Are We Overbuilding for AI?

AI Isn’t Just a Software Story

If you scroll through anything tech related right now, it feels like AI is everywhere.

New models are dropping constantly. Startups are raising huge rounds. Every company is trying to figure out how to integrate AI into what they’re doing.

But something that gets talked about way less is what actually powers all of this.

AI might feel like a software story, but underneath it is a massive physical system that has to exist for any of it to work.

And that is where things start to get interesting.

A Familiar Pattern

There is a tendency to treat AI as completely unprecedented, but in some ways it feels similar to earlier moments in tech history.

During the dot-com boom, there was a real shift happening. The internet was transformative. That part was not wrong.

What did go wrong was how capital responded.

There was a rush to build infrastructure in anticipation of demand. Fiber optic networks were laid across the country. Telecom capacity expanded quickly. A lot of it ended up being more than what was actually needed in the short term.

The technology still won in the long run. But the timing and scale of investment did not always match reality.

AI could be setting up for a similar dynamic.

The Rush to Build

Right now, there is a huge push to build out data centers. Large tech companies are spending billions to secure compute. Developers and investors are racing to acquire land, lock in power, and get projects online as fast as possible.

On the surface, it makes sense. Demand for AI compute is growing quickly and no one wants to be caught without capacity.

But this kind of environment also creates incentives to build ahead of confirmed demand. Not necessarily because the long-term vision is wrong, but because everyone is trying to get there first.

The Constraint No One Can Ignore

What makes this different from a typical tech cycle is how resource-intensive AI infrastructure actually is.

Data centers require significant and reliable power, large amounts of water for cooling, and physical land in locations that can support both.

Power is probably the biggest constraint. Many regions are already struggling to keep up with demand from new projects.

Water is becoming a bigger issue as well, especially in areas where supply is already limited. These are not abstract constraints. They are very real, and they are not easy to scale quickly.

Why Location Is Starting to Matter More

This is where things get more interesting from an investment perspective.

Location is not just a secondary consideration for data centers anymore. It is becoming one of the main variables that determines whether a project actually works.

Coastal areas tend to have better access to global network infrastructure, including submarine cables that handle a large portion of international data traffic.

They also often have more established energy infrastructure and, in some cases, better access to water for cooling.

At the same time, inland markets can offer cheaper land and fewer regulatory barriers, which is why a lot of development is happening there.

So there is a tradeoff. The question is whether current development patterns are fully pricing in those tradeoffs, or if speed is taking priority over long-term viability.

What Could Happen Next

This does not mean AI is a bubble or that the demand is not real.

It means there is a possibility that infrastructure gets built faster than it is needed, or in places that are not optimal long term.

If that happens, you could see underutilized data center capacity in certain regions, pressure on returns for projects that were underwritten on aggressive growth assumptions, and a shift toward locations that better align with energy, water, and connectivity constraints.

In other words, the winners might not just be the ones who build first, but the ones who build in the right places.

Final Thought

AI is going to keep growing. That part feels clear.

But the story is not just about models, startups, or software. It is also about land, power, water, and geography. And those are much harder to scale.

The next phase of AI may not be constrained by algorithms, but by infrastructure.