Why Cheaper AI Could Mean More of It, Not Less

Making a powerful technology cheaper does not always shrink demand for it. Often it does the opposite, and that pattern is worth keeping in mind as the cost of running AI falls.

Most coverage of artificial intelligence focuses on what the technology can do. The figure I keep coming back to is a quieter one: what it costs to run. That number, more than any single capability, will shape how far AI spreads and how much it ultimately changes the economy.

The number behind the question

There are estimates suggesting that the cost of processing for a model with one trillion parameters could fall by roughly 90 percent by 2030. That is not a modest discount. It is the price of a powerful capability dropping to a tenth of what it is today.

The instinct, when a price falls like that, is to assume spending on it falls too. Cheaper AI, lower AI bills, end of story. But that instinct is usually wrong, and the reason has a name.

Efficiency does not lead to thrift

It is called the Jevons paradox: improvements in how efficiently a resource is used can increase, rather than decrease, the total amount consumed. The original case came from coal. More efficient steam engines did not reduce coal consumption. They expanded it, because efficiency made coal powered machinery worthwhile in far more settings.

This is not a one off historical quirk. The same shape shows up again and again. When lighting became cheaper, people did not pocket the savings and sit in the dark a bit longer. They lit more rooms, for more hours, in more buildings. When computing power grew cheaper decade after decade, the computing industry did not shrink to a quiet corner of the economy. It spread into nearly everything we do.

The pattern is consistent enough to treat as a rule of thumb. Make a genuinely useful capability cheap enough, and people invent uses for it that were unthinkable at the old price.

How this plays out for AI

Picture the organizations sitting just outside AI's reach today. A small business. A district of rural schools. A clinic with a tight budget. A single team inside a larger company with an idea that does not quite justify the expense. At current prices, the math does not work for them.

Now cut that price by 90 percent. All of those uses that were just barely not worth it tip over into being clearly worth it, and they do so more or less at once. That is the heart of the paradox. Demand does not creep up gently as the price eases. It can jump, because whole categories of use become viable at the same moment. And each new use tends to suggest the next one, so the growth compounds on itself.

The part that matters for workers

As someone who has spent a career studying labor markets, this is where I pay closest attention. The widespread fear is straightforward: cheaper, more capable AI replaces people and shrinks the demand for work.

The Jevons lens complicates that story. If falling costs set off a genuine expansion in how and where AI is used, that surge generates its own demand for human effort, for the people who design, deploy, supervise, and maintain these systems, and for the new kinds of work that fresh processes always seem to require. None of this dismisses the real disruption that specific jobs and workers will face, which deserves serious attention and serious policy. But it does undercut the assumption that cheaper automatically means fewer jobs. If history is any guide, the likelier result is a reshaping of work rather than a simple subtraction of it.

Planning for the cheaper world

The practical takeaway is about which world you plan for. It is tempting to make decisions based on what AI costs to run right now. The more useful question is what happens when that cost falls sharply and demand rushes in to fill the space.

That is a different mindset. It means looking ahead to uses that do not exist yet rather than only fine tuning the ones that do. It means recognizing that the binding constraint may shift away from cost and toward the harder problems of capacity, energy, infrastructure, and skilled people. What the coal story really teaches is that using something more efficiently and using less of it are two different outcomes, and they often pull in opposite directions. With AI, cheaper is likely to mean more, quite possibly far more, which is exactly why I keep my eye on that cost curve.

Adriana Kugler, Ph.D., is a labor economist and professor at Georgetown University.