I have been reflecting a lot lately on the paradox of Artificial Intelligence. On one side, we celebrate it as the next great leap in human innovation, capable of optimizing industries, transforming economies, and even helping us manage the planet more responsibly. On the other, the very intelligence we are building is hungry, extremely hungry, for energy. And that is something we cannot ignore if we truly believe in sustainability.

From my research, the scale of energy consumption behind AI is much larger than most people realise. The International Energy Agency projects that global electricity demand from data centres could nearly double by 2030, reaching around 945 terawatt hours, mainly because of AI-driven workloads. Goldman Sachs predicts an even steeper rise, up to 165 per cent more power by 2030 compared with 2023, with AI possibly consuming more than a quarter of all data centre energy worldwide. If you are curious, you can easily verify these numbers; they are publicly available from both organisations.

Training large AI models is an energy-intensive process. One model similar in scale to GPT -3 was estimated to consume about 1,287 megawatt-hours of electricity and emit roughly 500 tons of CO₂. To put that in perspective, that is about the same annual energy consumption of a hundred average homes. But what surprised me most in my research is that training is not even the biggest part of the story. Once these models are deployed, they run continuously, serving billions of requests. The so-called inference phase often consumes even more energy than the initial training.

Then there is the hidden footprint: water for cooling, hardware manufacturing, and the growing mountain of electronic waste. Data centres might represent only about 3 per cent of the world’s electricity use by 2030, but when you consider how concentrated this demand is, often in regions already struggling with grid stability or water scarcity, the picture becomes more complex.

Why does AI need so much energy? The answer is simple: scale. These systems rely on thousands of high-performance chips running simultaneously to process massive datasets. The more complex the model, the more power it draws, and the more heat it produces, which in turn requires more cooling. It becomes a loop: intelligence demands energy, and energy fuels intelligence.

In my opinion, the real question is not whether AI is sustainable today, but whether we can make it sustainable tomorrow. I do not see AI as the enemy of the environment; far from it. When applied with intention, AI can be a powerful ally in the fight against climate change. It can optimise shipping routes to save fuel, improve offshore asset management, predict equipment failures before they happen, and analyse marine pollution with precision we could not dream of before.

The issue is balance. We need to ensure that the environmental cost of creating and running these systems does not outweigh the benefits they bring. Efficiency is not a side topic; it is central to AI’s long-term credibility. Research shows that smarter algorithms, model compression, and renewable-powered data centres can drastically cut emissions. Some studies suggest that up to a 75 per cent reduction in AI’s carbon footprint is achievable with existing methods.

That is why I believe sustainability must become a design principle in AI development, not an afterthought. Companies should measure and disclose their AI energy usage just as they do with other sustainability metrics. Transparency is not only good ethics, but it is also good governance.

You can Google most of these figures yourself, but beyond numbers, what matters is the mindset. AI is not the villain of our story; it is a mirror reflecting our collective priorities. If we build intelligence without responsibility, we will only shift the problem elsewhere, from carbon in the air to heat in the servers.

In my view, the challenge and the opportunity are the same: to make intelligence serve sustainability, not consume it. The smarter our machines become, the wiser our energy decisions must be. That, I believe, is the real measure of progress.

This article is also published on Medium as part of my public research and writing.