Permitting Is the Real Constraint on the Renewable Build-Out, Not Panels.

The artificial intelligence boom has spent the last few years sounding like a chip story. GPUs. Model training. Cloud capacity. Server racks. Tech executives saying “compute” so often that the word nearly filed for overtime.

But the next AI bottleneck is looking a lot less like a semiconductor and a lot more like a substation.

On June 18, the Federal Energy Regulatory Commission ordered the six major U.S. regional grid operators under its jurisdiction to either justify or reform their rules for connecting data centers, manufacturing facilities, and other large power users to the transmission grid.1 In plain English: the people who run large parts of the U.S. electric system have been told to explain how the grid will handle the new giants knocking on the door.

And those giants are not knocking politely.

The fight is not simply “data centers bad” or “AI good.” It is more complicated, and more interesting, than that. Data centers can bring jobs, tax revenue, investment, and new demand for power projects. They can also require expensive grid upgrades, new generation, water, land, transmission capacity, and local patience.

That raises the question utilities and regulators cannot avoid: who pays?

AP reported that data centers now account for about 5% of U.S. electricity demand and could triple by 2035, citing the Electric Power Research Institute. In Virginia, one of the country’s largest data-center markets, AP reported that data centers account for more than 25% of electricity demand and could rise above 40% by 2030.

Inside Climate News reported another number that should make grid planners sit up straight: Dominion Energy is reviewing 70 GW of data-center interconnection requests in Virginia.3 Seventy gigawatts is not a normal utility-service conversation. It is a full-on grid diet plan, except everyone keeps ordering more dessert.

The risk is that existing customers pay for infrastructure built mainly to serve new large loads. FERC’s order directly addresses cost transparency and cost allocation, because the political reality is obvious: if household bills rise while trillion-dollar tech companies build power-hungry campuses, the backlash will not require a white paper.

It will arrive in the next election, the next rate case, and the next town hall.

The AI power boom is not a single-fuel story. That is what makes it so important for energy readers.

Utilities and developers are already looking at gas plants because they can provide dispatchable power and support reliability. Renewables remain critical because large tech companies still have clean-energy targets and corporate buyers can sign big power-purchase agreements. Storage matters because it helps manage variable supply and peak demand. Nuclear is back in the conversation because around-the-clock clean power suddenly looks very attractive when your customer runs servers 24/7. Transmission matters because none of this works if power cannot reach the load.

In other words, the AI boom is not just creating demand for electricity. It is creating demand for every part of the electric system that was already stressed.

The least glamorous components may become the most important. Transformers, switchgear, interconnection studies, queue rules, siting approvals, and substation equipment are not exactly cocktail-party conversation. But if they are missing, the AI story slows down.

The grid is the bouncer now.

FERC’s move also shows how quickly AI has become a political energy issue. The commission’s action lines up with a broader push to speed power access for AI and other large loads. AP reported that companies including xAI, Google, Microsoft, Meta, Oracle, OpenAI, and Amazon signed the Trump administration’s Ratepayer Protection Pledge.2

That pledge matters because it acknowledges the central tension: the U.S. wants AI infrastructure, but voters do not want to subsidize it through surprise electricity bills.

There is also a local infrastructure question. Communities are not just asking whether data centers create jobs. They are asking how much water they use, what kind of generation serves them, whether air emissions rise, whether transmission lines cut through neighborhoods, and whether regular customers get pushed to the back of the line.

That is why FERC’s order is bigger than a technical filing deadline. It is an early sign that the data-center boom is being pulled out of the “tech growth” box and placed in the “public infrastructure” box.

Those are very different boxes.

The next 30 and 60 days matter.

Within 30 days, grid operators and transmission owners must submit reports explaining how they will ensure enough generation exists for current and new large loads.1 Within 60 days, they must either defend their current tariffs or propose reforms.1

That means energy professionals should watch for three things. First, whether grid operators propose new large-load service classes or interruptible/flexible service options. Second, whether regulators tighten cost-allocation rules so ordinary customers do not absorb the bill for hyperscale upgrades. Third, whether utilities and developers use AI demand to justify new gas plants, transmission projects, storage deployments, nuclear deals, or all of the above.

AI may still sell itself as software. But from here, its growth story runs through substations, generation queues, utility commissions, and local politics.

The cloud is getting very grounded.