Last December, Richland Parish, a small rural county in northern Louisiana that is home to 20,000 people, became an unlikely epicenter of the artificial intelligence boom. Facebook’s parent company, Meta, selected it as the site of a new $10-billion data center that will be the company’s largest worldwide. But it waasn’t the bucolic charm that drew one of the world’s most powerful tech companies to Richland. It was the energy—and Meta needed a lot of it.
Meta’s data center needs an estimated 2.2 gigawatts of power to sustain the tens of thousands of computer chips that will be used to train and run the company’s AI models. That’s roughly twice the energy used by the entire city of New Orleans during peak consumption—and it would require building $3 billion worth of new power infrastructure, such as gas plants and transmission lines, before the end of the decade. Utility companies across the U.S. wooed Meta for the contract, which was ultimately awarded to Entergy, a major utility in the Southeast. What was left unsaid was how this new infrastructure would be paid for.
Research by the Environmental and Energy Law Program at Harvard Law School points to a troubling answer. The rush to build new data centers to meet growing AI demand risks shifting the infrastructure costs onto consumers in the form of increased electricity rates, effectively subsidizing the development of such facilities for trillion-dollar technology companies.
“The costs of expanding and operating energy infrastructure have historically been socialized by utilities because it was recognized that increasing access to electricity was in the public interest,” says Ari Peskoe, who directs the Law School’s Electricity Law Initiative.
A century ago, policymakers decided to grant utility companies a monopoly on electricity distribution to offset the costs of operating and expanding energy infrastructure. “Data centers have been part of our internet infrastructure for decades,” Peskoe explains, “but now there is dramatic growth in data center energy use, driven by the world’s wealthiest corporations racing to develop AI, which raises the question of whether the socialization of those costs makes sense anymore.”
During the past few years, the race to build AI data centers has driven an unprecedented demand for power.
This model worked reasonably well when the demand for energy grew at a steady and predictable rate. But during the past few years, the race to build AI data centers has driven an unprecedented demand for power that has made the old way of doing business untenable. In Texas alone, a single utility reported demand for 119 gigawatts of power from data centers, which is more than the current power generation capacity of the entire state.
“It’s taken around 100 years to develop 82 gigawatts in Texas,” says Eliza Martin, a legal fellow at the Environmental and Energy Law Program who coauthored the paper on Big Tech’s energy demands with Peskoe. “It’s hard to fathom how we’re just going to double that in the next couple of decades. We’re talking about energy demand for entire cities regularly materializing out of thin air.”
The researchers identified several mechanisms that shift data center costs onto regular ratepayers. Of particular concern, they say, are confidential special contracts between utilities and data centers that offer discounted rates to technology companies while concealing the overall project costs.
In Louisiana, for instance, Entergy is building three natural gas plants to power Meta’s $3-billion data center. “Entergy has told regulators their intentions are to put this amount in the calculations they use to set rates for everybody,” says Peskoe, “meaning they would socialize the cost of this infrastructure for everyone. They say they have a 15-year deal with Meta that will cover the cost, but they haven’t provided any information to anyone about what’s in the contract.”
This lack of transparency prevents proper scrutiny from regulators and the public, and utilities have used their monopoly power to distribute costs to ratepayers in similar ways in the past. For example, a recent lawsuit brought against Duke Energy, a major utility servicing the Carolinas, revealed that the company had offered discounted power to one of its largest customers because it planned to shift the cost to its captive ratepayers, effectively creating a $325-million subsidy borne by the consumer.
“We know that this happens in the utility context because of [instances] like the Duke case,” says Martin. “I don’t think it’s a far reach to say that’s also what’s happening with data centers.”
Quantifying the precise cost to ratepayers is an inexact science due to the subjective nature of the rate-setting processes used by utility commissions. But past research suggests that the costs for individual consumers may be substantial. One study on data centers in Virginia concluded that ratepayers in the region could see their electricity costs increase by between $150 and $450 per year by 2040 if current rate structures continue and data centers are built as planned.
For Peskoe and Martin, the key to ensuring that tech giants pay their fair share for data centers is having engaged regulators on state utility commissions and transparent processes for evaluating deals struck between utilities and data center operators. This can often be challenging given the intense political pressure from state legislatures and governors’ offices to approve data center projects with the hope that they will spur economic development.
“The people matter,” says Peskoe. “You really need vigilant regulators [who] are willing to put in the time and effort to oppose these deals that pass costs on to ratepayers.”
Alternatively, Peskoe and Martin suggest that regulators could cut utilities out of the picture completely to avoid the risk of ratepayers subsidizing the cost of data center development. Instead, tech companies could develop private energy parks that isolate data center costs from utility ratepayers because they would be designed to provide power solely to the data center.
Although the projections for data center energy demand in the U.S. vary wildly, Peskoe and Martin say that the overall trend is still up. Whatever the energy needs of data centers turn out to be, they will certainly be greater than what the existing energy grid can handle, as the Texas case illustrates. Billions of dollars in new energy infrastructure will have to be built to service these massive facilities, and the question of who will pay for them isn’t going away.
Peskoe and Martin note increasing interest in their research from public utilities commissions and ratepayer advocates who are turning their attention to the issue. Earlier this year, several consumer and environmental advocacy groups launched protests against the Louisiana facility arguing that ratepayers were subsidizing the costs of the energy infrastructure, and Peskoe and Martin expect to see more public pushback in the future unless regulations change.
“We’re not making a value judgment about artificial intelligence, we’re just asking who is going to pay for it,” says Peskoe. “From our perspective, if these companies want to pursue this data center development, they ought to pay for that.”
