For most of the AI boom, investors have been obsessed with computing power…

Who makes the fastest chips? Who builds the biggest data centers? Who supplies the networking equipment connecting millions of processors together?

But an increasingly serious problem is emerging further down the infrastructure stack.

You can buy all the Nvidia chips you want and you can build a billion-dollar data center.

But none of it matters if you can’t turn the lights on.

And across America, that’s becoming a very real problem.

At least 75 U.S. data-center projects worth roughly $130 billion encountered local opposition during the first quarter of 2026 alone, according to Reuters. 

Concerns include electricity consumption, higher utility bills, water usage, noise, and the strain enormous computing facilities can place on local infrastructure.

And the backlash has become serious enough that some projects are being delayed, reconsidered, or stopped outright.

New York recently became the first state to impose a one-year moratorium on construction of large new data centers consuming 50 megawatts or more.

Texas — arguably America’s hottest data-center market — has temporarily halted approvals for new projects seeking connections to its power grid while state officials conduct an audit.

And the numbers help explain why…

Roughly 90% of the astonishing 474 gigawatts of proposed new electricity demand being reviewed in Texas comes from data centers. 

That’s more than five times the state’s current peak electrical load.

And it tells you that AI doesn’t just have a power problem…

It has a grid problem.

More Electricity Than the Grid Can Deliver

Data-center demand isn’t slowing down…

Goldman Sachs estimates electricity consumption from U.S. data centers could rise from approximately 31 gigawatts in 2025 to 66 gigawatts in 2027.

Wood Mackenzie sees U.S. data-center capacity reaching roughly 110 gigawatts by 2030, compared with approximately 24 gigawatts today.

That’s an extraordinary amount of new electricity demand arriving very quickly. And the problem is that America’s electrical infrastructure wasn’t designed for it…

Building generating capacity takes time. 

So does constructing substations, transmission lines, transformers, and all the other infrastructure necessary to move electricity from where it’s generated to where it’s consumed.

So, in many regions, developers can secure the land, financing, equipment, and customers for a data center long before they can secure the electricity necessary to operate it.

And that’s why a new phrase has started becoming increasingly important in the AI infrastructure business:

Time to power.

Investors are beginning to prioritize energy projects based not simply on what electricity costs, but how quickly that electricity can actually reach a new data center.

And that’s where things get interesting. Because perhaps the solution isn’t endlessly expanding the grid to accommodate data centers…

Perhaps data centers can start bringing their own fully-powered grid.

Bring Your Own Power Plant

That’s already where the industry appears to be heading…

PJM Interconnection, which operates America’s largest electrical grid, has considered requiring massive electricity consumers — primarily data centers — to either develop their own power supplies or accept arrangements allowing their consumption to be curtailed during periods of grid stress.

Think about what that means…

For decades, the basic infrastructure model was straightforward:

Build a factory, office building, hospital, or data center and plug it into the electrical grid.

But AI data centers consume electricity on an entirely different scale. That scale could eventually turn energy generation into part of the data center itself…

And one technology may be particularly well suited to doing it.

The Nuclear Battery

You’ve probably heard about small modular reactors, or SMRs. 

But there’s another category of advanced nuclear technology that’s considerably smaller, but potentially more lucrative…

They’re called microreactors.

The Department of Energy describes them as compact nuclear reactors small enough to potentially be transported by truck. 

And they’re being developed to supply reliable electricity to locations ranging from remote communities to military bases.

Now we can add AI data centers to that list…

You see, instead of building an enormous conventional nuclear power station, developers envision compact reactors that could be manufactured in factories and delivered where the electricity is needed.

So, a data-center campus could potentially operate several.

Need more computing capacity? Add more servers.

Need more electricity? Add more reactors.

And the result starts looking less like the traditional relationship between a power plant and an electricity customer and more like a giant nuclear-powered battery sitting beside the data center.

Except unlike traditional batteries, nuclear reactors can continuously generate power day and night.

There’s no waiting for the sun. There’s no dependence on wind.

And there’s far less dependence on an already overcrowded electrical grid.

That’s why the Department of Energy is taking the concept seriously.

So seriously that it’s already provided support for projects exploring microreactors and sites capable of hosting nuclear-powered data centers.

And that brings us to one particularly interesting design…

Meet Morpheus

A company called Nuclea Energy is developing a microreactor called Morpheus.

And one of the markets Nuclea specifically identifies for the technology is AI and data centers.

Morpheus is designed around a compact lead-cooled nuclear architecture intended to operate in locations where conventional energy infrastructure can be difficult, expensive, or impossible to build.

That original mission included remote communities, mining operations, military installations, and extreme environments.

But suddenly the same characteristics look remarkably well suited to America’s AI infrastructure problem…

A data-center operator doesn’t necessarily need another gigantic regional power plant.

It needs reliable electricity right beside the computers.

Microreactors offer the possibility of distributed nuclear generation capable of operating continuously with a very small physical footprint.

They’re also quiet, low-carbon, and capable of operating for extended periods without the constant fuel deliveries associated with diesel or natural-gas backup generation.

In other words, something resembling a long-duration nuclear battery.

Morpheus isn’t commercially deployed yet, and that’s an important distinction. 

Advanced reactors still face licensing, manufacturing, financing, deployment, and first-of-a-kind cost challenges. 

And even the Department of Energy warns that new nuclear designs will take time to deploy and early reactors can be expensive.

But those hurdles don’t diminish the problem they’re attempting to solve. They simply help explain why solving it could become so valuable.

The Picks-and-Shovels Opportunity Is Changing

For the past several years, investors have made fortunes betting on the infrastructure behind AI…

Semiconductors, servers, networking equipment, cooling equipment, data centers, utilities, transformers, and electrical contractors have all benefited.

But infrastructure bottlenecks evolve.

First, we didn’t have enough chips. Then we didn’t have enough data centers…

Now, increasingly we don’t have enough accessible power.

And that’s why the next stage of the AI infrastructure boom may produce a new class of winners…

Companies capable of allowing data centers to generate electricity themselves.

That’s bigger than simply selling another megawatt of power because it could remove one of the biggest constraints preventing billions of dollars of computing infrastructure from being built.

Remember those 75 projects worth $130 billion that encountered resistance in just three months?

Every delayed data center makes a solution that can reduce dependence on local power infrastructure a little more valuable.

Every overloaded transmission line makes distributed generation more attractive.

Every community worried that an enormous AI facility will raise residents’ electricity bills strengthens the argument for making that facility supply more of its own power.

And every additional AI model requiring another warehouse filled with processors increases the urgency.

The AI revolution still needs chips. It still needs servers. It still needs data centers…

But increasingly, the most valuable piece of equipment on tomorrow’s AI campus might not be sitting inside the data center at all.

It might be the little nuclear reactor sitting outside it.