I think energy price arbitrage through AI load shifting is going to be a real business.
Something I’ve been thinking about that I wanted to share with the team (and with anyone building software).
We all know AI is mostly compute, and compute is mostly energy. Here’s the part that’s less obvious:
Energy can’t move instantly. But a lot of AI workloads can.
Solar and wind produce power that’s hard to store. When they overproduce, the price drops - sometimes it goes negative. Utilities pay large consumers to take excess off the grid.
Now imagine you’re running code reviews, test pipelines, or refactoring tasks. None of these need to happen this second. They need to happen sometime today. What if you scheduled them for when energy is cheapest? What if your CI runs chased the sun - Frankfurt in the morning, Texas at noon, Chile in the evening?
Some of your overnight refactoring runs wouldn’t cost you anything. In fact, someone would be paying you to do them.
This isn’t science fiction. The datacenters are being built. The variable energy is already here. The piece that’s missing is the scheduler - the orchestration that says “this workload isn’t urgent, park it where the electrons are cheapest.