The Economics of Artificial Intelligence

We can complicate this. Or we can name it.

$$C_{\mathrm{AI}} = \alpha C_{\mathrm{Compute}} + \beta C_{\mathrm{Energy}}$$

where $C_{\mathrm{AI}}$ denotes the total cost of AI, $C_{\mathrm{Compute}}$ represents the cost of Compute, and $C_{\mathrm{Energy}}$ represents the cost of Energy with $\alpha + \beta = 1$.

Two stacks. One game.

On one side, the four horsemen — Alphabet, Microsoft, Meta, Amazon — are building for as much compute as you can imagine. Trillions deployed. Infinite scale as a moat.

On the other, the energy chain — Saudi Aramco, Rosatom, NextEra, Tesla, CATL — is stretching from capture to generation, storage to systems. From electrons and batteries to FLOPs.

“Eventually the cost of intelligence will become the cost of energy.”

Not quite. Markets pick winners by where scarcity lives.

Historically, technology has always earned the premium. And although energy is essential, compute will still capture the margin.

So, in reality $\alpha \gg \beta$.

Compute is energy’s favorite and forever customer. Energy is compute’s lifeline.

This extreme co-dependency isn’t a bug; it’s the business model.

If you run energy, this isn’t a threat. It’s an invitation: sell certainty at scale.

If you run compute, it’s a reminder: buy watts (and time) before you start falling behind.

Over time, the cost of intelligence will certainly trend down. But the value of control won’t.

More soon on what this means for the energy sector and the glide path of intelligence costs.