Begin with the places where water is already scarce. In Texas, where drought remains a persistent emergency, data centers could consume as much as 399 billion gallons of water annually by 2030, compared with 49 billion gallons in 2025. By one estimate, that volume would be enough to lower Lake Mead, the largest reservoir in the United States, by more than 16 feet in a single year. The machines that make AI-generated output appear effortless still need cooling, and that cooling is increasingly happening in regions with little water to spare — regions often chosen for cheap land, tax incentives, and fiber-optic connectivity rather than long-term water availability.

Now consider where consumers are spending as digital output becomes cheaper. In 2025, a record 159 million fans attended a Live Nation show, revenue exceeded $25 billion, and, for the first time, more attendees came from outside the United States than from within it. The surge aligns with a broader post-pandemic consumer shift toward live experiences — travel, dining, and entertainment — documented across the leisure economy. As machine-made output moved closer to being free, the price of being present for a one-time, shared event continued to rise.

Those facts may appear unrelated. They point to the same economic shift. AI has not eliminated scarcity; it has relocated it. Scarcity is moving downward into the physical inputs AI consumes, and upward into the human experiences it cannot reproduce. Neither form of scarcity is always visible on the dashboards companies use to track what AI makes cheaper.

Many executive teams are asking sensible operational questions: where can AI reduce costs and increase output? Those questions are useful, but they can obscure a more difficult one. When everything a company produces becomes abundant, what becomes scarce, and does the company still control any of it?

One of the clearest early examples comes from the labor market, where freelancers felt the shift quickly. After ChatGPT arrived, some of the workers who might have seemed best positioned were among the first affected. A study of more than three million postings on a global freelancing platform found that, within months, demand for tasks that generative AI performs well fell sharply. Translation into Western European languages dropped about 30%, while highly commodified writing, such as "About Us" pages, fell by half. A separate analysis of the same market found that the most experienced and highest-priced freelancers were hit at least as hard as others. Skill did not protect them.

Their work had not deteriorated. It had become abundant. Abundance, rather than incompetence, destroyed its price.

Abundance can obscure the real constraint

The abundance produced by AI is both genuine and misleading. What becomes abundant is the output itself, not the conditions that make that output reliable or useful. Producing an answer has become easier than ever. Producing the right answer, and being accountable when it is wrong, remains difficult.

The freelance market illustrates the mechanism on a small scale. AI did not only replace some work; it also devalued work that continued to be done by humans. The scarce ingredient was never simply "competent text." Competent text is now close to free. What remained scarce was what that text used to represent: judgment, responsibility, and the assurance that a person had attached their name to the words.

Scarcity becomes harder to see

In the industrial economy, scarcity was often visible and measurable: inventory, seats, engineer-hours. In the AI economy, scarcity becomes more diffuse and sometimes nearly invisible. It can hide in a compute queue behind what appears to be ample supply, or in the seconds of latency that reveal how few machines are actually available. In other cases, the constraint is not technical at all. It may be legal, operational, or tied to the limited attention a customer has left.

These constraints are easy to miss until they cause damage. A streaming service can display a thousand titles instantly, but the real rationed resource is the viewer's evening and the few minutes a person will spend choosing what to watch. A logistics platform can price every route in milliseconds, then encounter a ceiling unrelated to computing power: trucks waiting in line at a single loading dock.

Herbert Simon identified the shift in 1971, observing that an abundance of information creates a scarcity of attention. What was then an insight has become a basic operating condition across sectors. Scarcity does not disappear. It goes into hiding.

Many organizations still behave as if scarcity has been abolished, even while absorbing its consequences: congestion, frustrated customers, and trade-offs that were never made explicitly. They continue measuring units sold and headcount, while the business quietly loses value through scarcities it does not track.

Who controls the new scarcity?

The crowded freelance market makes one point clear: anyone producing cheap output at scale is often renting scarcity from someone upstream. The same applies to companies. When a business drives marginal cost toward zero, it has often moved the bottleneck one step up the chain, to the party that controls compute and the energy that powers it.

Data-center electricity demand rose 17% in 2025 and is on track to roughly double, from about 485 terawatt-hours that year to around 950 terawatt-hours by 2030, with demand from AI-focused sites tripling. Capital spending by the five largest technology companies passed $400 billion in 2025 and is expected to rise another 75% this year. By the end of the decade, the International Energy Agency expects data centers in the United States to consume more electricity than the production of aluminum, steel, cement, and chemicals combined.

The binding constraints are not always talent or capital. They include grid connections, gas turbines, transformers, and gallium, a material essential to the semiconductor chips that AI hardware depends on, of which China refines about 99%. Water now belongs on that list as well. The pressure on reservoirs in Texas is an early example, not an exception. Direct water consumption by U.S. data centers is projected to double or more by 2028, and much of the new construction is occurring in basins that were already water-constrained. The IEA has said these bottlenecks are already limiting its more aggressive growth scenarios. In many cases, a company celebrating near-free production has simply handed durable margin to whoever owns the scarce input. As the agency's director has said, there is no AI without energy.

The same logic applies on the demand side. When everyone can generate a similar offer, what becomes scarce is the place where the customer looks: the channel, or the assistant that filters the choices. Whoever controls that point of attention controls the scarcity that matters. Everyone else competes in the abundant layer.

Price signals what matters

When abundance becomes normal, price changes its role. It stops merely indicating how much is available and begins signaling what is being prioritized.

A premium price no longer reflects only higher cost. It can reveal an underlying scarcity: faster access, lower exposure to risk, or a stronger guarantee. It does not necessarily buy more volume; it buys time and reassurance. A price set too low is no longer only a margin problem. It can trigger a rush that overloads the system and hides trade-offs that should have been made openly.

Price becomes the tool through which a company recognizes a scarcity and decides whether to protect it or charge for it. That is a strategic decision, not just an arithmetic one. Because AI can measure once-vague constraints in real time, it can make those choices sharper and their consequences immediate.

Some companies are already charging for things that few would have thought to sell five years ago. A cloud provider sells not just capacity, but guaranteed priority when demand spikes. A financial firm stops selling data, which has become abundant, and instead sells traceability of origin and confidence that the data will withstand regulatory scrutiny. In construction software, Graitec has built its AI strategy around a similar premise, betting that engineers will pay not merely for faster generation, but for designs auditable enough to sign their names to. A wealth manager charges less for the portfolio, an allocation a robo-advisor can now assemble for a few basis points, and more for the judgment that keeps a client invested during a crash, the coaching Vanguard says advisor's alpha adds beyond stock selection, and the accountability no model will provide.

In each case, price stops rewarding the effort of production and starts assigning value to a scarcity the customer cannot find elsewhere.

The question executives often miss

Leaders navigating this transition most effectively share a habit. They do not only ask what AI allows them to produce in greater quantities. They can state clearly what in their business has become abundant, and what has become rare.

Three questions reveal where a company stands. What has become abundant in the business that is still priced as if it were scarce? What has become scarce that is not yet being measured? And among those new scarcities, which does the company truly control, and which have already been captured by a platform, compute provider, or competitor without management noticing?

The trap is usually the same. Faced with abundance, companies often try to compete harder on the abundant layer by producing more of it, faster. That is precisely the layer where margins tend to collapse toward zero, because competitors have access to the same tools. Value migrates to the bottleneck and settles with whoever controls it.

Freelancers who recovered did not succeed by producing more words. Those now doing well repositioned around what remained scarce. Freelancers with AI-complementary skills earn around 40% more than peers who do not, and higher-value contracts have increased. They returned to selling scarcity.

AI will continue making output abundant. That is its promise, and much of it represents progress. But abundance is not where margin resides. Margin moves to whatever remains scarce: access that is difficult to secure, trust that is difficult to earn, and power and priority that cannot simply be created on demand.

The companies that produce the most with AI will not necessarily capture the value. Value goes to whoever can identify the new scarcity, control it, and price it before a competitor does.

That requires discipline rather than instinct. Every strategy now needs a scarcity map: what the company is making abundant, where that creates a bottleneck, who controls that bottleneck, and whether pricing reflects any of it. The difficult part is building the business around that scarcity instead of the output AI has made cheap, especially while every incentive still pushes toward producing more.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.

This story was originally featured on Fortune.com.