Discover the dilemma of AI energy consumption and carbon footprint, where progress tests sustainability

Preface

I have long felt that time itself is not a straight line but a living fractal…

Artificial Intelligence has become the defining tool of our era. It promises to diagnose diseases faster than doctors, balance national power grids, and even model the planet’s changing climate. However, behind the bright headlines and enthusiastic press releases lies a quieter, darker story: AI is devouring energy, and in doing so, it may be writing its own warning.

a hungry machine

The scale of power needed to build modern AI systems is staggering. When OpenAI trained GPT-3, it burned more than 1,200 megawatt-hours of electricity, enough to power hundreds of American homes for a year. The result was approximately 500 tons of CO₂ released into the atmosphere.¹ And that was just the training phase. Once a model is live, it never stops eating: millions of queries come in, servers run 24 hours a day, and cooling systems work to keep the heat at bay.

Today, data centers already consume as much electricity as a medium-sized country between Saudi Arabia and France. By 2026, their appetite is expected to double. In the United States, the share of national electricity used by data centers is expected to increase from less than five percent in 2023 to nearly twelve percent in 2028.² In simple terms, the lights in one in eight homes could be on just to keep the machines thinking.

It’s not just electricity either. Training a model like the GPT-3 once consumed around 700,000 liters of water in cooling.³ If current trends continue, AI’s global water use by 2027 will exceed what the entire UK uses in a year.⁴ And the servers powering this revolution, refreshed every few years, will add millions of tons of e-waste to an already overflowing landfill problem.

Progress with a shadow

AI is often hailed as the tool that can help us solve climate change. Algorithms already improve the distribution of renewable energy, predict extreme weather conditions and adjust agricultural efficiency. Yet the irony is stark: the very technology designed to help stabilize our planet, because of its size and hunger, is accelerating the tension.

Governments and corporations are in a bind. If they slow AI development, they risk being left behind in an innovation race that is reshaping the global economy. If they continue, they will deepen the environmental footprint. It’s a modern version of the old industrial paradox: coal once gave us railroads and electricity, but it also poisoned the air; Oil gave us cars and global trade, but locked us into a century of carbon emissions. AI may now be going down the same path.

The deepest fear

When people talk about AI, the conversation often drifts toward a familiar anxiety: the idea that machines might one day outthink or outmaneuver humanity. The cinematic image of robots turning against their creators has always dominated the imagination. But perhaps the real danger is quieter, more prosaic. AI does not need to rebel to undermine its creator. It may simply demand more energy than we can sustainably provide, straining our power grids, draining our rivers, and pushing carbon targets further away.

This is how the decline of species begins: not with a dramatic takeover, but with the inability to sustain the systems they themselves built. The more powerful the AI becomes, the more it consumes. And by struggling to keep pace, humanity risks weakening its own foundations.

Searching for a way forward

The industry is not blind to the problem. Researchers are finding ways to make algorithms more agile, eliminating redundant calculations without sacrificing accuracy. Newer models are being designed to require only a fraction of the computing power of their predecessors. Technology companies are rushing to locate data centers in colder climates or power them with renewable energy. Some are experimenting with edge computing (processing data closer to where it is generated rather than funneling it to massive central servers) to reduce transmission costs.

However, these solutions run into a familiar dilemma: efficiency often drives demand. As models become cheaper and easier to use, more people use them and the overall footprint continues to grow. Economists call this the Jevons paradox, and it haunts every industry, from energy to agriculture. AI is unlikely to be an exception.

A matter of choice

The truth is that the dilemma is not purely technical. It is also ethical. Do we continue to push AI at any cost, trusting that future innovations will solve the problem? Or do we begin to demand a slower, more deliberate path, one that measures progress not just in teraflops and feeds, but also in the carbon and water it consumes?

Communities are already feeling the pressure. In Memphis, Tennessee, for example, residents discovered that a hastily created new artificial intelligence data center was running unauthorized gas turbines. Nitrogen dioxide in the area increased nearly 80 percent, raising real public health fears in a neighborhood that was already struggling.⁵ History is a reminder: this is not an abstract problem. It has human faces, human lungs and human costs.

Holding the line

Artificial Intelligence is not going to disappear. Nor should it be. It has the potential to revolutionize medicine, energy and education in a way that no previous technology has matched. But intelligence, whether natural or artificial, is not measured solely by speed or power. It is measured by foresight, moderation and balance.

If humanity cannot satisfy the energy hunger of the machines it has created, then those machines will become less of a triumph than a burden. Perhaps that’s the most sobering twist of all: the idea that AI could degrade its creator not through conquest, but through consumption.

The smartest future is one in which progress and sustainability are not enemies but partners. Anything less risks leaving us with shiny machines in a diminished world.