Artificial intelligence has survived just about every obstacle thrown at it. Not enough advanced chips? Nvidia (NVDA) and its competitors are racing to build more. Not enough data centers? The world’s largest technology companies are spending hundreds of billions of dollars to build them. Money’s never been an issue.

But AI is beginning to run into an obstacle that can’t be solved simply by writing another check: There may not be enough electricity.

And we’re starting to see what that looks like in the real world. Britain’s planned flagship AI supercomputer was supposed to come online in 2027, giving researchers access to some of the world’s most advanced computing power and playing a central role in the country’s AI ambitions.

There’s just one problem. The electric grid can’t give it the 90 megawatts of power it needs. The project is now facing delays, potentially into the 2030s, because the grid doesn’t have enough capacity. And it isn’t alone. There are reportedly 315 data centers waiting for grid connections in Britain.

That’s remarkable when you think about it. We spend a lot of time talking about the race for AI chips, models, and talent. But none of those things matter if you can’t plug the computers into the wall.

A $9 Trillion Reality Check

That’s why another story this week caught my attention. According to Barron’s, developers have proposed an astonishing 509 gigawatts of new U.S. data-center capacity. That’s roughly nine times the capacity currently operating, with a theoretical price tag approaching $9 trillion.

Here’s the catch. A new Columbia Business School analysis estimates that 45% of those proposed projects could ultimately be abandoned, while another 25% could be delayed until 2033.

At first glance, that sounds like bad news for the AI boom. I don’t see it that way.

The problem isn’t that companies suddenly don’t want AI. It’s that building an AI data center requires a lot more than ordering thousands of Nvidia GPUs. You need land, transformers, substations, transmission lines, and cooling systems. You may need gas turbines, batteries, or even an entirely new power plant.

And most importantly, you need electricity. Lots of it.

The International Energy Agency (IEA) expects global data-center electricity consumption to roughly double from about 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030. Electricity consumption from AI-focused data centers alone is expected to more than triple.

Here in the United States, the IEA expects data centers to account for nearly half of electricity-demand growth through 2030.

We’re essentially adding an entirely new source of electricity demand to a grid that wasn’t built for it.

Too Fast for the Grid

This is why I believe the AI investment story is entering another phase. The first phase was about chips. Nvidia became the poster child because every AI company needed GPUs.

But as more chips get installed, the bottleneck is moving. Now companies need the physical infrastructure required to actually run them.

We’re already seeing some data-center developers get tired of waiting for grid connections.

Reuters reported this week that developers are increasingly looking at installing their own small gas turbines behind the meter. Enverus estimates nearly 30 gigawatts of this behind-the-meter gas generation could be added in the United States by 2030, with data centers accounting for 88% of it.

That’s not happening because the AI boom is slowing down. It’s happening because it’s moving so quickly that our existing infrastructure can’t keep up.

And that creates another enormous investment opportunity.

Selling Shovels in a Power Rush

McKinsey estimates the global data-center buildout could require nearly $7 trillion of investment through 2030. Not all of those projects will happen. But even if the Barron’s analysis is correct and a large percentage are delayed or canceled, we’re still talking about trillions of dollars flowing into physical infrastructure.

That’s why I’m increasingly interested in the companies one layer beneath the AI boom. GE Vernova (GEV) sells gas turbines, transformers, and grid equipment. Its Electrification equipment backlog has already climbed above $40 billion, and data-center orders are surging.

Then there’s Eaton (ETN), which makes the electrical equipment needed to manage and distribute all that power. Vertiv (VRT) provides power-management and cooling systems inside data centers. And Quanta Services (PWR) helps build the transmission and electrical infrastructure needed to connect everything together.

These companies don’t need to predict which AI model wins. They just need AI to keep consuming more electricity. That’s the kind of picks-and-shovels opportunity I love because the bottleneck itself creates demand.

For years, investors have asked whether we have enough chips to power the AI revolution. As trillions of dollars pour into data centers, I think we’re approaching a much bigger question:

Do we have enough power to turn them all on?