Powers that be – Who controls the switch?
We have spent the last two years asking one question – How much electricity does AI need? Gigawatts, datacenters, and rising utility bills have dominated the headlines. It is a fair question. But what if it is the wrong one?
What if we should be asking: Who will control electricity?
At its heart, the grid is a giant real-time balancing act. Supply must match demand every second, across millions of generators, batteries, buildings and now EVs. No human team can do that at scale. Software already does a good part of it, and AI is about to do the rest.
In my earlier post on AI’s energy paradox, I described how AI agents could decide whether your rooftop solar surplus is used, stored, sold or held. Now zoom out. The same decision layer, scaled to a city or a country, ends up answering questions that were once left to regulators and utilities:
- Which neighborhood gets power first when the grid is stressed?
- Should the datacenter, the cold-storage warehouse or the EV fleet ramp down first?
- Whose battery gets drained to keep everyone else’s lights on?
- What is a “fair” price at 7 pm on the hottest day of the year?
Notice something? None of these are technical questions. They are questions of value, dressed up as math.
AI does not decide in a vacuum. It optimizes whatever it is told to optimize (based on the training data and rules). Lowest cost? Highest profit for the operator? Lowest carbon? Maximum reliability? Each objective produces a different grid, with different winners and losers. Whoever writes the objective function quietly writes the policy. And unlike legislation, nobody gets to vote on it.
I find it ironic. Policy makers are racing to put controls on datacenter appetite, while the far bigger regulatory question sits unnoticed in the software between the generator and your socket.
None of this is a doomsday scenario. Smarter coordination can mean less curtailment, lower bills and a cleaner grid. That is the upside, and it is real. But it comes with conditions. Transparency about what the system is optimizing for. Who can override it? Who is accountable when it gets it wrong?
Like ethics, governance cannot be bolted on later, especially when the thing being governed decides who gets light and who gets a blackout.AI’s energy story so far has been about appetite. The next chapter is about authority.
The question was never just how much power AI will use. It is who, or what, decides where the power goes.