AI is changing what land is worth

Power Play | Mini-Series Issue #3 | September 2026

Hello friend,

Four episodes into Power Play, the question I’m asking has changed.

At the beginning, I was focused on scale. How much electricity will AI need? How quickly can we build enough infrastructure to support it? And what happens if the forecasts are wrong?

Those questions still matter. But the deeper we get into the reporting, the more I find myself thinking about something else: before we decide what needs to be built, how do we decide what is already available to use?

That sounds like a technical question. But it turns out to be an economic, social, and sometimes surprisingly human one too.

Catch up on Power Play →

Episode 3 starts with a word I hadn’t given much thought to before: underutilized.

When companies are looking for places to build new infrastructure, rural land near transmission can be described as underutilized or empty. That language makes the opportunity sound obvious. Here is an asset that isn’t doing enough. Here is a way to put it to better use.

But “underused” depends entirely on who gets to define useful.

Land can already have owners, communities, history, agreements, cultural meaning, or an existing purpose that doesn’t show up neatly on a spreadsheet. Calling it empty can make a development decision sound neutral when it really involves a choice about whose definition of value takes priority.

That idea stuck with me because it travels well beyond land.

Companies do this all the time with unused office space, idle inventory, excess capacity, customer data, even people’s time. Sometimes we’ve genuinely found an inefficient asset. Other times, we’ve simply decided that the current use matters less than the new one we have in mind.

The lesson: Whenever something is described as underused, it’s worth asking: underused by whom?

Episode 4 takes the idea of “unused” in a completely different direction.

One of the findings I found most surprising is how little flexibility a large electricity user may need to change the infrastructure equation. The episode explores research suggesting that if a data centre could shift or reduce its demand for roughly 22 hours over the course of a year, that flexibility could help some new loads connect without requiring the same level of infrastructure built around the absolute peak.

Just 22 hours out of an entire year.

That’s a tiny window compared with the scale of the infrastructure decisions being made around peak demand.

We tend to think of infrastructure as physical: another power plant, another transmission line, another substation. But coordinating when demand happens can create capacity too. You haven’t built anything new. You’ve changed how the existing system behaves.

That opens up a very different set of solutions. Episode 4 gets into energy storage, virtual power plants, and other ways of finding or creating flexibility in infrastructure we already have. The premise is less “how fast can we build?” and more “how much more can the existing system do if we use it differently?”

The lesson: Sometimes the most valuable infrastructure isn’t something new you build. It’s the ability to use what you already have differently.

There’s another theme in Episode 4 that feels very familiar from the business world.

There are plenty of clever technical solutions to parts of the grid problem. But a technology being possible doesn’t mean it will be adopted.

It still has to fit into utility economics, regulation, market structures, data access, incentives, and the way companies actually make decisions. In other words, the technology may not be the hardest part.

That distinction matters.

We tend to talk about innovation as though the best technology naturally wins. In reality, a technically elegant solution can sit on the sidelines for years if the incentives, rules, or business model around it don’t support adoption.

The lesson: A technical solution only matters if the system around it allows it to work.

Four episodes in, I’m thinking less about whether America can simply build enough power for AI.

I’m thinking about the choices underneath that question.

What counts as unused? Who gets to decide? When is building more actually necessary, and when could we get more from what’s already there? If the technology exists, what stops us from using it?

The story is getting less tidy as we go. I think that’s probably a good sign.

Catch up on Power Play →