Energy, upside‑down.

Stand outside at noon and you're standing under a firehose. The sun delivers more than 20,000 times the energy humanity uses, every day, for free. Capturing it used to be expensive. It isn't anymore — solar is now the cheapest electricity most of the world has ever had access to.

So here's the strange part: almost no one believes solar will become the grid's primary source. We've made an abundant resource look like a difficult one. This is the first dispatch from Off Grid Labs, and it's about how that happened — and what we're doing about it.

The grid was built to take orders

The traditional grid runs on a simple promise: whenever you flip a switch, generation rises to meet you. Demand leads; supply follows. That works beautifully when your generators are a stack of fuel you can burn faster on command.

The sun doesn't take orders. It shows up in the morning, peaks at midday, and leaves at night — on its own schedule, not yours. So utilities try to bend it back into the old shape: they balance the grid by ramping their least-efficient plants up and down around the solar, and by bolting on expensive batteries to time-shift the surplus. The more wind and solar come online, the more ramping and the more batteries they need. We're paying — at enormous scale — to make an abundant resource behave like a scarce one.

Flip the model

There's a different move, and it's almost embarrassingly simple. Instead of forcing generation to follow demand, follow generation with demand. Use energy when the sun is making it. Store the rest in the cheapest form you can, and lean on it when the sun is down.

The objection is obvious: you can't tell people to only use power when the sun is out. But you don't have to — not if the load you're moving doesn't care what time it is. And the largest new load on the planet is exactly that kind of work.

That work is compute — AI inference and training. A model doesn't care whether it answers at noon or midnight; a training run is happy to speed up when the sun is high and coast when it dips. Compute is the most movable large load we've ever built. The trick is putting it where the power already is.

Put the compute where the sun is

So that's what we build: off-grid AI nodes — solar, storage, and accelerated compute in one box, dropped onto cheap land with good sun. Panels feed the GPUs directly while the sun is high; a battery bank carries the load into the night. We schedule inference and training to soak up the cheapest hours of daylight, the way a smart factory runs its heaviest machines when power is cheapest.

It's the flexible load the grid always wanted, finally pointed the right way. No ramping a fossil plant to babysit the solar. No fighting for a substation upgrade. No metered cloud bill that compounds forever — just sunlight in, tokens out.

And because the node lives behind its own panels, it's resilient by default. When the grid or the internet goes down, the sun is still up. Your models keep answering and your data never leaves the box — quietly, the way infrastructure should.

Builders, not landlords

The energy transition doesn't only need bigger batteries and smarter market models to keep the old system balanced. It needs people willing to stop fighting the resource and start designing around it. That's the whole thesis: the sun is abundant and cheap, the most flexible load we have is compute, and compute can run wherever the power lands — on hardware you own instead of rent. Put those three facts together and the answer flips upside-down: you can own your AI.

This is dispatch one. We'll keep posting from the field — what we're building, what's working, and what the sun taught us that week. If you want the next one, leave your email below.