How AI and automation are changing energy and compute

Every transformation on the other five screens runs on this one. Healthcare, finance, work, art, blockchains: each of them is, underneath, a question of how much computation you can afford, and computation is energy with extra steps. If energy gets cheap, everything above it gets cheap. If it does not, nothing does.

I used to think of this as an abstract argument. Then I started running automations that call models thousands of times a day and watched the bill. Compute is now a line item for a small company the way rent is. That changes how you think about the grid.

AI is the demand

Training and running models is the fastest-growing new load on the electrical system, and the people building it have noticed that the constraint is not chips. It is power, cooling, and the years it takes to connect a site to the grid. The result is something I did not expect: the software industry has become the most motivated energy investor on the planet. Nuclear restarts, geothermal, solar at scales that used to be science fiction, batteries by the gigawatt-hour. The demand is pulling the supply into existence.

AI is also the supply

The grid is the largest machine humans have built, and it is run with surprisingly little automation. Models are changing that at every layer.

• Forecasting. Knowing an hour ahead how much the wind will blow and how much a city will draw lets you run cheaper plants more and expensive ones less. This is the same skill as predicting which support tickets will come in Monday morning, applied to electrons.
• Balancing. Batteries, thermostats, car chargers, and factories can all shift when they draw power. Coordinating millions of them is a problem only software can solve, and it turns a demand problem into a storage solution.
• Finding and building. Models now help locate geothermal resources, design fusion experiments, discover battery chemistries, and shorten the paperwork that keeps a transmission line in review for a decade. The paperwork alone is a bigger lever than most people realize.

Compute gets cheaper at the same time

The other half of the equation is how much work a unit of energy buys. Chips get more efficient. Models get smaller for the same answer. Scheduling moves heavy jobs to the hours and places where power is cheapest. I have seen the cost of a given automation fall by an order of magnitude in a couple of years without my changing a line. That trend is the quiet engine behind everything else on this site.

Why it matters beyond the data center

Cheap energy means cheap water, because desalination is energy. Cheap fertilizer, cheap steel, cheap cooling in places that are getting hotter. It means a family's electricity bill stops being a tax on everything else they do. When I say I am trying to rebuild the world with automation, this is the foundation. Get this right and the rest is engineering.