The master variable
For two centuries, every leap in human welfare rode on cheaper energy: the steam engine, the electric grid, the Haber-Bosch process that now feeds half the planet. Energy isn’t one input among many — it’s the one the others are made of. So when we hear that AI could double the world’s data-center draw, we don’t hear a crisis. We hear the first demand signal in a generation big enough to actually finance new supply.
The instinct to ration is the worst trade on the board. Throttling intelligence to save watts is like dimming the library lights to shave the electric bill — you save pennies and forfeit what the building is for. The answer to scarce power is more power: solar where it’s sunny, gas where it’s needed now, geothermal, and nuclear at scale. Build all of it, and build it fast.
The ratepayer advocates are right that the costs have to land fairly, and we’ll meet them there — a boom that raises an retiree’s power bill makes enemies we don’t need. But the grid realists underrate what a giant, bankable, around-the-clock customer can do. For the first time in decades someone wants to buy enough firm power to pay for building it, and that money builds plants that outlast the boom. You don’t reach the future by being frugal with electrons.
Where we concede ground: If our boom quietly raises everyone else’s power bill, we’ve earned the backlash and should pay to prevent it.
What would change our mind: A decade of this buildout with no measurable gain in productivity, discovery, or living standards.
Read the full synthesis: Should AI data centers get their own power plants?