A 141-Year-Old Brick Factory’s Power Bill Rose 90%. Data Centers Drove PJM Capacity Prices Up 1,038%. Here’s the Displacement Math.

$1,600 a month. That was the capacity charge on the electricity bill at the Belden Brick Company in Sugarcreek, Ohio, a 141-year-old manufacturer whose products sit in the walls of the Texas Alamo and Notre Dame University. Then data centers arrived in the region, and the capacity charge jumped to $12,000 a month. Total electricity costs surged 90 percent in a single year. Not gradually. Not over a decade. In one year. The bricks did not get heavier, and the kilns did not get hungrier. The grid got a new tenant, and the new tenant outbid the old one.

Belden Brick is not an outlier but the opening exhibit in a systemic displacement that is rewriting the economics of American manufacturing in the thirteen states and the District of Columbia covered by PJM Interconnection, the largest power grid operator in the United States and the grid that hosts more data center demand than any other system on Earth.

PJM’s capacity prices, the wholesale rates that power generators charge to guarantee electricity availability on peak-demand days, jumped from $28.92 per megawatt-day in 2024 to $329.17 per megawatt-day today. That is a 1,038 percent increase across two auction cycles, driven almost entirely by one variable: data center interconnection requests flooding the queue faster than generation capacity can be built.

Meanwhile, the U.S. Energy Information Administration released its July 2026 Short-Term Energy Outlook on Tuesday projecting that commercial electricity consumption will surpass residential consumption this year for the first time in American history. Commercial: 1,550 billion kilowatt-hours, against residential at 1,508 billion. Data centers are classified as commercial loads, and that crossover is not a statistical footnote. It is the moment the American grid shifted from serving households as its primary customer to serving server racks, and no one held a vote.

The Revenue-per-Megawatt Gap

Understanding why the grid is choosing data centers over factories requires one calculation: revenue generated per megawatt of electricity consumed.

Start with hyperscale cloud providers, since their economics are the most transparent of any industry that consumes electricity at scale. Amazon Web Services reported approximately $115 billion in trailing twelve-month revenue as of Q1 2026, operating an estimated 7 to 8 gigawatts of data center capacity across its global fleet. That yields roughly $14.4 million to $16.4 million in gross revenue per megawatt-year. Microsoft’s Intelligent Cloud segment, which includes Azure, generated approximately $105 billion over the same period on an estimated 6 to 7 gigawatts, producing $15 million to $17.5 million per megawatt-year. Google Cloud, smaller but growing at 28 percent annually, generated about $46 billion on roughly 4 gigawatts: $11.5 million per megawatt-year.

Now compare manufacturing against that backdrop. According to the Bureau of Economic Analysis, U.S. manufacturing value added reached approximately $2.85 trillion in 2025. Industrial electricity consumption, per EIA data, totaled roughly 1,065 billion kilowatt-hours, which translates to an average load of about 122 gigawatts. That gives manufacturing a sector-wide average of approximately $23,400 in value added per megawatt-year. But that average spans semiconductor fabs, which are extraordinarily electricity-efficient per dollar of output, and primary metals and brick kilns, which are not.

For a brick manufacturer like Belden, the math is considerably worse than the sector average, because kiln-fired ceramics are among the most energy-intensive processes in all of construction materials manufacturing. The U.S. brick industry generates roughly $4.5 billion in annual revenue across approximately 8.3 gigawatts of thermal and electrical load. Revenue per megawatt-year: about $540,000, which is 26 times less than what a hyperscale data center produces from an equivalent megawatt of grid capacity.

Load Type	Economic Output per MW-Year*	Ratio to Brick Manufacturing
AWS (hyperscale cloud)	~$14.4–$16.4M	27–30x
Microsoft Azure	~$15.0–$17.5M	28–32x
Google Cloud	~$11.5M	~21x
U.S. manufacturing (avg value added†)	~$23,400	~0.04x
Brick manufacturing	~$540,000	1x (baseline)

*Cloud and brick figures are gross revenue. †U.S. manufacturing average is BEA value added, not gross revenue; gross manufacturing output per MW-year is roughly $57,000. The core cloud-vs-brick comparison uses the same metric.

When PJM runs a capacity auction and data centers bid against factories for access to a constrained grid, the outcome is predetermined by these ratios. A data center can absorb a capacity charge of $329 per megawatt-day, which works out to $120,147 per megawatt-year, because that charge represents less than one percent of its annual revenue per megawatt. For a brick manufacturer, that charge consumes 22 percent of revenue per megawatt. At Belden Brick’s scale, the $124,000 annual increase in capacity charges alone could represent the difference between operating and closing a kiln line.

The Historic Crossover

The EIA’s July outlook makes the structural shift visible in a single chart. U.S. electricity demand hit a record 4,195 billion kilowatt-hours in 2025, the second consecutive all-time high, and the agency projects it will rise to 4,269 billion in 2026 and 4,399 billion in 2027. Within that total, something unprecedented happens this year: commercial electricity sales are forecast to reach 1,550 billion kilowatt-hours while residential sales fall to 1,508 billion. Never before has the commercial sector exceeded residential consumption in the history of U.S. electricity statistics.

That crossover matters because of what “commercial” now means. Historically, commercial electricity demand meant office buildings, hospitals, schools, grocery stores. Today it means those things plus several hundred hyperscale data centers, thousands of enterprise colocation facilities, and an interconnection queue at PJM alone that contains 30,000 to 60,000 megawatts of pending data center requests, according to Joseph Bowring, president of Monitoring Analytics and PJM’s independent market monitor. AEP Ohio has 13,000 megawatts in its local queue. PJM has warned it could face electricity shortfalls as early as 2027.

“The problem is very simple,” Bowring told the Akron Beacon Journal. “Thirty thousand to sixty thousand megawatts of load want to come on the PJM system. There’s not enough capacity.”

PJM Chief Operating Officer Stu Bresler was more circumspect but no less alarmed, telling Reuters in May that the demand ramp is unlike anything the grid has experienced “since, probably, the Industrial Revolution.” He is not exaggerating, and anyone who has watched PJM’s interconnection queue data over the past eighteen months would struggle to find a comparable demand acceleration anywhere in the developed world outside of wartime mobilization. The grid operator is now considering a market overhaul to address the structural mismatch between how fast data centers can be built and how slowly new power generation can be permitted and connected.

Who Pays the Capacity Premium

Capacity charges are not distributed equally, and the asymmetry is structural rather than accidental. They fall disproportionately on industrial customers because those customers have higher peak-load contributions relative to their total consumption. While capacity charges represent approximately 10 percent of a residential electricity bill, they can constitute 20 to 30 percent of an industrial manufacturer’s bill, according to interviews conducted by Reuters with manufacturers, energy attorneys, and industry advocates.

PJM’s auction arithmetic compounds this disparity through a cascade that touches every ratepayer in the territory. When data centers flood the interconnection queue, the grid operator’s demand forecast rises, and when forecasted demand rises while generation buildout lags behind it, the clearing price in the capacity auction increases, and when the clearing price increases, every load-serving entity in PJM’s thirteen-state footprint passes that cost through to its customers, including factories that had nothing to do with the demand spike. Data centers pay capacity charges too. For them, it is rounding error on a cloud invoice. For a factory running century-old kilns, it is existential.

The numbers from the U.S. Department of Energy tell this story at the state level with brutal clarity. Industrial electricity prices in Pennsylvania rose 31 percent year-over-year as of December 2025, and in Ohio they rose 26 percent, while the national average for industrial customers over that period was 7 percent and residential prices in those states rose 14 percent and 9 percent, respectively. That differential is not random. Pennsylvania and Ohio are simultaneously legacy manufacturing states and emerging data center hubs, which means their factories and their data centers are competing for capacity on one constrained grid, and the factories are losing.

Five of the eight U.S. states classified as emerging data center hubs are in the Rust Belt, according to Synergy Research Group. This geographic overlap is the arithmetic of displacement. Simple as that. Data centers are not being built in the desert or on a separate grid; they are being built alongside the factories they are outbidding for identical constrained electrons on aging transmission infrastructure.

Ohio’s Attempted Fix

Ohio is the first state to formally respond to this dynamic with binding regulatory action. In July 2025, the Public Utilities Commission of Ohio approved a settlement requiring new data center customers of AEP Ohio to pay for a minimum of 85 percent of the electricity they request each month, regardless of whether they consume it. If a data center tells AEP it needs one gigawatt, it will be billed for at least 850 megawatts even if it draws zero. Both the settlement and financial viability requirements extend to exit fees if projects are canceled, and the terms remain in place for up to twelve years.

Its logic is defensive: utilities have been building grid infrastructure based on data center demand projections that may never materialize, and when those projections overshoot, the stranded costs land on existing ratepayers. As Bradley H. Belden, president of Belden Brick, wrote in the Akron Beacon Journal: “Paper projections were elevated to the level of real projects, shaping public narratives and policy decisions even when the underlying demand remained uncertain. The result is a system where nearly all of the risk is shifted onto customers.”

Belden’s framing cuts to the built-in asymmetry at the heart of the dispute: when demand forecasts are high, utilities build infrastructure, regulators approve rate increases, and factories pay higher bills, but when forecasts are revised downward, no one gives the factories a refund. Customers absorb the downside; utilities earn their regulated return on equity regardless; and the data centers that triggered the demand forecast in the first place may not even exist yet when the bills arrive.

PUCO Chairperson Jennifer French said the ruling would protect ratepayers while providing a “dependable and reasonable environment” for data center companies. The Data Center Coalition, a national trade group, has argued that data centers “provide significant benefits to Ohio” including job creation, tax revenue, and the downstream economic activity that cloud computing enables: small businesses running on AWS, telehealth platforms serving rural hospitals, remote work infrastructure that kept the American economy running during the pandemic. An AWS-funded study by E3 Consulting estimated that Amazon’s data centers generate $33,500 per megawatt in surplus value annually that utilities can pass on to other ratepayers. For a 100-megawatt facility, that is $3.35 million in annual ratepayer benefit, which is real money but represents roughly 2.8 percent of the $120 million annual capacity cost increase that a facility of that size helps trigger across PJM’s territory at current auction prices.

The Displacement Threshold

There is a calculable point at which capacity prices make manufacturing uneconomical in PJM territory, and the grid is approaching it faster than most participants acknowledge.

At the 2024 baseline capacity price of $28.92 per megawatt-day, annual capacity cost per megawatt was $10,556. For Belden Brick, with an estimated plant load of 3 to 5 megawatts, that represented $31,700 to $52,800 per year in capacity charges, manageable alongside energy and transmission costs on a brick manufacturer’s margins of roughly 8 to 12 percent.

At today’s $329.17 per megawatt-day, annual capacity cost per megawatt is $120,147. For that 3-to-5-megawatt plant, capacity charges alone reach $360,400 to $600,700 per year. On a manufacturer operating at 10 percent margins on $15 to $25 million in annual revenue, capacity charges now consume 2.4 to 4.0 percent of revenue, a threefold increase that arrives as a fixed cost, immune to any operational efficiency the factory can achieve.

The displacement threshold, the capacity price at which the total electricity cost exceeds a manufacturer’s pre-tax profit margin, varies by subsector. For a brick manufacturer at 10 percent margins, running on 4 megawatts with a $20 million revenue base, the electricity bill (energy + capacity + transmission + distribution) was approximately $1.2 million per year at pre-spike prices, or 6 percent of revenue. At current capacity prices, total electricity cost rises to approximately $1.8 million, or 9 percent of revenue. Margin left: under 1 percent, meaning one more capacity auction clearing at a premium pushes the factory into a loss.

Primary metals manufacturers, paper mills, glass plants, and chemical producers face similar arithmetic. The Industrial Energy Consumers of America, whose members consume roughly 7 percent of total U.S. electricity, has called for regulatory scrutiny. “This can have short- and long-term impacts on whether or not these facilities can continue to operate,” said president Paul Cicio.

What We Don’t Know

This analysis relies on public revenue data for cloud providers that may overstate per-MW productivity because not all revenue is proportional to data center capacity. AWS revenue includes services that run on customer infrastructure, and cloud margins include software value that is not directly tied to electricity consumption. At that level, a $14 to $17 million per megawatt figure is a ceiling, not a measurement; the true number is likely lower but unavailable because hyperscalers do not disclose capacity-level unit economics.

On the manufacturing side, the comparison carries the opposite limitation. Sector-wide value added per megawatt averages energy-intensive subsectors like primary metals and cement with efficient subsectors like pharmaceuticals and electronics assembly, producing a mean that may understate the productivity of many individual factories. Belden Brick’s revenue-per-megawatt is likely below the subsector mean for clay product manufacturing because its kilns are among the most energy-intensive processes in the building materials industry.

Most critically, the 30,000-to-60,000-megawatt figure for pending data center interconnection requests at PJM is a gross number that does not account for speculative applications, projects that will never break ground, or demand that will ultimately be served by on-site generation rather than grid power. The Ohio PUCO’s 85 percent minimum payment rule is designed partly to filter out speculative demand, but the rule applies only to AEP Ohio and does not cover other utilities in the PJM footprint. How much of the queue is real demand versus optionality remains the single most consequential unknown in American energy planning.

The Strongest Counterargument

The most serious case against the displacement narrative is that capacity price spikes are self-correcting. High clearing prices are the market’s signal to build more generation. At $329 per megawatt-day, nearly every proposed power plant in PJM’s territory clears the profitability threshold, which should attract investment in natural gas, solar, battery storage, and eventually nuclear capacity sufficient to bring prices back down. In this view, the spike is painful but temporary, a feature of the market mechanism doing exactly what it was designed to do: sending a price signal that accelerates construction.

A coalition of governors led by Pennsylvania’s Josh Shapiro has already intervened, successfully pushing PJM to impose capacity price caps that limited the most recent auction’s increase to 22 percent rather than allowing another unconstrained spike. Market reformers argue that PJM’s proposed market overhaul could further dampen volatility by modernizing how capacity is procured and allocated.

This counterargument is correct on mechanism but wrong on timing. A combined-cycle natural gas plant takes 3 to 4 years to build. Transmission lines take 7 to 10 years to permit and construct. A data center takes 12 to 18 months. The gap between demand growth and supply response is not a market failure that high prices will fix in a quarter. It is a physical constraint that high prices will fix in a decade, and in the intervening years, the factories that cannot absorb the premium will close. Some will reopen when prices normalize. Most will not. Kilns that go cold do not always relight, and skilled workforces that scatter do not always return.

What You Can Do

If you manufacture in PJM territory: Model your electricity cost exposure at capacity prices 50 percent above current levels. That is not a worst case; the Carnegie Mellon Open Energy Outlook projects an additional 8 percent national average increase from data center growth alone, and PJM’s concentrated exposure could double or triple that regional premium. Consider on-site generation, demand response programs that compensate you for curtailing peak load, and whether your state’s public utilities commission has created a data-center-specific tariff that separates their costs from yours.

If you regulate utilities: Ohio’s 85 percent minimum payment rule is a template, but it only works if it applies across the PJM footprint, not just to one utility in one state. A data center that cannot get favorable terms from AEP Ohio will apply to a neighboring utility without that rule, producing regulatory arbitrage that benefits no one except the developer.

If you invest: The commercial-surpassing-residential crossover creates a durable investment thesis: grid infrastructure companies (Quanta Services, MasTec, MYR Group) that build transmission are selling into a demand cycle that will not normalize until generation catches supply. But capacity price caps imposed by state governors introduce political risk into what the market considers a pure supply-demand play.

If you vote in a PJM state: Ask your state legislators whether they have created rate structures that distinguish between data center load and manufacturing load. Lumping a 100-megawatt data center and a 4-megawatt brick factory into the same “large commercial” rate class, then allocating capacity costs by peak load contribution, is the mechanism by which factories subsidize the infrastructure that serves data centers. The fix is not complicated, at least not technically; it is a tariff redesign, which is not a radical proposition in a regulated industry whose entire business model rests on tariff structures. But tariff redesigns require public pressure because the beneficiaries of the current structure have more lobbyists than the people it harms.

The Bottom Line

The American grid is making an economic choice. It is not being stated as a choice, because it is happening through the opaque machinery of wholesale capacity auctions, interconnection queues, and load-serving entity allocations rather than through a vote or a policy declaration. But the math is a choice. When 30,000 to 60,000 megawatts of data center load compete for grid access alongside factories that have operated for a century, and the capacity price mechanism allocates costs proportionally to peak demand without regard for who caused the demand to rise, the result is a transfer of grid access from low-revenue-per-megawatt users to high-revenue-per-megawatt users. That transfer has a name in economics: displacement. It happened when mechanized agriculture displaced tenant farmers. It happened when container shipping displaced longshoremen. It is happening now when data centers displace factories, and it is happening on a grid that was built with public money, regulated as a public utility, and justified by the promise of serving the public equally.

The bricks in the Alamo were fired in an Ohio kiln. Whether the kiln that made them survives the decade depends on a wholesale electricity auction that most Americans have never heard of, run by a grid operator that most Americans cannot name, and driven by a demand spike that most Americans will experience only as a higher number on their monthly bill. The grid is choosing. It just isn’t telling anyone.