The benefits concentrate in wealthy nations; the costs spread globally.
En el momento en que la inteligencia artificial se convierte en infraestructura global, Naciones Unidas advierte que su coste ambiental amenaza con superar sus promesas tecnológicas. Para 2030, los centros de datos que sostienen estos sistemas emitirán tanto CO₂ como el Reino Unido en su totalidad y consumirán agua suficiente para saciar la sed del planeta durante casi dos años. Lo que hace este informe especialmente inquietante no es solo la escala, sino la asimetría: los beneficios se acumulan en pocas manos, mientras la carga se reparte sobre quienes menos la eligieron.
- Los centros de datos de IA alcanzarán en 2030 emisiones equivalentes a las de todo el Reino Unido y un consumo eléctrico comparable al de Francia entera, según el nuevo informe de la ONU.
- El agua necesaria para refrigerar esos servidores —9,3 billones de litros— equivale a 1,6 años de consumo de agua potable para toda la humanidad, una cifra que convierte lo abstracto en urgente.
- Solo el 16% de los países dispone de infraestructura para desarrollar IA, y Estados Unidos y China concentran el 90% de esa capacidad, dejando a las naciones del Sur Global con los costes ambientales pero sin los beneficios.
- Expertos como el profesor Alphonso Valencia reconocen la gravedad del diagnóstico, aunque señalan que las proyecciones podrían cambiar si mejoran la eficiencia energética, se descentraliza el cómputo o surgen nuevas regulaciones.
- Sin intervención deliberada, la trayectoria actual apunta a una profundización de la desigualdad digital y ambiental a escala planetaria.
La infraestructura que hace posible la inteligencia artificial está a punto de convertirse en uno de los mayores consumidores de energía y agua del mundo. Un informe publicado esta semana por Naciones Unidas cifra el impacto: para 2030, los centros de datos necesarios para operar sistemas de IA producirán unos 400 millones de toneladas de CO₂ al año, una cantidad equivalente a las emisiones totales anuales del Reino Unido. En términos de electricidad, el consumo rozará los 1.000 teravatios-hora, tanto como consume Francia en un año, o lo suficiente para abastecer durante cinco años a 1.300 millones de personas en el África subsahariana.
Luego está el agua. Los servidores generan un calor constante que exige refrigeración permanente. La ONU estima que los centros de datos consumirán unos 9,3 billones de litros de agua a nivel global —el equivalente a cubrir las necesidades de agua potable de toda la población mundial durante 1,6 años—. La cifra resulta casi abstracta hasta que se traduce en esos términos humanos.
Pero la carga no se distribuye de forma equitativa. El informe, elaborado por el Instituto de Agua, Medio Ambiente y Salud de la Universidad de las Naciones Unidas, subraya una brecha digital que se agrava: solo el 16% de los países cuenta con la infraestructura necesaria para desarrollar IA, y Estados Unidos y China controlan aproximadamente el 90% de esa capacidad instalada. Los beneficios económicos y tecnológicos se concentran así en un puñado de naciones ricas, mientras los costes ambientales recaen de forma desproporcionada sobre los países en desarrollo del Sur Global.
Alphonso Valencia, director de Ciencias de la Vida en el Centro Nacional de Supercomputación de Barcelona, reconoce la importancia del informe y afirma que "aporta claridad y nos confronta con lo que es, sin duda, un problema enorme". Aunque advierte que las proyecciones podrían variar si se producen mejoras en eficiencia, si el cómputo migra hacia dispositivos individuales o si nuevas regulaciones reorientan el desarrollo tecnológico. La trayectoria no está sellada, pero sin intervención, el rumbo es inequívoco.
The infrastructure that powers artificial intelligence is about to become one of the world's largest consumers of energy and water. A new United Nations report released this week lays out the scale of what's coming: by 2030, the data centers required to run AI systems will produce roughly 400 million tons of carbon dioxide annually—a volume equivalent to the entire yearly emissions of the United Kingdom, one of the world's most industrialized nations.
The cloud that underpins the internet is not some ethereal thing. It is concrete and steel: vast warehouses filled with servers processing billions of data points without pause. That relentless computational activity, essential for AI to function, will consume nearly 1,000 terawatt-hours of electricity within four years. To put that in perspective, it is as much power as France uses in a year, or enough to supply electricity to 1.3 billion people across sub-Saharan Africa for five years. If data centers were a country, the UN research notes, they would rank eleventh globally in electricity consumption.
Then there is the water. Those servers generate tremendous heat and must be cooled constantly to avoid failure. The UN estimates that data centers will draw approximately 9.3 trillion liters of water globally for this purpose. That volume would satisfy the drinking water needs of all 8.1 billion people on Earth for roughly 1.6 years. The scale is almost abstract until you consider it that way—the entire planet's hydration, consumed by machines.
But the environmental burden is not distributed evenly. The UN report, produced by the United Nations University's Institute for Water, Environment and Health, highlights a deepening digital divide. Only 16 percent of nations possess the specialized infrastructure required to build and operate AI systems. Of those, the United States and China control approximately 90 percent of installed capacity. This means that the benefits of artificial intelligence—the economic gains, the technological advancement—concentrate in a handful of wealthy countries, while the environmental costs spread across the globe, falling disproportionately on developing nations in the Global South that lack the resources to absorb them.
Alphonso Valencia, an ICREA professor and director of Life Sciences at Barcelona's National Supercomputing Center, acknowledges the report's importance. He told Spain's Science Media Center that the research "provides clarity and confronts us with what is, without question, an enormous problem." Yet he also notes a caveat: the UN's projections, while grounded in real data, could shift depending on whether efficiency improvements materialize, whether AI computation moves from centralized data centers to individual devices, or whether new regulations reshape how the technology develops. The trajectory is not fixed. But without intervention, the path ahead is clear.
Citações Notáveis
The research provides clarity and confronts us with what is, without question, an enormous problem.— Alfonso Valencia, ICREA professor and director of Life Sciences at Barcelona's National Supercomputing Center
A Conversa do Hearth Outra perspectiva sobre a história
When we talk about AI consuming as much water as the planet needs for drinking in 1.6 years, are we talking about actual scarcity—places running out—or is this more about the principle of the thing?
Both. In water-stressed regions, it's already scarcity. But even in places with abundant water, you're talking about freshwater being diverted to cool servers instead of irrigating crops or filling reservoirs. The principle matters because it reveals where power actually sits.
Which is where?
The US and China control 90 percent of AI infrastructure. They get the innovation, the economic value, the jobs. Meanwhile, a data center in Southeast Asia or Africa cools servers for companies headquartered elsewhere, consuming local water and energy, generating local emissions. The benefits flow north and west. The costs stay put.
The report says only 16 percent of nations have this infrastructure at all. That's a staggering number.
It is. And it's not accidental. Building a data center requires capital, technical expertise, stable electricity grids, cooling systems. Most countries can't afford it. So they become dependent on whoever can—which means they're locked into a relationship where they're absorbing environmental costs for technologies they didn't choose and won't control.
Is there a way out of this, or are we locked in?
The expert quoted in the report mentions three possibilities: efficiency improvements in how data centers operate, shifting computation to devices instead of centralized servers, or new regulations that reshape the whole model. None of those are guaranteed. But they're not impossible either. The question is whether governments act before the infrastructure is too entrenched to change.