Making things more efficient doesn't save resources. It makes us use more of them. This isn't pessimism dressed as economics—it's a pattern so consistent that it's been running the global economy since before the light bulb was invented.
Most people assume efficiency is a form of conservation. Cut the fuel consumption of cars in half, and you've cut total fuel consumption in half. Make AI ten times cheaper, and companies use less computational power overall. It feels intuitive. It should work. Instead, the opposite happens almost every time. When a technology becomes more efficient—and therefore cheaper—demand explodes. We don't save the resources we've freed up. We spend them on new things we previously couldn't afford. The efficiency gain vanishes into the economy like water into sand.
William Stanley Jevons noticed this in 1865, watching what happened after James Watt's steam engine became dramatically more efficient. According to the Economics Help analysis of Jevons's original observations, when the steam engine became cheaper to operate, coal consumption didn't fall. It tripled. The efficiency didn't conserve coal—it democratized access to steam power, triggering an industrial explosion that burned more fuel than ever before. Jevons realized then what economists still grapple with: efficiency and conservation are not the same thing. One is a technical achievement; the other is a choice about what to do with the money you've saved.
This dynamic is repeating itself with uncanny precision in 2025. According to Forbes analysis of the AI economy, as large language models become exponentially cheaper to run, usage explodes rather than plateaus. The cost of generating intelligence drops, which should theoretically mean we use less computational power. Instead, companies deploy AI for tasks they'd never have considered affording before. Automating customer service was a luxury; now it's standard. Generating variations of marketing copy, summarizing documents, personalizing recommendations at scale—these all become economically viable only once the price per unit of intelligence drops. The rebound effect is so powerful that efficiency gains in computation are barely visible in total energy consumption figures. We're running more AI, on more servers, consuming more power, despite each individual query becoming measurably cheaper.
Why does this pattern hold so stubbornly? The mechanism is straightforward. When something becomes cheaper, two things happen: existing users do more of it, and new users who couldn't afford it before enter the market. A 50% drop in AI costs doesn't mean companies cut their AI usage in half to pocket savings. It means they expand their AI use by an order of magnitude, because entirely new applications suddenly pencil out financially. A manufacturer couldn't justify AI-powered quality control at $1,000 per hour of processing. At $10 per hour, it's obvious. The efficiency unleashes latent demand—demand that existed but was economically suppressed until the price fell far enough. This is the Jevons Paradox, and it works because human behavior responds to price signals more powerfully than efficiency improvements change the underlying cost structure.
The implications are unsettling for anyone betting on efficiency as an environmental strategy. Making something twice as efficient doesn't halve consumption—it might double or triple it. But there's a flip side worth considering: efficiency still matters, because the baseline of consumption would be higher without it. We just can't assume that efficiency alone solves resource constraints. It buys time, shifts demand, redistributes access—but it doesn't make scarcity disappear. The real question isn't whether to make things more efficient. It's what we do with the breathing room efficiency creates.