Who bears cost for unbearable weight of resource-hungry AI?

Given the boom in artificial intelligence, the industry has been made to look like it is weightless and comes without cost. But there is no such thing as a free lunch.

A growing body of research is shedding light on the uncomfortable reality that AI is emerging as one of the world's most resource-hungry industries. According to a recent report by the United Nations University Institute for Water, Environment and Health, by 2030, AI-driven data centers could consume 945 terawatt-hours of electricity annually — roughly triple the combined electricity consumption of Pakistan, Bangladesh and Nigeria. Their water use could also reach levels equivalent to the basic needs of 1.3 billion people.

For years, Silicon Valley sold the digital economy as a cleaner alternative to heavy industry. The internet was supposed to dematerialize economic activity. Instead, AI is rematerializing it. The cloud turns out to be made of steel, concrete, copper, electricity and water.

The challenge is that AI companies currently enjoy a remarkable privilege: they are allowed to externalize a significant portion of their environmental costs.

When a data center is built in a drought-prone region, local residents often pay the price through strained water supplies and higher utility bills. When electricity grids require upgrades to accommodate massive computing clusters, the costs are frequently socialized while the profits remain private. Gains accrue to shareholders and technology executives; infrastructure burden falls on everyone else.

It is astonishing that many AI companies disclose model capabilities in meticulous detail while revealing comparatively little about the electricity, water and land footprints of their systems. The European Union is already moving toward mandatory sustainability disclosures for large data centers. That should become the global norm. Citizens have a right to know how much water a chatbot consumes, just as consumers know the fuel efficiency of a car.

Economists have long argued that markets work best when costs are visible. If an AI model requires vast computing resources, that expense should be reflected in its business model rather than hidden in public utility bills. By extension, AI companies should pay the full costs of grid expansion, water extraction and environmental mitigation associated with their operations. Governments should reward efficiency rather than scale alone. The industry's current incentive structure resembles an arms race in which larger models automatically attract more investment. Yet UNESCO research suggests that smarter design and model optimization can dramatically reduce energy consumption without sacrificing performance. The future of AI should be measured not only by intelligence per parameter, but by intelligence per watt.

Policymakers should resist the temptation to frame this as a choice between innovation and regulation. It is neither. Aviation did not flourish because governments ignored safety. The pharmaceutical industry did not gain legitimacy by avoiding oversight. The most successful technological revolutions have always combined entrepreneurial dynamism with public guardrails.

AI's environmental footprint is not an argument against AI. Indeed, the technology may help optimize power grids, accelerate scientific discovery and improve climate forecasting. But those benefits will lose political legitimacy if communities conclude that they are sacrificing water, land and affordable electricity so that technology companies can train ever-larger models.

The defining question for the next decade, therefore, is not whether AI will become more powerful. It almost certainly will. The question is whether society can ensure that the industry's resource bill is paid by those generating it, rather than by everyone else. If AI is to become humanity's most transformative technology, it must first learn a very human lesson: there is no such thing as a free kilowatt-hour.