Every AI token begins as an electron. But before the electron moves, a dollar does.
The interface hides this. You type into a small box and an answer appears seconds later. It feels like software: clean, weightless, infinitely replicable. Underneath is something closer to an industrial plant - power contracts, transformers, cooling loops, network fabrics, accelerators, depreciation schedules, debt covenants and teams trying to keep very expensive machines busy every second of the day.
This is the insight behind Jensen Huang’s “five-layer cake” of AI. From the bottom up, the layers are energy, chips, infrastructure, models and applications. Energy supplies the electrons. Chips convert them into computation. Infrastructure makes thousands of chips behave like one machine. Models turn computation into reusable capabilities. Applications convert those capabilities into economic value. Every successful application pulls demand through the layers below it, all the way to the power plant.
It is an elegant map of the AI economy. It is also incomplete.
The cake sits on a balance sheet.