The Question
Here is a puzzle physicists have enjoyed for years: how much does the internet weigh? Information is carried by electrons moving through wires and by photons — particles of light — racing through fibre. If you add up the tiny extra mass of all the electrons that are in motion representing data at any single instant, one popular estimate lands at around fifty grams. That is roughly the weight of a single strawberry, or a few sugar cubes. The figure is a physicist's illustration rather than a measured quantity, and different assumptions give different answers, but the point survives every version: the thing that connects eight billion people is essentially weightless.
That makes what comes next genuinely counterintuitive. This weightless system is an industrial-scale energy consumer, and its appetite is climbing fast. The question is not whether digital infrastructure uses a lot of power — it plainly does — but how high that share will climb, and how quickly. The answer matters for electricity bills, climate targets, and whether the grids we already have can cope with what is coming.
What the Evidence Shows
The numbers are large and moving quickly. In its 2025 Energy and AI report, the International Energy Agency estimated that data centres — the warehouses full of computers that store and process the world's data — consumed around 415 terawatt-hours in 2024, about 1.5% of global electricity. In its base case, that more than doubles to around 945 terawatt-hours by 2030, slightly more than Japan uses today, or just under 3% of global demand. Even in its high-growth case, data centres reach about 4.4% of global electricity in 2035; in its high-efficiency case, less than 2%.
Data centres are only part of the story. Add the networks that move data across continents, the billions of phones, laptops, and connected devices drawing power at the edges, and the electricity spent mining cryptocurrencies, and the total digital footprint is larger than the data-centre figure alone — though estimates vary widely depending on what is counted. There is also a hidden cost in water: large facilities can use substantial quantities for cooling, straining supplies in dry regions.
"AI is one of the biggest stories in the energy world today – but until now, policy makers and markets lacked the tools to fully understand the wide-ranging impacts."
— Fatih Birol, Executive Director of the International Energy Agency, April 2025The great uncertainty is whether efficiency can keep pace with demand. Every generation of chips does more calculation per unit of energy, and new cooling methods — such as running liquid directly over processors instead of blowing cold air — cut waste sharply. History shows these gains are real and large. But artificial intelligence is expanding so fast that total consumption is rising even as each computation gets cheaper to run. Efficiency is winning the race per calculation and losing it in total, and that tension is what makes the 2035 figure a forecast rather than a certainty.
The internet is the lightest object humans have ever built, and one of the hungriest. Both statements are true at once.
Why This Is Happening
Artificial intelligence has changed the arithmetic. Training and running large AI models consumes far more power than traditional web services, and the biggest technology companies are racing to build enormous new data centres, some drawing as much electricity as a small city. This is the largest force pushing digital energy demand upward. But even the IEA's fastest-growth scenario has data centres at well under a twentieth of global electricity in 2035.
Demand for data keeps compounding, decade after decade. Streaming video, cloud storage, video calls, and always-connected devices have made data traffic grow relentlessly. Each convenience feels weightless to the user, which is exactly why consumption is easy to overlook — the cost is hidden in distant buildings and on someone else's electricity bill. Invisible demand is the hardest kind to restrain.
The infrastructure is being built ahead of the grid. Companies are commissioning data centres faster than utilities can add clean power, so some operators are turning to dedicated generation — including new deals to buy nuclear and renewable output directly. This locks in years of rising consumption and puts real strain on regional grids, in some cases delaying the retirement of fossil-fuel plants that were due to close.
What Could Happen
AI-driven data-centre growth runs ahead of even the IEA's high case, and the combined footprint of data centres, networks, devices, and crypto rises to roughly a tenth of global electricity by 2035.
Data-centre demand roughly doubles or triples, as the IEA expects, and networks and devices add more — but the total digital share of global electricity stays well below ten percent in 2035, because the rest of the world's electricity use is also growing fast as transport and heating electrify.
AI adoption accelerates beyond current expectations and efficiency gains lag, driving digital electricity use above ten percent and forcing emergency grid measures, higher power prices, and hard political choices about who gets electricity first. Possible, but efficiency improvements and grid limits make this extreme outcome less probable than the central case.
What Can We Do
Individuals cannot slow this trend much alone, but informed choices by consumers, companies, and policymakers together shape whether the digital surge is powered cleanly or carelessly.
Push for transparency on digital energy use. Cloud and AI providers rarely disclose the electricity and water behind their services. Demand — as a customer, investor, or regulator — that they report it clearly. What gets measured gets managed, and honest numbers are the precondition for any sensible policy or purchasing decision.
Favour clean-powered infrastructure. The problem is not that data centres use power; it is where that power comes from. Support and choose providers that match their consumption with genuinely new renewable and nuclear generation, rather than draining existing clean supply that others would have used.
Value efficiency as much as capability. More efficient chips and cooling are the single biggest lever for keeping the footprint in check. Policymakers can set efficiency standards for data centres, and buyers can reward the most efficient providers. Every calculation made cheaper to run buys the grid breathing room.
Keep the paradox in mind. The internet feels free and weightless, so its physical cost is easy to ignore. Simply understanding that every stream, query, and AI request draws real power somewhere makes both citizens and leaders more likely to treat digital growth as the serious energy question it has become.