The mainstream financial press has officially found its latest villain: the artificial intelligence data center. According to the prevailing narrative, tech giants are building massive, power-hungry server farms that suck up electricity, strain utility infrastructure, and force American manufacturing companies to foot the bill through sky-high power rates. It is a neat, tidy story about Silicon Valley elites ruining things for blue-collar factory towns.
It is also completely wrong.
I have spent years analyzing utility rate structures and industrial energy procurement. When you look at how the American power grid actually operates, the panic over AI energy consumption reveals a profound ignorance of utility economics. The assumption that data centers inherently drive up costs for manufacturers is built on flawed logic. In reality, hyperscale data centers are the best thing to happen to industrial energy consumers in decades.
The Flawed Premise of the Data Center Scapegoat
The "lazy consensus" argues that electricity is a simple fixed-supply commodity. If Big Tech buys more of it, the price goes up for everyone else. This ignores how regulated utilities actually set rates.
Utilities do not price electricity based on simple supply and demand. They price it based on capital recovery and capacity utilization. A utility builds power plants and transmission lines to meet peak demand—usually a hot summer afternoon when everyone turns on their air conditioning. The rest of the year, that expensive infrastructure sits underutilized.
Enter the AI data center. Unlike a factory that runs two shifts five days a week, or a residential neighborhood that spikes at 6:00 PM, a data center pulls a flat, highly predictable load 24 hours a day, 7 days a week, 365 days a year. In grid parlance, they have a near-perfect load factor.
When a utility adds a massive, constant customer like an AI data center, it spreads the fixed costs of the entire grid across a much larger volume of sold megawatt-hours. Because the fixed infrastructure costs are shared among more total units of energy, the cost per kilowatt-hour can actually decrease for every other ratepayer on the system. Industrial manufacturers are not subsidizing tech companies; tech companies are diluting the overhead costs of the local utility.
Manufacturers are Asking the Wrong Questions
If a manufacturing plant sees its energy bill spike, executives immediately point a finger at the new data center down the road. They are misdiagnosing the disease.
The real culprit behind rising industrial power bills is not demand from AI; it is the staggering cost of grid modernization and clumsy regulatory mandates. Over the past decade, utilities have been forced to retire reliable, fully depreciated coal and gas plants prematurely to meet political carbon targets. Replacing that baseline generation with intermittent resources requires billions of dollars in new transmission infrastructure and battery storage.
Under standard regulatory frameworks, utilities pass these multi-billion-dollar capital expenditure costs directly to consumers, plus an assured 9% to 11% return on equity. That is why your power bill is going up.
Consider the mechanics of a typical industrial rate case. A factory owner might complain: "Why should we pay for grid upgrades when the data centers are the ones moving in?"
Here is the brutal truth they miss: data centers routinely pay for their own infrastructure. Hyperscalers like Microsoft, Google, and Meta do not just plug into a standard substation. They fund their own high-voltage substations, pay for dedicated transmission line extensions, and frequently sign long-term Power Purchase Agreements (PPAs) that finance new, independent generation capacity. They are bringing their own chairs to the party, while traditional manufacturers expect the utility to keep providing free seating.
The Vulnerability of the Contrarian Reality
Let us be completely transparent about the downsides. This capital-intensive energy model is not without risk.
If a utility overestimates AI growth and builds massive new natural gas generation facilities to support projected data center demand, and then the AI bubble bursts, those capital costs remain trapped in the utility's rate base. If the tech companies pack up or go bankrupt, the remaining industrial and residential customers will be left holding the bag for those stranded assets.
Furthermore, local grid congestion is a real, operational bottleneck. Even if a data center finances its own connection, it can create localized transmission constraints that prevent power from moving smoothly from west to east or north to south. This forces utilities to run more expensive, less efficient "peaker" plants locally, which temporarily drives up real-time wholesale electricity prices.
But these are engineering and planning challenges, not existential threats driven by tech greed.
Dismantling the Peak Demand Myth
Let us look at a concrete thought experiment to understand how this plays out financially.
Imagine a regional utility with 1,000 megawatts of peak capacity. Industrial manufacturers use 400 megawatts during the day but drop to 100 megawatts at night. The utility has to maintain 1,000 megawatts of infrastructure all day and night, meaning its equipment sits idle or underutilized for half the day. The fixed costs of that idle equipment are baked into the daytime rates charged to manufacturers.
Now, add a 200-megawatt AI data center that runs at 200 megawatts consistently, day and night.
During the day, total demand rises to 600 megawatts—well within the existing 1,000-megawatt capacity. No new power plants are needed. At night, demand stays at 300 megawatts. The utility's assets are now working twice as hard, generating twice as much revenue from the exact same physical infrastructure. Because the utility is a regulated monopoly with capped profits, that extra revenue must be used to offset the overall rate base. The manufacturer's daytime rate should, logically, go down.
Without Data Center:
Peak Capacity: 1000 MW
Daytime Load (Manufacturers): 400 MW
Nighttime Load: 100 MW
Utilization Rate: Low (High fixed cost burden per MW)
With AI Data Center:
Peak Capacity: 1000 MW
Daytime Load: 600 MW (400 MW Mfg + 200 MW AI)
Nighttime Load: 300 MW (100 MW Mfg + 200 MW AI)
Utilization Rate: High (Fixed costs diluted across more volume)
The manufacturers screaming about data centers are essentially complaining that a new tenant moved into their apartment building and started paying half the rent.
Stop Lobbying for Protections; Do This Instead
Manufacturing executives need to stop crying to public utility commissions for protectionist tariffs against data centers. It is a losing strategy that shows a total lack of market sophistication. Instead, industrial energy buyers need to adapt their procurement strategies to copy the tech sector.
- Co-Locate with Generation: Stop relying blindly on the public grid. Look at what companies like Amazon are doing by buying data centers directly adjacent to nuclear power plants. Manufacturers with high thermal loads should co-locate with independent power producers or build behind-the-meter generation.
- Monetize Interruptibility: Data centers cannot shut down without catastrophic data loss. Manufacturing plants often can. If you can curtail your operations for four hours during the hottest days of the year, you can sell that interruptibility back to the grid or to the data centers themselves for an absolute premium.
- Demand Flex-Rates: Force your utility to offer dynamic, time-of-use pricing that rewards you for consuming energy when the data centers are running smoothly and the sun is shining, rather than sticking to flat, outdated industrial tariffs.
The narrative that AI is starving American industry of power is a convenient excuse for manufacturing companies that failed to modernize their energy procurement strategies over the last twenty years. The power is there. The capital is there. Stop blaming the servers for your inability to manage a balance sheet.
