Artificial intelligence may live in the cloud, but the infrastructure supporting it is becoming impossible to ignore…

Every prompt, generated image, automated report, and AI-assisted discovery ultimately runs through a physical data center filled with processors that consume enormous amounts of electricity and generate tremendous heat.

And as AI models grow more capable, the campuses supporting them are becoming larger, more expensive, and considerably more demanding.

As I’m sure you’ve heard, communities are beginning to push back. And they’re pushing back hard enough to make a big impact…

More than $130 billion worth of proposed American data center projects were reportedly blocked or delayed during the first three months of 2026 alone.

Think about that…

Around the world, governments and municipalities have imposed moratoriums, rejected developments, or demanded stricter conditions before allowing hyperscale facilities to connect to local infrastructure.

The objections vary from one location to another…

Some residents worry about noise, disappearing farmland, changing landscapes, and the relatively small number of permanent jobs created after construction ends.

But two complaints consistently rise above the others….

Data centers consume electricity.

And that problem could determine whether the AI infrastructure boom continues accelerating or crashes into the limitations of the existing power grid.

The Physical Cost of Artificial Intelligence

For years, data centers were treated as relatively unobtrusive pieces of infrastructure.

They were windowless buildings that stored websites, processed financial transactions, and delivered streaming video.

But AI has changed the scale of the business…

Traditional computing workloads tend to rise and fall. AI training campuses can run thousands of specialized processors continuously.

And inference facilities must remain ready to respond to users at any hour, day or night.

That means a modern AI campus will require hundreds of megawatts of reliable electricity.

S&P Global Market Intelligence estimates that U.S. data center electricity demand could reach 82.3 gigawatts in 2026.

That’s this year! And that’s up 28% from 2025 and more than double the level of three years earlier.

That kind of demand can’t be hidden behind a fence. And a facility consuming as much electricity as a city quickly becomes a local political issue…

Residents understandably want to know whether their utility will need to build new substations, transmission lines, and power plants to serve it.

But more importantly, they want to know who will pay for all that infrastructure.

Because technology companies routinely promise to cover the direct cost of connecting their campuses. But electricity markets are complicated…

A new industrial customer can affect transmission congestion, capacity auctions, generation planning, and utility capital spending across an entire region.

And that creates a simple and politically powerful fear: The data center gets the electricity, while local families get the bill.

Research into the historical relationship between data centers and electricity rates is not unanimous on that conclusion….

One 2026 academic study found that data center growth may actually have lowered average retail rates between 2015 and 2024 by spreading fixed grid costs across a larger volume of electricity sales.

However, its authors also warned that future supply constraints could reverse that effect.

And that warning matters more than the historical evidence, because the grid is no longer operating under the conditions that existed a decade ago.

Transmission congestion is growing. New power plants and transmission lines can take years to approve.

Transformers, turbines, and other electrical equipment remain difficult to secure.

Meanwhile, data center developers increasingly want gigawatts of capacity on timelines measured in months rather than decades.

Whether data centers have raised rates in the past is almost beside the point. Communities are worried that the next wave will.

And perception alone is enough to stop a project dead in its tracks.

The Grid Was Not Built for This

The AI industry simply can’t keep plugging enormous new campuses into a grid that was designed for slower, more predictable growth.

Even where enough generating capacity exists in theory, the electricity may be located hundreds of miles from the proposed data center.

Moving it requires transmission infrastructure that can take a decade or longer to permit and construct. And grid congestion is already becoming expensive…

In the PJM market (that covers all or part of 13 states and the District of Columbia) transmission congestion cost about $1.8 billion during May and June 2026 alone.

And adding more intermittent generation like wind and solar can help produce electricity, but an AI data center requires power around the clock.

Batteries can smooth short interruptions and shift electricity between hours, but providing continuous backup for a multihundred-megawatt campus would require an enormous amount of storage.

Natural gas offers dependable generation and can be deployed faster than many alternatives, which is why turbines, fuel cells, and behind-the-meter gas plants are becoming part of the data center conversation.

But gas still produces emissions, requires pipeline access, and exposes operators to fuel-price volatility. Plus, the backlog on new gas turbines is now nearly a decade long.

The more durable solution here is to stop treating the grid as the data center’s only power source….

And realize that the power plant can become part of the campus itself.

Nuclear Power Without the Giant Nuclear Plant

Traditional nuclear reactors are exceptionally reliable once operating, but building one is a monumental undertaking.

Projects can require billions of dollars, vast construction sites, and decades of regulatory review.

But micro modular reactors are designed to change that equation…

Rather than constructing a custom-built gigawatt-scale plant, developers envision smaller reactors manufactured largely in factories, transported to their destination, and installed in standardized configurations.

Multiple units could be added as a campus grows, allowing operators to match electricity supply more closely with computing demand.

And that makes them particularly interesting for data centers…

A microreactor can theoretically provide steady power without relying on weather, long-distance transmission, or frequent fuel deliveries.

It can operate behind the meter, reducing the amount of electricity the campus has to pull from the public grid.

And because nuclear fuel contains an extraordinary amount of energy, a compact reactor can potentially run for years before refueling.

And one company pursuing this opportunity is Nuclea Energy…

Nuclea is developing the Morpheus Micro Modular Reactor, a factory-fabricated, lead-cooled nuclear system intended for remote communities, industrial facilities, military infrastructure, mining operations, and data centers.

The proposed design is intended to scale from approximately 3.5 megawatts to 50 megawatts of electrical output.

Several units could therefore be deployed together to support a much larger campus.

Morpheus also differs substantially from conventional water-cooled reactors…

The design uses lead as its primary coolant and is intended to operate at low pressure.

You see, lead has a very high boiling point, allowing it to absorb substantial heat without boiling under normal operating conditions.

And because the primary cooling system doesn’t depend on water or steam, the design also avoids the hydrogen production risks associated with some traditional reactor accidents.

Nuclea says Morpheus will incorporate passive heat removal and a negative temperature coefficient, meaning the nuclear reaction naturally slows as the reactor temperature rises.

And perhaps most importantly for the AI industry, it is being designed as a sealed, transportable system rather than a miniature version of a traditional nuclear station.

That could eventually allow a data center developer to install a reactor alongside computing modules, cooling equipment, backup systems, and energy storage.

The campus would still maintain a grid connection, but it would no longer depend entirely on the grid for its continuous baseload needs.

Powering AI Without Punishing the Public

The AI infrastructure boom isn’t slowing because the world lost interest in artificial intelligence…

It is slowing because the physical systems required to support AI haven’t caught up with the software.

And if AI campuses continue relying primarily on public grids, developers will continue to face increasingly fierce resistance from residents who fear higher bills, declining reliability, and massive infrastructure costs.

But micro modular reactors offer another path…

They could allow data centers to bring their own dependable power, just as they already bring their own servers, networking equipment, cooling systems, and backup generators.

And instead of forcing local utilities to rebuild the grid around a single enormous customer, reactors such as Morpheus could turn each data center into a largely self-contained energy campus.

The technology still has to be licensed, manufactured, financed, and proven in the real world.

But the companies that solve the industry’s electricity problems won’t merely supply another component to the data center boom.

They’ll make the next stage of that boom possible.

And in a world where communities are already blocking more than $100 billion worth of proposed AI infrastructure…

That may be one of the most valuable problems any company can solve.