AI’s power appetite just collided with Trump’s manufacturing agenda, and Rust Belt factories are eating the cost. According to Ars Technica, US manufacturers across the industrial Midwest are paying sharply higher electricity bills as data centers strain PJM Interconnection, the largest grid operator in the country. The squeeze hits steelmakers, brick factories, and other energy-hungry producers right where it hurts: profit margins.
What stands out here is the contradiction. Trump has championed “Made in America” manufacturing and the AI data center boom at the same time. Those two goals are now pulling against each other on the same power grid.
The numbers tell the story
The Ars Technica report, drawing on Reuters analysis, lays out how steep the climb has been:
- Belden Brick, a 141-year-old Ohio manufacturer, watched its monthly electricity bill jump from $1,600 to $12,000, driven by a higher capacity charge.
- Metallus, an Ohio steelmaker, says its electricity costs rose 70 percent since 2024, adding roughly $15 million in energy spending a year.
- PJM’s capacity prices, paid to power generators based on supply and demand forecasts, exploded from $28.92 per megawatt-day in 2024 to $329.17 in 2026.
That last figure is the tell. Capacity prices climbed more than tenfold in two years, and data center demand is a big reason why.
Why manufacturers feel it first
Steel is electricity-intensive in a way most industries aren’t. Power makes up 20 to 40 percent of the total cost of producing it. Each electric arc furnace pulls between 40 and 200 megawatts, and the whole US steel industry draws up to 11 gigawatts at peak. When capacity charges spike, steelmakers can’t absorb it quietly. The Steel Manufacturers Association warns its members in PJM territory are paying tens of millions more per year.
There’s a genuine irony baked in. Data center construction needs an estimated 1 million tons of steel annually, so the AI buildout is a customer for these same mills. But per the Wall Street Journal reporting cited by Ars Technica, the energy demand driving those orders is also inflating the cost of making the steel. Factories are selling into the boom and paying for it at the meter.
Why it matters now
This is significant because it reframes the AI infrastructure debate. The conversation has mostly been about chips, capital, and water. Power was treated as a line item. It’s now a live political and economic fault line.
Data centers concentrate enormous, steady demand in specific grid regions faster than new generation can come online. PJM’s capacity auctions price that scarcity, and every business customer on the grid pays the higher clearing price, whether or not they benefit from AI. Manufacturers, with thin margins and heavy loads, feel it first and worst.
For policymakers, the tension is real. You can’t promise a manufacturing revival and unlimited data center growth on a grid that isn’t expanding fast enough to serve both. Something gives.
Practical takeaways
For businesses and practitioners watching this unfold:
- If you run energy-intensive operations in PJM territory, model capacity charges as a rising cost, not a fixed one. The 2026 auction result won’t be the last jump.
- If you’re planning data center capacity, expect siting and interconnection to get harder as local grid stress becomes a political issue. Cheap, fast power is no longer a safe assumption.
- If you invest across this theme, watch grid operators, independent power producers, and nuclear and gas generation. Scarcity pricing rewards whoever can add reliable megawatts.
- Track the regulatory response. States that courted data centers with incentives are now watching residents and factories foot higher bills. Rate design and cost-allocation fights are coming.
The deeper question is whether the grid can grow fast enough to serve AI and industry at once. Right now it can’t, and the cost of that gap is landing on century-old factories in Ohio. Expect this to become a bigger fight, not a smaller one, as more data centers come online.
Full details are available at the original Ars Technica report.