends beyond the replacement of existing jobs to opportunities that may narrow before countries have fully used them.

Service exports can allow a country to employ educated workers and earn foreign currency without first building a large industrial base. If routine software production, customer support or administrative services require fewer people, some of that opportunity may weaken. Domestic employment statistics could initially show limited damage because the jobs that would otherwise have been created never existed.

This remains a conditional argument. Lower costs can expand demand for exported services, and local firms can use AI to compete for work previously beyond their capabilities. Relationships, language, accountability and knowledge of particular markets can preserve advantages. Countries should test how specific export activities are changing rather than conclude that service-led development has ended.

Lower direct exposure also offers incomplete reassurance. The ILO's 2025 index found less exposure in low-income economies than in high-income ones, partly reflecting different occupational structures. A country can nevertheless be affected through weaker demand from trading partners, changing investment flows or reduced remittances. Vulnerability includes its capacity to respond to a shock, not simply the number of tasks a model can perform. ILO exposure study ↗

There is an opposite possibility worth taking seriously. Where professional services are scarce, AI may extend access to useful knowledge at a cost that previously made provision impossible. A small enterprise can obtain better administrative support. A teacher can prepare material in a local language. A public agency can process an application more quickly. Such improvements can raise productivity and strengthen demand within the domestic economy, provided the systems are reliable and people can actually use them.

The World Bank's 2025 assessment identifies connectivity, computing capacity, locally relevant data and skills as foundations for participation. It also documents how unevenly AI capabilities and adoption are distributed. These foundations suggest a practical sequence for preparation. World Bank report ↗

Governments should begin with the constraints that prevent ordinary firms and public institutions from using the technology productively. Reliable electricity, affordable connectivity and capable staff may deliver more value than an expensive attempt to train a national frontier model. Domestic computing capacity can be justified where demand, security needs and operating costs support it. In other settings, diversified international access or regional cooperation may be more sensible. Sovereignty includes the ability to change supplier and maintain essential services.

Preparation must also reach beyond access. A country needs businesses that can retain customers, assume responsibility for outcomes and earn income from locally valuable expertise. Supporting these firms can involve shared facilities, practical training, access to finance and public procurement that allows smaller suppliers to compete.

Education should preserve routes into expertise. If routine junior work contracts, employers and universities will need other ways for people to practise under supervision and learn to recognise mistakes. Training everyone to operate the same interface would create a fragile advantage. Domain knowledge and the ability to verify work are more durable, though they too must be connected to actual employment opportunities.

Social protection needs to fit economies where many people work informally. Benefits tied exclusively to formal payrolls will miss much of the population. Affordable basic services and portable income support can help households withstand transitions. Payment and identification systems can improve delivery, provided they include workable routes for people without reliable digital access and protect against exclusion or misuse of personal data.

The financing cannot be assumed. A government whose most productive firms depend on imported AI services may have little access to the profits generated abroad. Cloud payments can require foreign currency, while domestic revenues arrive in local currency. Tax administration, regional cooperation and the terms of international investment therefore belong within AI strategy. Commitments to permanent transfers should rest on credible fiscal resources rather than projected windfalls from a technology boom.

Resource-rich countries face a related choice. Supplying energy, minerals or land for data centres can generate income, but ownership of those resources does not guarantee a broad development benefit. Projects should be judged by public revenue, infrastructure contributions, local capability and their demands on scarce electricity and water. A large investment announcement can coexist with limited lasting employment.

The relevant test of development is whether adoption leaves local people and businesses with a lasting capacity to generate income and exercise meaningful choice. Imported technology can contribute greatly to that capacity; dependence becomes a concern when countries lack workable alternatives or the ability to retain a fair share of the benefits. Reproducing foreign concentration within a protected national monopoly would offer little reassurance. The aim is wider participation on terms that countries can help shape.

The principles of a different economic bargain

The abundance paradox exposes an institutional question that productivity figures cannot answer. On what basis should people share in output when an increasing part of its production may require little of their labour? Existing arrangements combine wages, property rights and public provision. If the balance between production and employment changes substantially, the balance between these claims may have to change with it. No single institutional answer follows from the technology.

Competition policy remains necessary. People and businesses should be able to move their data, change providers and reach customers without unreasonable dependence on a gatekeeper. These obligations must be technically workable and respect privacy and security. Open models can help, but openness offers limited protection where the practical means of using them remain inaccessible. The objective should be the ability to enter and compete, assessed across the whole service.

Wider ownership deserves equal attention. Employee participation, diversified pension provision, citizen investment funds and public equity obtained on fair terms when public resources finance private development can distribute returns beyond a small circle. Each requires scrutiny. Concentrating an employee's savings in the company that also pays their wage increases personal risk. Public funds can be captured politically. A poorly designed ownership scheme can distribute losses more effectively than prosperity. Diversification, independent governance and transparent rights are essential.

There is also room to negotiate how productivity gains appear in working life. In some activities, they could support shorter hours without equivalent reductions in pay. In others, they may justify higher pay or an expansion of service. None of these follows automatically, and a universal reduction in hours without regard to productivity would impose real costs. Worker representation can help determine which arrangements are feasible and how gains are shared.

Tax systems will need to remain effective if income shifts away from payrolls. This calls for attention to profits, capital income and economic rents, meaning returns sustained by scarcity or market power beyond what is needed to attract investment. It does not require treating every use of AI as a taxable harm. The IMF's work on fiscal policy and generative AI makes the case for preparing social protection and revenue systems for a range of possible disruptions. IMF fiscal-policy study ↗

A universal basic income is one possible component of a future settlement. Its appeal is a dependable income floor that does not disappear when a person changes employment. Its adequacy, cost and interaction with existing services require serious examination. Cash alone cannot create housing or healthcare where supply is constrained, and part of the benefit may be absorbed by higher rents. Redistribution must therefore be considered alongside the production and ownership of necessities.

Purchasing power is only one dimension of participation. A person can receive an adequate income and still have little influence over the conditions of their economic life. Transfers would address only part of the problem if control of production remained beyond the reach of most citizens. Opportunities to create enterprises, acquire assets and influence common rules also matter. The question is whether people can act within the economy with some independence, as well as consume what it provides.

Governments and businesses should begin measuring the transition accordingly. Productivity and investment remain important, but so do median disposable income, the cost of essentials, entry into professional work and the distribution of capital income. Persistent productivity gains accompanied by weaker household incomes would justify stronger action on distribution. Healthy business formation and rising real incomes would support a different assessment. Policy should respond to these patterns without waiting for a definitive count of jobs lost to AI.

Such changes could remain recognisably capitalist. Private enterprise, markets and profit can coexist with wider ownership, shorter working lives and a larger role for collectively financed services. What would change is the weight placed on employment as the principal basis of economic membership. Markets operate within arrangements of property and obligation that societies can revise. We cannot yet know how far revision will need to go, but the present arrangement should not be mistaken for an unchangeable economic necessity.

Work has purposes beyond income. It can provide a sense of competence, recognition and connection with others. Those goods deserve attention even if less paid labour is required. Yet the value of a person cannot reasonably be made to depend on whether their abilities remain more profitable to employ than a machine. Reducing necessary labour should allow more room for worthwhile activity, provided people retain the resources and freedom to choose how to use that room.

The companies building the Intelligence Economy have a stake in this too. Their long-term prospects depend on the societies in which their services are used: on customers with income, institutions with legitimacy and markets in which new businesses can grow. Even the most capable provider cannot secure those conditions through technical excellence alone.

We should address these questions while several futures remain open. Infrastructure contracts, ownership arrangements and procurement decisions made today will shape the options available years from now. The transition may take a long time, but its terms are already being formed. Waiting for the economic effects to become unmistakable would leave less room to decide how the gains should be shared.

The promise of abundance is a greater freedom from necessity. The paradox appears when the institutions governing access to it leave people more dependent and less secure. Resolving that tension will require an account of prosperity that gives as much attention to people's ability to lead their lives as to the economy's ability to produce. The Intelligence Economy's success should ultimately be judged by how far those two capacities advance together.

Sources

1. US Federal Trade Commission. FTC Issues Staff Report on AI Partnerships and Investments Study. January 2025. The report identifies potential competition concerns; it does not establish the inevitability of monopoly. FTC staff report ↗

2. Stanford Institute for Human-Centered Artificial Intelligence. The 2025 AI Index Report, Research and Development. The inference-cost comparison holds a benchmark capability level approximately constant and should not be read as the cost of every AI workload. Stanford AI Index ↗

3. Gmyrek, P., et al. Generative AI and Jobs A Refined Global Index of Occupational Exposure. ILO Working Paper 140, 20 May 2025. Exposure describes potential effects on tasks, not a forecast of redundancies. ILO exposure study ↗

4. Acemoglu, D., and Restrepo, P. Automation and New Tasks How Technology Displaces and Reinstates Labor. Journal of Economic Perspectives 33(2), 2019, pp. 3–30. Automation and New Tasks ↗

5. Brynjolfsson, E., Li, D., and Raymond, L. R. Generative AI at Work. NBER Working Paper 31161, April 2023, revised November 2023. The 14 per cent estimate refers to this working-paper version and its customer-support setting. Generative AI at Work ↗

6. Humlum, A., and Vestergaard, E. Still Waters, Rapid Currents Early Labor Market Transformation under Generative AI. NBER Working Paper 33777, May 2025, revised March 2026. Previously circulated as Large Language Models, Small Labor Market Effects. Danish labour-market study ↗

7. Brynjolfsson, E., Chandar, B., and Chen, R. Canaries in the Coal Mine Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, revised 12 August 2026. The authors describe the findings as descriptive, not causal, and discuss limitations in sample representativeness and pre-existing trends. Stanford employment study ↗

8. Mian, A., Straub, L., and Sufi, A. Indebted Demand. Quarterly Journal of Economics 136(4), 2021, pp. 2243–2307. A theoretical mechanism linking inequality, borrowing and demand; it is not an empirical estimate of AI's effects. Indebted Demand ↗

9. International Monetary Fund. World Economic Outlook Resilience Under Strain, Urgent Choices. October 2026. IMF World Economic Outlook ↗

10. Georgieva, K. The World Can't Afford to Get AI Wrong. IMF Blog, 15 October 2026. IMF Blog ↗

11. World Bank. Digital Progress and Trends Report 2025 Strengthening AI Foundations. World Bank report ↗

12. Brollo, F., et al. Broadening the Gains from Generative AI The Role of Fiscal Policies. IMF Staff Discussion Note 2024/002, June 2024. IMF fiscal-policy study ↗