Somewhere in America right now, a data center humming with AI workloads is demanding as much electricity as an entire city. The bill for building all the new infrastructure to supply that power has to land somewhere. Spoiler: the utilities aren't planning to eat it themselves.
The Demand Nobody Planned For
For most of the last few decades, the people who run America's electric grid had a pretty comfortable job, planning-wise. Demand grew slowly and predictably. You built a little capacity here, upgraded a line there, kept the lights on. It was boring in the best possible way.
AI broke that. According to Axios, data centers are now showing up to grid operators requesting amounts of electricity that used to be associated with powering entire cities. Not a neighborhood. Not a suburb. A city. And they want it fast, they want it reliable, and they want it cheap.
What 'Once-in-a-Generation' Actually Means Here
Axios frames this as a 'once-in-a-generation decision' about how America's electricity system should grow, and that framing is doing a lot of work. What it really means is that the rules written when the biggest load on the grid was a steel mill or a shopping mall are completely unequipped for what's happening right now.
The grid was not built for this. The transmission lines, the substations, the interconnection queues, the regulatory frameworks governing all of it, none of it was designed with the assumption that a single corporate campus might need the output of a mid-sized power plant. We are, to use the technical term, in uncharted territory.
PJM and FERC Are in the Hot Seat
The debates aren't just abstract policy arguments happening in think tanks. Axios reports they are actively unfolding right now at PJM, the nation's largest grid operator, covering 65 million people across 13 states and the District of Columbia, and at the Federal Energy Regulatory Commission. These are the institutions that actually decide who connects to the grid, in what order, and under what terms.
PJM's interconnection queue, the line of energy projects waiting to connect, has been a disaster for years. Thousands of projects, including solar and wind farms that could be generating clean power right now, are stuck waiting years for approval. Drop a hyperscale AI data center into that mix and you have a genuine resource allocation nightmare. Who jumps the line? Who pays for the new transmission needed to serve the guy who just jumped the line? These are not rhetorical questions. They are the actual questions being argued over right now.
The 'Who Pays' Question Is the Whole Fight
Here is where it gets genuinely contentious. When a massive new load connects to the grid, somebody has to pay for the upgrades to the transmission system required to serve it. The traditional model spread those costs across all ratepayers, on the theory that everyone benefits from a stronger grid. Big Tech, shockingly, is very fond of this model. Regular electricity customers, less so.
The alternative is to make the entity causing the need for new infrastructure pay for it directly. This is called cost causation, and it is the kind of phrase that sounds boring until you realize it's the difference between your electric bill going up because Google needed a new substation and Google paying for the substation itself. Consumer advocates and many state regulators are pushing hard for cost causation. The data center industry is lobbying just as hard in the other direction.
Scarcity, Access, and the Queue Nobody Wants to Talk About
Beyond the money fight, there is a harder question lurking underneath all of this: what happens when there simply isn't enough power to go around? Grid operators are already flagging reliability concerns in parts of the country where demand is spiking faster than new generation can be built and connected.
Axios notes the debates include not just who pays but who gets access to scarce power. That is an extraordinary sentence to be writing in the United States in 2026. We are talking about electricity rationing scenarios, about industrial customers and residential customers potentially competing for a finite resource, because the AI industry's appetite for power has genuinely outrun the system's ability to supply it. The AI companies will tell you this is a story about bureaucratic permitting delays. It is also, undeniably, a story about what happens when an industry scales faster than the physical world can keep up.
The Dingo Take
Let's be honest about what this is. Some of the most profitable companies in human history, firms sitting on hundreds of billions in cash, are engaged in an active lobbying campaign to make sure that ordinary ratepayers subsidize the infrastructure required to run their AI products. The argument, dressed up in the language of grid modernization and shared benefit, is essentially: we are too important to pay our own bills. It would be funny if it weren't going to show up as a line item on your electric statement.
The PJM and FERC proceedings happening right now are genuinely consequential and almost completely invisible to the public. Regulatory proceedings are where industry gets what it wants when nobody is watching. The data center lobby has armies of lawyers and lobbyists working these dockets full time. The people who will pay if they win are busy with their actual lives. That asymmetry rarely resolves in favor of the people with the actual lives.
The AI boom is real, and the infrastructure challenge it creates is real. None of that obligates the public to hand Big Tech a subsidy disguised as a rate structure. If building the grid for AI is worth doing, and it probably is, the companies profiting most from it should be writing the biggest checks. Any regulatory outcome that ends differently than that is not a compromise. It's a mugging.
