We are afraid of the wrong building

Opinion piece · September 20, 2026

Sergio M.L. Tarrero, President of Alianza Futurista

Translated from the Spanish original, written for readers in Spain: the infrastructure, the grid operator, and the draft regulation discussed here are Spanish. Read the original version in Spanish.

A huge warehouse is going up beside a country road. It has its own electrical substation, a water contract, and a sign bearing the name of a company almost nobody in the village can pronounce. Inside there will be computers. Around it, neighbors wondering where that electricity and that water are going to come from.

That scene is repeating itself in Aragón, in Madrid, in Extremadura, and the unease it produces is reasonable. But I think we are aiming it at the wrong object.

Let us start with the numbers, which are less dramatic than the headlines suggest. According to the International Energy Agency, data centers worldwide consumed some 415 terawatt-hours in 2024: around 1.5% of the planet’s electricity. The forecast for 2030 is that this will double, to a little under 3%. And there is a figure that is rarely mentioned: the growth of data centers accounts for less than 10% of the increase in global electricity demand this decade. Industry, the electrification of transportation, and air conditioning all come ahead of it.

Globally, then, it is a modest problem. Locally it is another matter, and there it is worth being demanding. Unlike electric cars, which are spread everywhere, data centers concentrate: a single campus can consume as much as a medium-sized city. Red Eléctrica, the Spanish grid operator, has spent months warning that grid connection points are running out and that reinforcing the network requires investment nobody has budgeted for. In a country heading toward structural drought, cooling water competes with agriculture.

That is not a planetary catastrophe: it is a concrete political problem, and therefore a solvable one. The Government has before it a draft royal decree that would require these facilities to have renewable backing, limits on water consumption, and measures against speculation with grid connection permits. Let us argue about the figures, the deadlines, and who pays for the substation if the company packs up tomorrow. But let us argue about it as what it is: zoning and infrastructure, not the end of the world.

And I would add two reasons not to take it for granted that this will grow forever.

The first is that a good part of this computing will end up moving onto our own devices. It is worth explaining why, because it is not obvious.

Today, when someone puts a question to one of these tools, the question travels to a company’s servers, almost always American, and there it stays. Many people assume this works like an internet search, something that evaporates. It does not. In 2025, a federal court in New York ordered OpenAI to preserve its users’ conversations, including those they had deleted. It then ordered twenty million of them to be handed over in a lawsuit, and in January 2026 a judge upheld the order, reasoning that users had voluntarily handed those texts over and therefore had little privacy to claim over them. Those affected were not told.

What you tell your lawyer is protected by attorney-client privilege. What you tell your doctor is protected too. What you write in an artificial intelligence chat is not: OpenAI itself asked for an equivalent protection to be created and the court refused it. If a law firm works through a case strategy there, if a company pours in its industrial plan, if someone asks about a health problem or a family conflict, all of it sits as documentation in the hands of a third party. And documentation in the hands of a third party can be subpoenaed. Encryption does not solve the problem, because nobody is spying on the connection: what happens is that a company exists which can be ordered to hand the content over.

That is why I think this move will happen, and not out of idealism. It will happen because law firms, clinics, notaries, and engineering firms will discover that there is material they cannot put on somebody else’s server, and that demand will create product. And it will happen because the capacity is arriving: the computing power needed to reach a given level of intelligence has been halving every eight months, so what today requires a building will fit in your pocket within a few years. More laptops are already sold with artificial intelligence capability of their own than without it. On the horizon there are also photonic chips and neuromorphic computing, architectures that promise to do the same calculations with a fraction of the energy. I am not betting on any one of them; I am pointing out that today’s physical limits are not eternal.

The second is that the appetite for building is not infinite either. There comes a point at which more computing power buys little additional capability and, in exchange, adds a great deal of risk.

And here is what nobody photographs from the road.

This month, the heads of the world’s leading artificial intelligence companies—Anthropic, OpenAI, Google DeepMind, and xAI—have publicly called for a slowdown in the rate at which their own systems become more capable. They asked for it themselves, the people who make money from this. The President of the United States replied that whoever wins the artificial intelligence race wins, and that those who sound warnings are negative forces talking about things that are not going to happen. China retorted that Western alarmism gets in the way of global governance, while tightening its own internal control.

The result: there is no binding international rule covering the most advanced systems. None. There are no mandatory minimum safety standards, no independent auditors with real power to halt a deployment, no obligation to report serious incidents.

Let us compare. The data center on the country road has a water meter, a building permit, an environmental review, and a public comment period. It can be measured, photographed, regulated, appealed, and, if need be, stopped. The absence of global rules for frontier artificial intelligence has no meter, cannot be seen, and is bounded by nothing.

We are afraid of the building because the building is visible.

I do not write this out of technophobia. I have spent more than twenty-five years defending these technologies and I believe they can give us a far better world: more health, more years of life, more knowledge. Precisely for that reason we are not going to stop using them, and pretending otherwise is a waste of time. At Alianza Futurista we wrote in July to the Secretary-General of the United Nations calling for an international body comparable to the International Atomic Energy Agency, binding agreements, and mandatory reporting of incidents. And it is worth knowing that Spain co-chairs, together with Costa Rica, the process that created the UN scientific panel on artificial intelligence: on this we have a seat at the table that almost nobody knows about, even though that panel can describe the risk and cannot stop anything.

Because the ceiling on this growth is going to arrive anyway. The only question is whether an agreement sets it or an accident does.

And that, unlike the climate or geology, is not a destiny: it is a decision. There, it seems to me, lies the only solid reason for hope.