A single query consumes what an oven uses in one second
En la carrera por construir inteligencias artificiales cada vez más capaces, Sam Altman ha puesto sobre la mesa una verdad incómoda: cada pregunta que millones de personas hacen a ChatGPT tiene un coste físico real, medido en electricidad y agua. Lo que parece insignificante a escala individual —una fracción de teaspoon, un segundo de horno— se convierte en una crisis de infraestructura cuando se multiplica por 2.500 millones de consultas diarias. La humanidad se enfrenta, una vez más, a la paradoja del progreso: las herramientas que prometen liberarnos exigen, para existir, recursos que el planeta no puede conceder sin límite.
- Sam Altman ha roto el silencio cómodo de la industria al publicar los números exactos del coste energético e hídrico de cada consulta a ChatGPT, forzando una conversación que muchos preferían aplazar.
- Con 2.500 millones de peticiones diarias, lo que parece un consumo trivial por usuario se convierte en una demanda masiva de energía y agua que amenaza con superar la capacidad de las infraestructuras existentes.
- Google ya explora la posibilidad de trasladar centros de datos al espacio, una señal de que las soluciones terrestres convencionales están alcanzando sus límites físicos y ambientales.
- La sostenibilidad ha dejado de ser una nota al pie en los planes estratégicos de las grandes tecnológicas para convertirse en el cuello de botella que podría frenar toda la expansión de la IA.
- La industria no tiene aún una respuesta clara: la carrera por la eficiencia energética y las fuentes alternativas de energía se ha vuelto tan urgente como la carrera por la inteligencia misma.
Sam Altman ha sido inusualmente transparente sobre uno de los problemas que más le preocupan: la cantidad de electricidad y agua que necesitan los sistemas de inteligencia artificial que su empresa y sus competidores construyen a toda velocidad. En un artículo publicado en junio, el fundador de OpenAI desglosó las cifras con una claridad deliberada. Cada consulta a ChatGPT consume 0,34 vatios-hora de energía —el equivalente aproximado a lo que un horno usa en poco más de un segundo— y unos 0,00032 litros de agua, algo así como un quinceavo de cucharadita.
Tomados de forma aislada, esos números parecen irrelevantes. Pero Altman los publicó precisamente para obligar a pensar en escala: ChatGPT recibe alrededor de 2.500 millones de consultas cada día. Cuando se multiplica ese consumo individual por miles de millones de peticiones, lo trivial se convierte en una crisis real de infraestructura energética y en una preocupación ambiental legítima.
El problema no es exclusivo de OpenAI. Google, Meta y el resto de la industria enfrentan el mismo muro: centros de datos que consumen energía a tasas sin precedentes y requieren enormes cantidades de agua para refrigerarse. La competencia por desarrollar mejores modelos de IA se ha convertido también en una competencia por asegurar la infraestructura física que los sostenga. Google ha llegado a considerar trasladar parte de esa infraestructura al espacio, lo que ilustra hasta qué punto las opciones terrestres están quedándose cortas.
La disposición de Altman a cuantificar y hacer públicos estos datos refleja un cambio más profundo en el discurso del sector. Durante meses, la conversación pública giró en torno a las capacidades de la IA. Pero en paralelo, ingenieros y directivos han tenido que enfrentarse a una pregunta más difícil: ¿es posible construir la infraestructura que estos sistemas exigen sin chocar con límites físicos y ambientales que obliguen a frenar? La respuesta, por ahora, no está clara.
Sam Altman has been unusually direct about a problem that keeps him awake at night: the staggering amount of electricity and water required to run the artificial intelligence systems his company and its competitors are racing to build. In a blog post from June, the OpenAI founder broke down the math in terms anyone could understand. A single query to ChatGPT—one person asking the chatbot a question—consumes 0.34 watt-hours of energy. To make that tangible, he offered a comparison: it's roughly equivalent to what an oven uses in a little over one second, or what an efficient lightbulb draws in a couple of minutes. The water footprint is smaller but no less real: each query requires about 0.00032 liters of water, which Altman described as roughly one-fifteenth of a teaspoon.
These numbers sound almost negligible in isolation. A teaspoon of water, a second of oven time—the kind of thing you wouldn't think twice about in daily life. But Altman's point was to force a reckoning with scale. ChatGPT receives approximately 2.5 billion queries every single day. Multiply that individual consumption across billions of requests, and the picture shifts dramatically. The energy demand becomes a genuine infrastructure crisis. The water consumption becomes a genuine environmental concern.
This challenge is not unique to OpenAI. Every major technology company racing to develop and deploy advanced AI systems faces the same wall. Google, Meta, and the rest of the industry are all grappling with data centers that consume power at unprecedented rates and require enormous quantities of water for cooling. The competition to build better AI models has become a competition to secure the physical infrastructure to support them—and that infrastructure is hitting real limits.
The scale of the problem has become so acute that some companies are exploring radical solutions. Google has begun considering whether to relocate data center infrastructure to space itself, a sign of how constrained terrestrial options have become. The company is not alone in recognizing that business as usual will not work. Energy consumption and water usage have become central strategic questions for the entire industry, not afterthoughts to be solved later.
Altman's willingness to quantify and publicize these numbers reflects a broader shift in how the AI industry talks about its own sustainability. For months, the public conversation has centered on what these systems can do—their capabilities, their speed, their intelligence. But underneath that conversation, engineers and executives have been forced to confront a harder question: can we actually build the infrastructure these systems require without running into physical and environmental constraints that will force us to stop? The answer is not yet clear, and the industry knows it.
Notable Quotes
A single query consumes about 0.34 watt-hours of energy, roughly equivalent to what an oven uses in a little over one second or an efficient lightbulb in a couple of minutes— Sam Altman, OpenAI founder
Each query also requires about 0.00032 liters of water for cooling, roughly one-fifteenth of a teaspoon— Sam Altman, OpenAI founder
The Hearth Conversation Another angle on the story
Why did Altman choose to publish these numbers so publicly? It seems like the kind of thing a company might want to keep quiet.
Because the numbers were already becoming a problem he couldn't hide. Users were asking, regulators were asking, and the scale of the operation made it obvious something had to give. Better to frame it himself than have it framed for him.
But 2.5 billion queries a day—that's an enormous number. Does that surprise you?
Not really. ChatGPT is the fastest-adopted technology in history. What surprises me is that the infrastructure held up this long. The real surprise is what comes next, when demand keeps growing and the energy grid can't keep pace.
So this is a hard ceiling? We can't just build more data centers?
You can build more data centers, but you need power to run them and water to cool them. Both are finite. Google looking at space infrastructure isn't a joke—it's an admission that Earth-based solutions might not scale fast enough.
What does this mean for the average person using ChatGPT?
Right now, nothing. But if energy becomes the limiting factor for AI development, it changes which companies can afford to build these systems and how fast they can improve them. It becomes a question of resources, not just talent.