We are living in one of those moments in time.
En el escenario más grande de la industria tecnológica, el CEO de Google admitió en voz baja lo que ninguna diapositiva de presentación se atrevió a decir: que incluso la empresa que más invierte en infraestructura de inteligencia artificial en el mundo no puede satisfacer la demanda que ella misma ha despertado. Sundar Pichai, hablando con periodistas tras el Google I/O 2026, confirmó con una sola palabra —'absolutamente'— que la era del entusiasmo sin límites ha encontrado su frontera. Lo que emerge no es una crisis de visión, sino una colisión entre la ambición tecnológica y la física del hardware: los costos suben, la capacidad no alcanza, y el mundo quiere más.
- Google gasta entre 180 y 190 mil millones de dólares anuales en infraestructura, y aun así su propio CEO reconoce que la demanda supera lo que pueden ofrecer.
- Los precios del hardware —procesadores, memoria, componentes de centros de datos— siguen subiendo, lo que significa que el mismo presupuesto compra cada vez menos capacidad de cómputo.
- Directores de tecnología en grandes empresas ya han agotado sus presupuestos anuales de IA con meses por delante, y Pichai advierte que la presión solo aumentará.
- La respuesta de Google es estratégica y pragmática: promover Gemini 3.5 Flash, un modelo más barato y eficiente, como la solución principal para la mayoría de los casos de uso empresarial.
- El mensaje implícito es profundo: la industria entra en una era de 'cómputo restringido', donde la eficiencia reemplaza a la potencia como valor central.
El Google I/O 2026 cerró con la pompa habitual: cifras récord de inversión, centros de datos en expansión, el futuro prometido desde el escenario. Pero horas después, en una entrevista con periodistas, Sundar Pichai dijo lo que ninguna presentación había dicho. Cuando le preguntaron si Google enfrentaba más demanda de inteligencia artificial de la que podía atender, respondió con una sola palabra: "Absolutamente."
Esa admisión tiene un peso particular viniendo del líder de la empresa que más invierte en infraestructura de IA en el planeta. El problema es estructural: los costos del hardware siguen subiendo, y con el mismo presupuesto Google obtiene menos capacidad que hace apenas unos meses. Al mismo tiempo, los clientes que ven una demostración de Gemini 3.5 Flash corriendo a 800 tokens por segundo exigen acceso inmediato. Pichai describió conversaciones con directores de tecnología que ya han agotado sus presupuestos anuales de IA con meses por delante, y advirtió que la situación empeorará.
En ese contexto, el lanzamiento de Gemini 3.5 Flash adquirió un nuevo significado. Pichai lo presentó como la respuesta práctica a este momento: un modelo rápido, con capacidades de frontera, y —en sus palabras— "extraordinariamente eficiente en costo". Su recomendación a las empresas fue directa: usen una combinación. Reserven los modelos costosos para lo que realmente los requiere, y muevan el grueso del volumen a Flash. Así es como se opera en la era del cómputo restringido.
La lógica es coherente: si la capacidad es escasa, guiar a los clientes hacia modelos más ligeros libera recursos para los casos que verdaderamente los necesitan. Google resuelve dos problemas a la vez —la factura del cliente baja, y la empresa distribuye mejor un recurso limitado. Pichai mencionó también otros cuellos de botella: permisos para construir centros de datos, disponibilidad de energía, componentes específicos en escasez. El retrato que emergió de la entrevista fue más honesto que cualquier keynote: la era del crecimiento sin restricciones ha encontrado un muro. Y la respuesta de Google no es construir más rápido. Es enseñarle al cliente a usar menos.
The morning keynote at Google I/O 2026 had closed with the usual pageantry—slides about record infrastructure investment, the company's $180 to $190 billion annual spending on data centers, the relentless march forward. But hours later, sitting across from journalists in a live interview, Sundar Pichai said something the stage presentation had carefully avoided. When asked whether Google now faced more demand for artificial intelligence than it could actually supply, the CEO answered with a single word: "Absolutely."
That word carries weight because it comes from the leader of the company that spends more on AI infrastructure than any other on Earth, and it amounts to a public acknowledgment that even that spending is not enough. The math behind the admission is straightforward and grim. Hardware costs are rising. Memory prices, processor costs, the price of every component that goes into a data center—all moving upward. With the same budget, Google gets less computing power than it did months ago. Meanwhile, customers who watch a demonstration of Gemini 3.5 Flash running at 800 tokens per second demand immediate access. That pressure accumulates on a company that must also feed its own products serving billions of users worldwide.
Pichai framed it as a difficult balance rather than a failure, but the substance is unmistakable. He described conversations with chief technology officers across major companies who have already exhausted their annual AI budgets with months still remaining in the year. "And I think the problem is going to get worse as we move through the year," he said. This was not a May anecdote. It was a trend he expects to deepen—a signal that the constraint is structural, not temporary.
The solution Google had already unveiled that morning took on new meaning in this context. Gemini 3.5 Flash, Pichai explained, exists precisely for this moment. It is fast. It delivers frontier capability. And it is, in his words, "extraordinarily efficient in cost" compared to models of similar power. His recommendation to enterprises was direct: use a mixture. Reserve the expensive models for work that genuinely requires them. Move the bulk of your volume to Flash. This is how you operate in an era of what he called "constrained computing."
The logic binds the two halves of the interview together. If Google has more demand than supply, pushing every customer toward the most expensive model no longer makes business sense—each query to a large model consumes scarce capacity the company cannot afford to waste. The rational move is to guide customers toward a lighter model that handles much of the work at a fraction of the cost and energy. This frees the expensive capacity for cases that truly need it. Google solves two problems at once: the customer's bill shrinks, and the company distributes a scarce resource more efficiently. Pichai noted that Google itself uses a combination of Pro and Flash internally. Most companies, he suggested, should learn to do the same.
Pichai touched on other constraints too. He spoke of Chinese open-source models and said his concern is not whether American companies adopt Chinese technology, but whether the United States does enough to stay at the frontier. He acknowledged systemic bottlenecks across every layer of infrastructure—permits to build data centers, available power, specific components in short supply. The picture that emerged from the interview was sharper than any stage presentation could offer. The era of unbounded enthusiasm for AI, the era of models that grow without restraint and investment that seems infinite, has hit a wall. That wall is called computing capacity. And Google's answer is not to build faster, because it cannot. The answer is to sell a smaller model and teach the customer to use it well. The most revealing line Pichai offered came almost in passing: "We are living in one of those moments in time." In that sentence, he had defined the entire year ahead for the industry.
Citações Notáveis
Demand for AI services absolutely exceeds our available computing capacity— Sundar Pichai, CEO of Google
The problem is going to get worse as we move through the year— Sundar Pichai, describing enterprise AI budget constraints
A Conversa do Hearth Outra perspectiva sobre a história
When Pichai said "absolutely" to the question about demand exceeding capacity, was he admitting failure, or something else?
Neither, exactly. He was naming a constraint that every company at scale eventually hits. The difference is that most companies hide it. Pichai said it out loud, which changes how you read everything else he said that day.
Why would he do that? What's the advantage of admitting you can't meet demand?
Because the real story isn't about capacity. It's about what you do with the capacity you have. If you tell customers the truth—we're constrained, here's how to use what we have efficiently—you actually build trust. And you solve a business problem: cheaper models at scale generate more revenue than expensive models that sit idle because you're rationing them.
So Flash isn't just a product. It's a strategy for living within limits.
Exactly. It's Google saying: we've hit the wall, and instead of pretending we haven't, we're going to help you work within it. That's a different message than "we have unlimited capacity."
Does this mean the AI boom is over?
No. It means the boom is entering a new phase. The phase where you can't just throw compute at every problem. You have to think about efficiency, about which tasks actually need the expensive tools. That's maturation, not collapse.
And the companies that figure out that mixture first—the ones using Flash for routine work and Pro for complex tasks—they win?
They win on cost, yes. But they also win on speed and flexibility. They're not waiting in a queue for scarce expensive compute. That's a real competitive advantage.