AI Has a Power Problem — But Who Actually Benefits?

There’s a growing narrative forming around AI and energy:

without electricity, there is no AI

That sounds obvious, but it’s becoming a real constraint.

The Setup

Data center electricity demand in the U.S. is expected to rise from ~25 GW to over 100 GW by 2035.

At the same time:

- grid expansion takes years - permitting is slow - infrastructure is already strained

This creates a gap between: - demand (AI compute) - and supply (power availability)

The Immediate Reaction

The market has started focusing on:

on-site power generation

This is where companies like BE come in.

The pitch is simple: - deploy power in ~55 days - bypass grid delays - support AI infrastructure directly

That’s a real advantage.

But This Might Be Just One Layer

The more interesting question is:

does one solution capture the value… or does the system split it?

If you think about data centers like an electrical system, there are multiple layers:

- generation (power supply) - distribution (grid + infrastructure) - load (AI compute demand)

On-site generation solves one piece: speed

But other parts of the system may still matter more over time: - large-scale generation (utilities, nuclear) - distribution infrastructure (grid upgrades, equipment) - efficiency at the compute layer

Where This Gets Interesting

If power really becomes the bottleneck for AI, then:

- it’s not just about who can generate power - it’s about who captures the most value across the system

Some companies may benefit from: - urgency (short-term demand spikes)

Others may benefit from: - scale (long-term infrastructure buildout)

The Takeaway

The narrative is probably right:

AI will drive a massive increase in power demand

But the conclusion might be too simple.

Instead of asking: “which stock benefits?”

It may be better to ask:

“which part of the system becomes most valuable as this constraint plays out?”

That’s a much harder question.

And a much more interesting one.