· 8 min read
I grew up in a small Moroccan town called Kenitra, which means “the little bridge” in Arabic, القنيطرة. Once a quiet town built around an American base, it sat just off the bridge that gave it its name. My parents were born there. I spent my early childhood beneath its calm skies, where evenings smelled of orange blossoms and faint sea-salt dust, where life moved at the pace of bicycles and slow moments.
Now, each time I return, the Kenitra I knew is harder to find. The small town has become a city, louder, denser, restless. The soft hum of the base has turned into the roar of engines, the quiet roads into industrial arteries. I barely recognise it. The city swells with cars, concrete, and a cacophony so foreign that I retreat into the cocoon of our old neighbourhood, where I can still hear the echoes of the stillness that raised me.
I think about that often, how places change when energy moves through them. How growth hums, sometimes beautifully, sometimes unbearably. I think about that when I look at our world today, a global Kenitra of sorts, expanding, electrifying, digitising, pulsing with invisible currents.
I remember my first phone, a white Nokia flip. 2006, maybe 2007. I was a nineties kid, and that little phone made me feel infinite.
When my parents brought home our first computer, a heavy screen, a whirring base, it felt like a portal. They set it up in the hallway, cautious, protective, as if the internet itself might slip through and steal us. They weren’t wrong. The web was magic, but also something else, something vast, unpredictable, alive. They knew, even then, that connection comes with risk.
And now, I watch as connection becomes intelligence. The child who played Snake now writes about large language models, and I feel a strange nostalgia, not for simplicity, but for tangibility. For the sound of a modem dialing up, for when the world wasn’t entirely stored in a cloud.
Fast forward a few years, and I often find myself grateful to be a nineties kid, part of that in-between generation that grew up with analog dust on its fingers and digital light in its eyes. We’ve watched the world rewire itself in real time. I remember when Google was just a rumor about two students in a California garage; now it’s the very reference for knowledge itself.
We’ve lived through it all, the rise of start-ups and scale-ups, the dizzying birth of fintech, the breaches and scandals that redrew the boundaries of trust, the anarchy of crypto, and now, the calm inevitability of artificial intelligence.
AI itself isn’t new. Algorithms and language models have existed quietly in the background since those early search-engine days. But generative AI changed the atmosphere; it made intelligence visible, accessible, conversational, almost tangible. It didn’t just compute, it created.
For years, the very ideas of blockchain and artificial intelligence were dismissed as hype, much as sustainability and ESG once were, idealistic, overpromised, too abstract for the real world. Yet here we are, in 2025, standing at the hinge of a new era. The truth is simple: if you don’t embrace the transformation AI brings, the way it’s reshaping our work, our language, and even our sense of humanity, you risk becoming fluent only in a past that no longer exists. Like the pandemic years that redefined our collective fragility, this technological shift isn’t waiting for permission. It’s already rewriting what it means to live, to think, to belong.
But there’s a truth humming beneath the poetry. Intelligence, whether human or machine-made, is greedy. It eats energy. It needs light, power, water, the tangible things of the world. Behind every chat, every query, every answer, every text message to a date, every email to a colleague, every list or fleeting thought, there are data centres the size of cathedrals, roaring, glowing, cooling, consuming. The cloud is not weightless; it is built of earth and steel and energy. The digital is deeply physical.
I remember sitting once in the House of Lords, listening to three men, two Americans and one Brit, speak about green data centres. We’d all been invited by a VC fund to discuss the intersection of technology and sustainability. I was one of the few corporate guests there, attending in a personal capacity while still with Accenture. Around me were politicians, founders, and investors, all fluent in the dialect of disruption. The trio spoke about their data centres, how theirs would be different, more ethical, less hungry. I couldn’t recall their names later, and couldn’t find them online, but the idea stayed with me. I began to read, to research, to realise there is so little being said about the environmental pulse of our digital age. We celebrate intelligence but ignore its heartbeat.
Large language models, these miracles of code, require staggering amounts of power, megawatts of energy, oceans of cooling water. The more we build them, the more they build us, and the more they demand. It’s a beautiful paradox: our brightest creation could darken the planet that made it possible.
In 2023, global data-centre electricity consumption exceeded 460 terawatt-hours, roughly two percent of all electricity used on Earth, about the same as the total energy demand of Sweden and South Africa combined. Analysts predict this could double by 2027, driven largely by the power-hungry compute required for AI training and inference.
To put that in perspective: running ChatGPT for a single day consumes as much electricity as powering thirty thousand U.S. households. Training GPT-4 is estimated to have used nearly one thousand megawatt-hours of electricity, enough to power one hundred and twenty average homes for a year. A hyperscale data centre can draw one hundred megawatts of power, the same as a medium-sized town, and many campuses now exceed one gigawatt, enough to light a million homes. Cooling alone accounts for thirty to forty percent of total consumption, much of it from evaporative systems that depend on fresh water.
In Northern Virginia, the world’s largest data-centre hub, facilities consumed nearly two billion gallons of water in 2023, a sixty-three percent increase since 2019. In Ireland, Google and Meta together used more than twenty billion litres that same year, straining municipal supplies during droughts. The cloud, it turns out, is made of land.
And yet, there is innovation in this tension. Some pioneers are rewriting the physics of infrastructure. In Hamina, Finland, Google’s data centre uses seawater from the Gulf of Finland for cooling, a remarkable conversion of a 1950s paper mill into a sustainable digital hub. In Odense, Denmark, Meta pipes its servers’ waste heat to warm over ten thousand homes, turning digital exhaust into domestic comfort. In Sweden, engineers are experimenting with data-centre co-location near wind farms, capturing residual heat for greenhouses and aquaculture. And in the United States, Microsoft has signed an agreement with Helion Energy to source electricity from nuclear fusion by 2028, an audacious promise to power intelligence with the same process that fuels the sun.
We are entering an age where the sustainability of intelligence may matter as much as its sophistication. Large language models are trained on vast oceans of data, but their true cost lies in the terawatt-hours of energy required to teach them to predict, reason, and write. A single training run of a major AI model can emit over five hundred tons of carbon dioxide, roughly equivalent to flying a passenger jet across the Atlantic three hundred times.
This is not a reason to retreat from progress, but to engineer it with conscience. The challenge ahead isn’t whether machines can think; it’s whether humanity can learn to think sustainably.
When I began my career in civil and environmental engineering, my obsession was how to decarbonise infrastructure, bridges, cities, power systems. Today, those same questions apply to the architecture of the digital world. The bridges we now build are made of data, not steel; the flows we manage are electrical, not hydraulic. But the ethics remain identical: how to connect without consuming, how to design without depleting.
A data centre, at its core, is a living ecosystem. It draws water, generates heat, and demands electricity. With foresight, it can also give back, supplying waste heat to communities, balancing renewable grids, recycling water through closed-loop systems. Even AI can help: adaptive cooling algorithms like DeepMind’s optimisation system cut Google’s cooling energy by forty percent, simply by letting an algorithm tune the variables humans overlooked.
If we can build intelligence that learns empathy, surely we can build infrastructure that learns restraint.
I used to fear a future where machines replaced us. Now I dream of one where they remind us of our humanity. Because the truth is, AI doesn’t strip us of meaning, it reveals how desperately we seek it.
And as I write this, in my quiet apartment hundreds of miles from Kenitra, I think of my parents old warning gaze, the hum of that first computer, the simple, cautious awe of discovery. I think of how they feared the unknown, and how that fear was love. I think, maybe this too, this new wave of intelligence, deserves to be loved carefully.
While AI takes over, may we remember to engineer wisely, imagine boldly, and consume gently, responsibly. Because the future isn’t just built, it’s powered. And every light that turns on in a data centre somewhere, should remind us of the stars we once looked up to before we built our own.
illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.
Curious how major companies measure up on climate? On illuminem’s Data Hub™, explore verified emissions data, net‑zero targets, and sustainability performance of thousands of firms — from industry leaders to emerging innovators.
