common sense is remarkably uncommon

How much common sense is left during a tech boom?

This post was prompted by the latest in a long line of news articles about how far tech companies are willing to go to pursue more AI. My immediate reaction: hold on, let’s have a quick sanity check!

When companies consider restarting old nuclear power plants (Amazon), plan to expand (Meta) or build new ones (Google), it is time to stop and think. (Never mind the risks of restarting nuclear plants that were shut down for good reason.)

According to the IEA, global electricity demand from data centers is expected to quadruple by 2030 to 946 terawatt-hours, 80% of that for AI alone. To put that it into perspective: that is the entire energy consumption of Japan, or Germany and France combined, or 1/30th of the world’s total energy demand.

This is the moment to ask: is AI really so valuable that we’re willing to dedicate this much energy to it?

Sure, AI already does some very useful things, but enough to justify this much power consumption? Let’s take a few steps back for some perspective:

• Yes, AI can help accelerate the development of viable fusion power and safer thorium nuclear reactors. But building nuclear plants takes a decade (sometimes two and therefore will not help us with the current energy transition).

• Yes, AI can help us to (even better) understand how badly the climate will change due to our energy hunger.

• Yes, AI can improve diagnostics in healthcare and lead to more efficient jet and electric engines.

• Yes, AI will radically reshape some jobs.

But at some point, we will have enough AI compute power, and newer generations of AI systems will use less energy, not more. That is the pattern we have seen with previous tech booms (think mobile networks or, further back, railways), and there’s no reason this time should be different.

Still, common sense tends to disappear during a gold rush. No one wants to be the one who underinvested.

So what will likely happen? A massive wave of overbuilding, where supply far exceeds demand. It’ll be funded by venture capital (eventually including ordinary people’s money) and when the bubble bursts, that capital will evaporate, along with the savings of well-meaning citizens.

The projected energy demand of AI and data centers is based on a simplistic extrapolation, assuming a hockey stick curve with no end in sight, a classic mistake. When will the curve flatten? At this point, that is anyone’s guess. But hopefully, it happens before we will be dedicating a significant portion of global energy just to power machines.

Otherwise, the bio-battery scenario from The Matrix may start looking less like fiction, and that can’t be the goal.

Lesson of the day: Use common sense, especially during a tech bubble.