By Norah Simon | November 2026
For decades, electricity demand was a foregone conclusion. Rising at the morning teakettle, falling with the evening lights, demand could be predicted with remarkable accuracy, and the system ran with the utmost equilibrium. But then came the machine intelligences, and a new hunger.
The question that troubles many people at present is a simple one, and it seems to be quite disquieting: does the omnivorous appetite of the machines draw electricity from the network too rapidly? Will the lights be put out? The short answer is complicated; the long answer tells of the financial instability of the economy, of the difficulties of the national and the international network, of a quiet redistribution of the social contract between states, companies and citizens.
The Scale of the Appetite
By 2026, according to a recent study by Gartner, the electricity used by all data centers in the world will rise by 26% over the figure for 2025 and reach 565 terawatt hours, which is about the annual electricity consumption of a medium-sized European country.3 One third of this consumption is due to the servers designed for artificial intelligence, and by 2030, total data center consumption is projected to grow by more than 50%.
For the training of a large language model, you can expect to need as much electricity as hundreds of households do in a year. For the inference, that is to say running queries on those models, there is a continuous, constant background. Unlike a factory which can be shut down at night, the AI tasks are mostly running continuously, clustered in massive data centres, often clusteredinonespots of the world.
The Real Threat: Volatility, Not Volume
It’s here that the plot becomes more subtle, and more dangerous. The greatest threat posed by an artificial intelligence to the power grid is not the fact that it might use too much electricity. It’d be extreme volatility.
It is not a smooth curve, it is a sharp edge. While the models are being trained, the consumption of energy drops or rises by hundreds of megawatts in a single second. In Virginia, home to the world’s largest concentration of data centers, dozens of data center sites disconnected from the power grid at once, due to a fault, which instantly cut off electricity supply of over 1,500 megawatt, roughly equivalent to the power of a large-scale fusion power station. The power grid swayed. The grid operators scrambled. A more widespread swing was narrowly avoided.
The scenario which keeps power engineers awake at night is this: not a gradually increasing demand, but a sudden, synchronized, universal shock that the grids were never designed to bear.
The Government’s Quiet Anxiety
A government is very nervous, in private. His dread is not an outside enemy but a polycrisis, a simultaneous and recurrent series of threats, and AI’s demand for power is at the crossroads of many of these fears.
- Loss of control over the agenda: Events unfold beyond government control, forcing administrations into a reactive posture.
- Fiscal and social strain: Governments must simultaneously address austerity, cost-of-living crises, and demands for increased military spending. Rising energy costs-driven partly by data center demand-add fuel to public discontent.
- Critical infrastructure vulnerability: Power grids are increasingly recognized as strategic assets. A grid destabilized by AI volatility is a grid more vulnerable to accidents, cyberattacks, and cascading failures.
That is why the wave of survival instructions given to citizens—the emergency packages of a few days, the brochures in Sweden, the “be ready” campaigns in Britain—should be understood in context. They are neither signs of an imminent catastrophe, nor signs of spies or secret machinations, but they are a part of – to be sure – a larger trend towards a more prudent public policy, in a world where all threats are interrelated: cyber-attacks, climatic disasters, political tensions and, yes, even the vulnerability of energy.
The Grid Fights Back
The good news is that the power companies and the regulators are not standing idly by. Countermeasures have already been taken:
- Regulatory Standards: The North American Electric Reliability Corporation (NERC) is expediting the development of mandatory standards requiring large data centers to disclose their electricity usage plans and equip themselves to handle impacts.
- Policy Constraints: Regions like Greece are requiring new data centers to bear the full cost of grid connection and upgrades, and to use at least 60% renewable energy, preventing them from becoming a burden on the public grid.
- Infrastructure Investment: Utilities are investing in grid-scale batteries, demand-response programs, and transmission upgrades to absorb volatility.
The goal is not to stop AI – that train has already left the station – but to make sure that the energy requirements of this AI are controlled, transparently and fairly, and not put on the backs of ordinary consumers.
What This Means for You
For the majority of ordinary households, the most probable scenario is not a long, general blackout, but an increase in energy prices and the risk of brief fluctuations in the supply. In some data-rich regions, such as Northern Virginia, grid operators have already warned that during periods of high demand they might have to resort to “temporary rolling blackouts,” short-term outages meant to prevent the entire grid from failing.
When the AI companies need to draw a great deal of energy, who pays for the grid’s increase? When a data center has to be taken down for any reason, who carries the risk? These are the questions that the governments have only just begun to resolve, and whose answers will determine the relationship between technology, citizenship and infrastructure for the next few decades. The deeper issue is one of justice.
The Bottom Line
AI is not going to turn the lights out tomorrow, but it is making the grid nervous, and nervous grids need to be managed carefully. A combination of an aging grid, climate stress, and political tensions, the challenge of energy is less about quantity than about stability, fairness, and resilience.
Governments are right to warn people of disruption. But the question is, can they hold the monopolists to account for the costs of public work? Can they set up a system of electricity which is flexible enough to endure the shocks of the time of AI?
The Machine was hungry. The question was whether we could feed it without going hungry ourselves.
Norah Simon is an expert on energy policy and technology. This article is based on publicly available information and official statements of November 2026.