You’re probably reading this on a device that’s drawing more power than your brain uses to comprehend it. We are terrified of AI taking over the world, yet we’ve completely ignored the absurd, catastrophic price tag of building it.
We are trying to build a supercomputer to replicate what a sandwich can power.
Your brain—the most sophisticated reasoning engine on the planet—operates on about 20 watts of power. That’s less than a dim lightbulb. You can solve complex spatial puzzles, feel empathy, and write code on the energy it takes to run a small desk fan. Meanwhile, training a single large language model consumes the electricity of a small nation.
We think we’re innovating, but we’re repeating a familiar, arrogant mistake. Look at lab-grown meat. Tech billionaires poured billions into replicating the cow in a lab, only to realize the cow is a perfectly engineered, self-replicating, solar-powered meat factory. We sacrifice natural efficiency for artificial control because we want to own the patent, not the process.
Our obsession with artificial intelligence isn’t about surpassing nature; it’s about monetizing it, no matter the environmental cost.
We need to stop celebrating the brute-force scaling of AI models. It’s not brilliant; it’s dangerous. We are treating compute power as a substitute for actual intelligence. Throwing megawatts at a server farm to generate a mediocre essay isn’t a technological triumph—it’s an indictment of our architectural laziness.
If we don’t course-correct, our pursuit of artificial minds will leave us in the dark. We will have built a machine that can write a poem about the forest, while burning down the forest to keep the server running.
True innovation isn’t replacing a 20-watt miracle with a 20-megawatt monstrosity. It’s learning how to work with the efficiency we were already given.
FAQ
Q: Isn't AI just getting more efficient every year?
A: Hardware efficiency is improving, but the sheer scale of model parameter growth is outpacing those gains. We're making the engine bigger, not necessarily smarter per watt.
Q: Does this mean we should stop building AI?
A: No, it means we need to pivot from brute-force scaling to architectural efficiency, looking closer at neuromorphic computing rather than just throwing more data centers at the problem.
Q: Is the AI boom just an energy bubble?
A: Yes. We are subsidizing massive energy consumption under the banner of 'innovation,' ignoring that nature already solved these compute problems millions of years ago.