Who knew that running the entire collective intelligence of humanity on a massive cluster of water-cooled GPUs would require, well, electricity? Apparently, the shock is starting to hit the grid.
For the last three years, the AI narrative has been a clean, silicon-scented fantasy of infinite growth. But as of this morning, the bill has arrived. Between a landmark Senate bill from Adam Schiff and a PJM report that reads like a horror story for regional utilities, the “AI infrastructure boom” is officially colliding with the messy, expensive reality of the North American power grid. And if you’re a human ratepayer, you’re currently the one holding the check.
The 75% Jump: PJM’s Q1 Warning
PJM Interconnection—the largest grid operator in the US, covering 65 million people—just dropped its Q1 2026 data. The headline number is a 75.5% jump in wholesale power costs compared to last year. But the real violence is in the capacity costs, which surged by a staggering 398%. In the PJM world, capacity is the premium you pay to ensure generators are ready when demand spikes. It’s the "preparedness tax," and it’s skyrocketing because hyperscalers are essentially camping out at the substation with their bags packed.
The geography tells the story. PJM is the home of Northern Virginia—the data center capital of the world. It’s where the "cloud" actually lives, in windowless concrete boxes that pull more power than mid-sized cities. As these facilities move from blueprints to live loads, the grid isn't just straining; it’s repricing its entire future around them.
The Schiff Bill: "Energy Cost Fairness" as De-Facto Regulation
Enter Senator Adam Schiff. His newly introduced “Energy Cost Fairness and Reliability Act of 2026” is the mechanism the industry has been dreading. It’s simple: it targets the growing electricity demands of AI data centers and requires them to pay for their own grid upgrades. No more socializing the cost of a 1.2 gigawatt training cluster across the utility bills of a thousand Maryland families.
This is "infrastructure policy as AI policy" in its purest form. If you can’t regulate the model’s weights, you can certainly regulate its watt-hours. By forcing developers to internalize the transmission capital costs—which Maryland’s People’s Counsel estimates at $2 billion for their state alone—the bill effectively introduces a "compute tax" that doesn't need a Silicon Valley lobbyist's approval.
The Singularity Soup Take
The honeymoon phase of “AI is just software” is over. We have entered the industrial phase, where model performance is gated not by clever math, but by the availability of switchgear and the patience of ratepayers. The Schiff bill is the first of many; as grid costs continue to leak into consumer bills, the political backlash will become a harder constraint on AGI than any safety board or copyright lawsuit.
What to Watch
Watch for the “Maryland Template” to spread. If the FERC complaint against PJM cost allocation succeeds, every state in the PJM and ERCOT regions will suddenly have a legal roadmap to force data centers to build their own substations. Also watch the “on-site generation” pitch; if the grid becomes too expensive or politically toxic, the big labs will move to small modular reactors (SMRs) even faster than planned.
Sources
ConsumerAffairs — "Senate bill would make AI data centers pay for power grid upgrades"
Startup Fortune — "AI data centers are pushing power prices higher on the PJM grid"
Reuters — "Are data centers pushing grid costs onto consumers?"