Imagine a household whose costs for reports, design and paperwork keep falling. Nearby delivery services are starting to use robots too. The family does benefit from technological progress. Yet at the end of the month, rent, energy and care bills are still a source of anxiety.
There is no contradiction here. A society can have ever-cheaper intelligence and still face an expensive life.
AI-driven abundance has to pass through production capacity, competition and income distribution before it becomes affordability that ordinary households can feel.If we skip those steps, it is easy to turn “some things cost less to produce” into “life will become easier for everyone.”
The price of intelligence and the cost of living are different bills
Digital services can be copied and delivered remotely. Physical services must happen at a particular place and time. Robots can make handling goods more efficient, but they still have to reach the site. More efficient construction does not automatically create more land in a popular neighborhood. Where density, redevelopment and infrastructure allow, however, the same land can accommodate more homes.
Likewise, reducing an institution’s administrative workload does not immediately expand every professional service it provides. Infrastructure, qualifications, space, transportation and organizational processes may still constrain delivery.
This does not mean those limits can never change. It is a reminder that households usually buy a complete outcome, while technology may initially reduce the cost of just one part of it.
To assess AI’s benefits, we need to look at what households actually buy. Cheaper content production may offer limited relief to someone whose budget is dominated by housing. For someone constrained by caregiving, reliable in-person help may be worth more than many free software tools. No abstract “intelligence price index” can stand in for everyone’s financial pressure.
Between lower costs and lower prices lies a market
Suppose automation lowers the cost of a service. The business might cut prices to attract customers, improve quality, increase its profit or first repay its equipment investment. Several of these outcomes can occur together.
How much reaches consumers depends on competition, barriers to entry, alternatives and how quickly new supply can arrive. A falling cost curve alone does not establish that retail prices will fall by the same proportion.
If a few parties control essential sites, channels or infrastructure, efficiency gains may accumulate at those bottlenecks. Even as robots get cheaper, access to operating locations and connections may become a new source of cost.
There is an encouraging possibility too: automation could lower entry barriers and strengthen competition, passing more gains to consumers. I do not think the outcome is predetermined. What matters is actual pricing, quality and additional supply—not just a manufacturer’s demonstrated cost per action.
For a business, “how much have we reduced our internal costs?” is an operating question. “What has the customer gained?” is the promise the market cares about.
Lower prices should also be examined alongside quality. If a cheaper service leaves customers handling more exceptions themselves, the apparent savings may simply shift costs to the other side.
Physical abundance takes time to build
AI’s development creates new resource demands too. The IEA’s Energy and AI report examines electricity demand from growing data centers and the conditions needed to supply it. Its future figures are scenario projections. Data-center electricity use cannot all be attributed to robots, but the analysis reminds us that expanding intelligent systems still requires physical infrastructure.IEA energy-demand analysis
Electricity, connections, equipment and site modifications must be coordinated. Software capabilities may improve faster than infrastructure expands. In the short term, sudden demand growth may put even greater pressure on some parts of the supply chain.
This explains why “AI will improve supply” and “some prices may rise” can both be true. Technological progress affects a changing process, not an endpoint reached in an instant.
Public policy and business investment need to identify bottlenecks early, rather than discovering after robot numbers rise that space, electricity or maintenance capacity cannot keep up. The capacity that matters is capacity that reliably delivers to customers.
High incomes for all still leave the question of what money can buy
A future in which more people receive income through wages, public transfers or participation in capital is worth discussing. But rising nominal income does not automatically mean purchasing power rises at the same rate.
Consider a simple scenario: more households suddenly have money for care services, while reliable local supply has not yet increased. The additional demand could produce queues, price increases or expanded supply. Which path follows depends on entry conditions and the pace of construction.
Analyzing cash support and prices therefore requires attention to funding sources, unused economic capacity, where demand goes and how responsive supply is in each sector. Moving directly from “total output will rise” to “there will be no inflation” omits crucial conditions. Claiming that every transfer inevitably produces runaway prices is equally careless.
The IMF’s discussion of fiscal policy in the AI era considers social protection alongside revenue collection and tax reform. This whole-budget perspective explains more than announcing a payment everyone will receive.IMF policy analysis
The promise I care about is how income support and access to essential services improve together. Giving people money to pay and ensuring they can actually obtain the service need to be designed together.
The strongest counterargument: AI can change the bottlenecks too
Of course, today’s limits are not tomorrow’s destiny. AI and robots may improve design, maintenance, construction and resource allocation, expanding supply. That is one reason Physical AI deserves investment.
But deployment evidence must bridge the gap between “could ease” and “has eliminated.” We should track delivery times, service coverage, maintenance costs and quality—and whether people across income groups share in the improvement.
Businesses can evaluate products through the customer’s total cost: purchase, connection, supervision, repairs and downtime. Local authorities can identify what prevents essential services from expanding. Household research should examine both spending and time burdens, so that money saved does not come at the cost of more unpaid management work.
I would like an AI report card that looks closer to everyday life: are basic services easier to obtain? Have low-income households’ actual burdens fallen? Do people have more choice when seeking services?
Technological abundance has to earn its value through these concrete changes. It can begin in a laboratory, but it will not be completed there.
If you could make AI reduce one cost first, would you choose housing, care, transportation or education? Where do you think the real bottleneck lies?