En los suburbios de Atlanta, los vecinos notaron cómo el agua dejaba de fluir con normalidad por sus grifos, sin saber que un centro de datos de inteligencia artificial, a treinta y cinco kilómetros de distancia, había extraído más de cien millones de litros del acuífero que compartían. Lo que ocurrió en Annelise Park no es una anomalía local, sino el síntoma visible de una tensión más profunda: la expansión ilimitada de la inteligencia artificial choca, por primera vez, con un límite que ningún capital puede disolver. El agua, a diferencia de la valoración bursátil, no se puede fabricar.
Tech's Water Crisis: AI Data Centers Drain Aquifers as Musk Eyes Billionaire Status
Residents in Annelise Park, Georgia experienced alarming water pressure drops due to nearby AI data center extraction of 100+ million liters from local aquifers.
The servers need water. The water is running out.
Mark
Why does an AI data center need so much water in the first place?
Mimi
It's all about cooling. These servers generate enormous amounts of heat—they're essentially thousands of computers running simultaneously at maximum capacity. Water is the most efficient way to dissipate that heat. You pump water through the systems, it absorbs the heat, and then you have to cool that water back down before you can reuse it. It's a cycle that never stops.
Mark
And there's no alternative? No way to make the servers less thirsty?
Mimi
Not at the scale we're talking about. You could use air cooling, but it's less efficient and requires more energy. You could locate data centers in cold climates, but then you're dealing with transmission losses and infrastructure costs. The industry has chosen water because it works. The problem is that choice was made without accounting for what happens when you do it everywhere at once.
Mark
The article mentions Madrid's water supply. Is that just a comparison, or is it actually relevant?
Mimi
It's both. It's a way to make the number comprehensible—most people understand a city's water consumption. But it's also a warning. Madrid is a major metropolitan area with millions of people. If AI data centers are using the equivalent of Madrid's annual supply, we're talking about infrastructure that was never designed to support that demand. And unlike a city, you can't just build another data center somewhere else. The servers need to be where the electricity is cheap and reliable.
Mark
What happens when the water runs out?
Mimi
That's the question nobody wants to answer yet. In places like Georgia, you're already seeing it—people's taps running dry because a single facility drained the aquifer. Multiply that across the country, across the world, and you're looking at a genuine constraint on growth. You can't scale your way out of a water shortage with better algorithms.
Mark
So the trillion-dollar valuations assume unlimited resources?
Mimi
They assume resources will be available at current prices. But if water becomes scarce, if electricity becomes more expensive because you have to generate it from less efficient sources, if you have to locate data centers in places where water is abundant but infrastructure is poor—all of that changes the math. The companies know this is coming. They're just hoping they can solve it before it becomes their problem.
O Pulso
- Los residentes de Annelise Park, Georgia, vieron caer la presión de sus grifos sin explicación aparente, hasta que las autoridades rastrearon el origen: un centro de datos de IA había vaciado silenciosamente el acuífero vecinal.
- El consumo anual de agua de los centros de IA equivale a cuatro o cinco meses del gasto total de Madrid; si se suma el agua necesaria para generar la electricidad que demandan, la cifra escala a dos años completos de suministro de la capital española.
- La fiebre de expansión tecnológica no se detiene: Musk apunta a convertirse en el primer billonario del mundo, Altman prepara la salida a bolsa de OpenAI, y la empresa NextEx pagó 66.800 millones de dólares para controlar el mayor corredor de centros de datos del planeta, cerca de Washington.
- A diferencia de las crisis financieras, que admiten ingeniería contable y maniobras políticas, la escasez de agua es un límite físico: el acuífero no puede ser refinanciado ni el algoritmo puede negociar con la sed de sus vecinos.
- La sostenibilidad hídrica emerge como el cuello de botella que podría frenar la expansión de la IA antes de que lo haga cualquier regulación o corrección de mercado.
En los suburbios de Atlanta, los vecinos notaron cómo el agua dejaba de fluir con normalidad por sus grifos, sin saber que un centro de datos de inteligencia artificial, a treinta y cinco kilómetros de distancia, había extraído más de cien millones de litros del acuífero que compartían. Lo que ocurrió en Annelise Park no es una anomalía local, sino el síntoma visible de una tensión más profunda: la expansión ilimitada de la inteligencia artificial choca, por primera vez, con un límite que ningún capital puede disolver. El agua, a diferencia de la valoración bursátil, no se puede fabricar.
En Annelise Park, un barrio de las afueras de Atlanta, los vecinos empezaron a notar hace unos meses que el agua salía con menos fuerza de sus grifos. La investigación de las autoridades locales condujo a un centro de datos de inteligencia artificial situado a treinta y cinco kilómetros, que había extraído más de cien millones de litros del acuífero subterráneo para refrigerar sus servidores. El episodio condensa, en una imagen doméstica y perturbadora, la magnitud del problema.
Las cifras globales son aún más difíciles de asimilar. Los centros de datos de IA consumen en un año lo que Madrid gasta en cuatro o cinco meses. Pero si se incorpora el agua necesaria para producir la electricidad que esos centros demandan —torres de refrigeración, embalses hidroeléctricos, toda la cadena de generación energética—, el consumo equivale a dos años completos del suministro de la capital española.
Este escenario coincide con un momento de euforia expansiva en el sector tecnológico. Elon Musk avanza hacia convertirse en el primer billonario de la historia, con planes de sacar SpaceX a bolsa en torno a su cincuenta y cinco cumpleaños. Sam Altman prepara la salida a bolsa de OpenAI con valoraciones que rozan el billón de dólares. Y la empresa NextEx adquirió Dominion Energy por 66.800 millones de dólares, no tanto por su negocio eléctrico tradicional como por el control de un corredor de 4.700 kilómetros cuadrados cerca de Washington donde se concentra la mayor densidad de centros de datos del mundo. Para facilitar la aprobación regulatoria, NextEx realizó una donación de 300 millones de dólares destinada a la renovación del salón de baile de la Casa Blanca.
Lo que distingue la crisis del agua de otras tensiones que enfrenta el sector es su carácter irreductible. Las valoraciones infladas pueden sostenerse con ingeniería financiera; las dependencias energéticas incómodas pueden camuflarse con etiquetas distintas en las facturas. Pero el acuífero no admite refinanciación. Los vecinos de Annelise Park no pueden negociar con un algoritmo. Por primera vez, la industria que ha aprendido a sortear casi cualquier obstáculo se enfrenta a uno que no cede ante el capital ni ante la influencia política.
In the suburbs of Atlanta, residents of Annelise Park began noticing something troubling a few months back: the water pressure in their taps was dropping, steadily and noticeably. When local authorities investigated, they found the culprit thirty-five kilometers away—an artificial intelligence data center that had extracted more than 100 million liters from the aquifer beneath their neighborhood just to keep its servers cool. It was a stark illustration of a problem that threatens to reshape the entire technology sector: the staggering thirst of artificial intelligence.
The numbers are difficult to grasp in their scale. AI data centers consume as much water in a single year as the city of Madrid uses in four or five months. That figure alone is striking. But when you factor in the water required to generate the electricity those centers demand—the cooling towers, the hydroelectric reservoirs, the entire infrastructure of power generation—the consumption reaches the equivalent of two years' worth of Madrid's total water supply. A single year of AI data center operations, measured this way, equals what Spain's capital city uses in twenty-four months.
This crisis arrives at a moment when the technology industry is experiencing a kind of fever dream of expansion and valuation. Elon Musk is positioning himself to become the world's first trillionaire, with plans to take SpaceX public around his fifty-fifth birthday in late June, a move that would likely push his wealth past the psychological barrier of one trillion dollars. Meanwhile, Sam Altman at OpenAI is preparing his own public offering, with valuations approaching a trillion dollars as well. The competition for dominance in artificial intelligence has become a competition for resources—and water is emerging as the constraint that no amount of venture capital can overcome.
The infrastructure race is already reshaping the energy sector. NextEx, a Florida-based utility company, just purchased Dominion Energy in Virginia for $66.8 billion, a transaction that on the surface looks like a standard corporate consolidation. But what NextEx actually wanted was control of a 4,700-square-kilometer corridor near Washington, D.C.—an area slightly larger than the Spanish province of Pontevedra—where the world's largest concentration of data centers sits. Between 250 and 400 facilities operate in that region, and an increasing share of them are dedicated to artificial intelligence, each demanding tens of thousands of gigawatts of electricity. To secure regulatory approval for the deal, NextEx made a patriotic donation of $300 million toward Donald Trump's planned ballroom renovation at the White House.
The water crisis is becoming one of the primary obstacles to AI's continued expansion. Unlike the financial engineering that has allowed companies like Tesla to maintain trillion-dollar valuations despite questionable fundamentals, or the political maneuvering that allows governments to quietly import Russian oil under different labels, there is no accounting trick that can create water where there is none. The residents of Annelise Park cannot negotiate with an algorithm. The aquifer cannot be refinanced. As the technology industry races toward ever-larger models and ever-greater computational demands, it is running headlong into a physical limit that no amount of venture capital, political favor, or stock market enthusiasm can overcome.
Citações Notáveis
The water crisis is becoming one of the primary obstacles to AI's continued expansion, unlike financial engineering or political maneuvering that can be negotiated.— Analysis of industry trends
Quer a matéria completa?
Leia o original em El Mundo ↗
Technology