Every time you use an AI tool, somewhere a building the size of a warehouse gets hotter. Here is what that actually means.
The Question Nobody Thinks to Ask
You type a question. A few seconds later, an answer appears. It feels like thinking. It feels weightless.
It is not weightless.
What just happened, invisibly, was this: your query travelled to a data centre, where thousands of specialised processors worked in parallel to generate your response, drawing electricity the entire time. That data centre runs twenty-four hours a day, every day, regardless of whether you are using it. It needs cooling. Cooling needs more electricity. The whole operation requires a power supply that never flinches, never dims, and never takes a weekend off.
Now multiply that by every AI query, every training run, every background inference happening simultaneously across hundreds of facilities on every continent. The numbers become very large, very fast.
What the Numbers Actually Say
Global data centres consumed around 415 terawatt-hours of electricity in 2024, roughly 1.5 percent of total world electricity consumption. To put that in human terms, it is comparable to the entire annual electricity demand of a major industrialised country.
That figure is already extraordinary. What follows is more striking.
From 2024 to 2030, data centre electricity consumption is projected to grow at around 15 percent per year, more than four times faster than the growth of total electricity consumption from all other sectors combined. By 2030, the IEA projects data centre consumption will more than double to around 945 TWh, roughly equivalent to Japan’s total electricity use today.
AI is the primary driver. Electricity consumption in accelerated servers, mainly driven by AI adoption, is projected to grow at 30 percent annually in the base case.
The grid these facilities depend on was not designed for this. Much of the infrastructure in the United States, which accounts for the largest share of global data centre consumption, was built decades ago. One study from Carnegie Mellon University estimates that data centres and cryptocurrency mining could lead to an 8 percent increase in the average US electricity bill by 2030, potentially exceeding 25 percent in the highest-demand markets.
This is not a complaint about artificial intelligence. AI is genuinely transformative, and its potential applications in medicine, science, and logistics are real. This is an examination of a structural collision: the most power-hungry technology the world has ever deployed is arriving faster than the energy system built to supply it.
Why the Obvious Solutions Are Not Enough
The standard response to AI energy demand is: build more renewables, add more nuclear, upgrade the grid. These are all reasonable directions. They are also slow, expensive, and they share a fundamental limitation that data centres expose more clearly than almost any other application.
AI infrastructure requires continuous, reliable power. Not 80 percent of the time. Not when the sun is shining or the wind is blowing. Every hour, every minute, without interruption. A data centre that loses power does not just slow down. It stops, and the consequences cascade across every service it supports.
Renewables currently supply about 27 percent of the electricity consumed by data centres globally, with natural gas as the largest single source in the United States at over 40 percent. Solar panels do not generate at night. Wind turbines stop when the air is still. Batteries store energy for hours, not days. The gap between what intermittent renewables can promise and what AI infrastructure actually requires is real, and it is not filled by good intentions.
The grid itself adds another layer. Transmission capacity, permitting timelines, and the physical reality of building new high-voltage lines through populated areas mean that even when new generation comes online, delivering it reliably to the concentrated clusters where data centres sit is a separate and difficult problem. Nearly half of data centre capacity in the United States is in just five regional clusters, placing intense and localised demand on electricity infrastructure that was never designed for this concentration.
A Different Kind of Answer
There is a design question worth asking here. What if the power source came to the data centre, rather than the data centre competing for space on an overloaded grid?
This is not as abstract as it sounds. It is the structural premise behind the work of the Neutrino® Energy Group, whose international team of engineers and scientists has spent years developing technology that converts multi-channel ambient energy flux directly into stable electrical current. The inputs, including particle momentum transfer, cosmic muon flux, electromagnetic fluctuations, and thermal gradients, are not only continuous. They are present everywhere, at all times, regardless of weather, geography, or time of day.
The governing framework is the Schubart Master Formula, developed by Holger Thorsten Schubart, the Architect of the Invisible:
P(t) = η × ∫V Φ_eff(r,t) × σ_eff(E) dV
The formula defines power output as a function of conversion efficiency, effective ambient flux, and the interaction cross-section of the engineered material stack. None of those variables depend on the sun, the wind, or a transmission line. The output is continuous because the inputs are continuous.
The Neutrino Power Cube delivers 5 to 6 kilowatts of continuous net output with no fuel, no moving parts, and no emissions. The architecture is modular and parallel: units can be deployed in any number, stacked to meet the load requirements of any facility, without permits for grid connection, without transmission infrastructure, and without the capacity auction bottlenecks that are already driving electricity prices higher in the markets where data centres concentrate. “The real transformation begins,” Schubart has said, “when we replace the fear of scarcity with an understanding of abundance.”
The Collision Course and What Reroutes It
The current energy transition was designed around a different demand profile than the one AI is creating. It assumed that electrification would grow steadily, that load would be distributed geographically, and that storage plus grid upgrades would cover the gap between intermittent generation and real-world demand. AI has disrupted each of those assumptions simultaneously.
The answer is not to slow AI down. The answer is to stop treating baseload power for computation as a grid problem and start treating it as a materials problem, which is what the Neutrino® Energy Group has been doing for years. A distributed, modular, continuously generating architecture does not replace the grid. It removes the dependency on the grid for the most demanding and most concentrated load profile the energy system has ever encountered.
“Energy is not something we create,” Schubart has said. “It is continuously present; we simply need to learn how to harvest it.”
Every data centre drawing power from a grid that was never built to carry this load is asking the wrong question. The right question is not where to find more capacity on the existing network. It is how to bring generation to the load, continuously, at any scale, anywhere it is needed.
That question already has a physics-grounded answer. The engineering is catching up.