AI Data Center Energy Costs Are Starving the Manufacturing R

There is a contradiction at the heart of the current administration's industrial policy, and it is playing out on the power grid of the Rust Belt. The same PJM Interconnection that serves the data centers powering America's AI ambitions is now pricing out the steel mills and brick factories those industrial policies were designed to revive. This is a structural collision between two national priorities, not a market hiccup — and on the current trajectory, both cannot win.

The numbers are stark. Capacity costs on the PJM grid soared from $28.92 per megawatt-day in the 2024 auction to $329.17 per megawatt-day in 2026 — a more than 1,000 percent jump. Of the record $16.4 billion in costs from PJM's most recent auction, about 40 percent came from data center demand, a Reuters analysis found. These AI data center energy costs are the primary driver of the price surge. For manufacturers operating on thin margins, the effect has been immediate and severe.

Belden Brick Company in Ohio has seen its monthly capacity charge jump from $1,600 to $12,000. The Steel Manufacturers Association reports that American steel companies concentrated in PJM territory are now paying tens of millions of dollars in additional power costs each year. Given that electricity accounts for 20 to 40 percent of total steel production costs, and the US steel industry draws up to 11 gigawatts at peak production, every upward tick in capacity pricing directly erodes the competitiveness that reshoring policies are meant to restore.

The counter-argument is straightforward: data centers create jobs, too, and AI infrastructure spending is a national security imperative. I take that point seriously. But the comparison collapses under scrutiny. A single data center employs a few dozen to a few hundred people once built. A steel mill supporting a mid-sized Rust Belt town employs thousands, with a much wider supply chain multiplier. The energy consumed by one hyperscale facility could instead power an entire industrial corridor. When capacity pricing rises by an order of magnitude, we are not balancing two equivalent goods. We are cannibalizing a proven engine of middle-class employment for a speculative one.

This is not an argument against AI investment. It is an argument for honest accounting. The administration cannot simultaneously champion a manufacturing revival and allow the grid operator to allocate capacity to the highest bidder when that bidder is a data center backed by trillion-dollar market caps. The AI data center energy costs being externalized onto manufacturers amount to an implicit subsidy for Big Tech at the expense of the industrial base.

PJM's capacity auction mechanism is designed to ensure grid reliability by paying generators enough to stay online. It was never designed to adjudicate between AI infrastructure and factory production as competing national priorities. That is a policy question, and it has not been asked, let alone answered.

The broader implication is that the US is building its AI future on top of an energy grid that was not designed for this load, and the strain is showing up first in the places the administration has promised to revitalize. Without targeted policy intervention, whether through industrial electricity rate carve-outs, accelerated grid build-out, or capacity market reform, the AI data center energy costs will continue to be paid by the manufacturing sector in higher power bills and thinner margins. That outcome is not accidental market dynamics. It is the predictable result of regulatory inertia meeting a demand shock.

Why This Matters

The administration's manufacturing and AI agendas are on a collision course that neither set of policy architects appears to have modeled. Every kilowatt-hour committed to a data center is a kilowatt-hour that cannot flow to a steel furnace at an affordable price. If the PJM capacity pricing trajectory continues, the administration will have to choose between subsidizing factory power bills (which taxpayers would ultimately fund) or watching the manufacturing revival stall under the weight of energy costs that no amount of tariff policy can offset. The AI data center energy costs are not just a utility line item. They are the most concrete measure we have of a policy contradiction that has yet to be acknowledged.

Photo by Winston Chen on Unsplash

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team.

Why This Matters

Related Articles