You know something is strange when a coalition of major oil, gas, and petrochemical companies, the Louisiana Energy Users Group (LEUG), is collaborating with environmental non-profits. This unexpected alliance is a result of Entergy, Louisiana’s biggest utility, and its plans to build three new natural gas plants to support Meta’s huge two gigawatt artificial intelligence (AI) data center—a power load comparable to twice the maximum power consumption of the state’s largest city, New Orleans.
The new gas plants will cost $3.2 billion, the burden of which will not be wholly paid by Meta and instead shifted to Entergy’s ratepayers: the general public. LEUG argues that Entergy is leveraging its “monopoly to…propose financial risks on its existing captive ratepayers…while it reaps the return-on-equity benefits.”
In Wisconsin, two new gas plants with a lifetime of 30 years are being brought online to serve Microsoft’s AI data center. The company responsible, We Energies, chose not to invest in renewable energy sources to feed the data center, despite Clean Wisconsin attorney Brett Korte claiming that, “wind and solar, paired with battery storage, can deliver the power Wisconsin needs safely and for less money.” Instead, the state will be forced to shoulder an estimated additional $80–127 million in health costs per year due to increased rates of respiratory and cardiovascular diseases.
Evidently, Big Tech companies are willing to recklessly sacrifice the public’s health and offload their economic costs onto consumers through unchecked infrastructure growth and hidden utility deals while they soak up the reward.
According to a Harvard Law report, many data centers, the most expensive of which are built to support AI, make confidential agreements with power companies to get special low rates instead of going through the usual rate case, the formal public process where a utility asks regulators to approve changes to the prices it charges customers based on its costs and allowed profit. Utilities’ fees are subject to the “cost causation” principle, often a legal requirement described in state law which calls for the person or company that causes expenses to pay for them, so infrastructure costs are distributed among ratepayers. Arguing that AI data centers benefit everyone, utilities bend the cost causation principle to spread the cost across everyone’s electricity bills; thus by charging tech companies less, utilities ensure tech firms do not take their business elsewhere, securing a deal that raises both their profits. But when the special discounted rates don’t cover the power company’s full cost of supplying data centers with electricity, regular people and small businesses are forced by the utility to pay the difference through even higher power bills.
To appease the state’s public utility commission (PUC) to approve the rates, utility companies claim these private deals don’t impact other ratepayers, an assertion that becomes difficult to verify due to lack of transparency. For example, a regional transmission organization in the Pennsylvania-New Jersey-Maryland (PJM) grid region, which coordinates the flow of electricity between utilities across multiple states, approved an additional $5 billion in new interstate transmission projects specifically to aid with data center load growth. With Virginia utilities expecting to cover about $2.5 billion and Maryland utilities about $500 million, and state formulas requiring residential ratepayers to pay more than half of transmission project costs in Virginia and 66% in Maryland, the public will end up bearing most of the expenses for data center projects.
Additionally, in Virginia, Dominion Energy has identified $2.4 billion in new “data center driven” transmission projects, whose costs will be spread across all utility customers in the PJM region. Consequently, the average Dominion residential customer could see their monthly energy bill rise by $14 to $37 in today’s dollars by 2040, largely due to the growing energy demands of data centers.
Yet state utility regulators often face political pressure from governors or company leaders to approve large data center projects lauded for economic benefits, such as regional investment and job creation, even when they serve private interests. For example, Entergy Louisiana’s CEO Phillip May cited the Meta data center as “support[ing] a brighter, more sustainable future for all of Louisiana… benefiting generations to come.” Louisiana’s governor also touted the project as signaling a “new chapter” for the state because it would “create opportunities for Louisiana workers to fill high-paying jobs.” Despite Entergy being later unable to provide evidence of new jobs, the utility regulators fear being “seen as the veto point” blocking the latest tech boom, worried that higher rates could push Big Tech to invest elsewhere. They settle for worse.
Perhaps if AI companies were willing to be more transparent on their energy consumption, energy contractors would be less likely to believe and fund their claims of “constant power, 24/7, 365 days a year.” In order to protect the average customer, states should reform ratepayer protections to prevent the cross-subsidization of private infrastructure by following researchers’ guidelines, namely by standardizing how PUCs review special contracts and requiring utilities to offer fixed terms and conditions for all data centers.
Notably, the Public Service Commission (PSC) of Georgia, which harbors 158 data centers, passed a rule in January which allows utility Georgia Power to bill customers with electricity loads of more than 100 megawatts in a manner that accounts for the added risks from their high demand. Data centers will also be required to pay for the generation, transmission, and distribution infrastructure needed to support their projects, with every such contract to be submitted to the PSC for review. “By approving this new rule,” PSC Chairman Jason Shaw said, “the PSC is helping ensure that existing Georgia Power customers will be spared additional costs” from data centers.
Alternatively, Ohio’s PUC now requires data centers to pay upfront for 85% of their projected electricity usage. This rate structure applies broadly to large-load customers and forces them to shoulder more of the infrastructure costs they impose,thus reducing risk to residential ratepayers.
Financial costs aren’t the only burdens being borne by the public either. In Memphis, TN., Elon Musk’s portable generators, which his company xAI uses to help power his supercomputer and AI chatbot Grok, have been revealed to illegally violate the Clean Air Act. The Southern Environmental Law Center (SELC) estimated the generators emit around 1,200 to 2,000 tons of nitrogen oxides per year, exceeding limits on toxic and carcinogenic pollution.
Shelby County, which already suffers from higher rates of cancer and asthma due to smog, held a public hearing in April on xAI’s Clean Air Act permits. Residents of the predominantly Black county elucidated the risk of their families’ lung conditions being worsened by the generators’ nitrogen oxide emissions. “I can’t breathe at home, it smells like gas outside,” resident Alexis Humphreys stated in the hearing. “How come I can’t breathe at home and y’all get to breathe at home?” In a protest that same month, referring to Musk, state Rep. Justin Pearson, D-Tenn., proclaimed, “They put our lungs and our air on the auction block and sold us to the richest man in the world.”
In its letter to the Shelby County Health Department, the SELC argued for an emergency order to cease the operation of the 35 generators or for xAI to pay a fine of $25,000 for every day the generators keep running. Despite pushback, in July, Musk secured Clean Air Act permits for 15 of his generators. In response, the SELC has filed an appeal on behalf of the NAACP.
An analysis by a team of researchers from UC Riverside also reveals the profound effects training AI models can have on public health, as the process emits air pollutants such as fine particulate matter comparable to 10,000 round trips by car between Los Angeles and New York City. The group projects that by 2030, under a medium-growth scenario, U.S. data centers will be responsible for around 600,000 asthma cases and 1,300 deaths annually, corresponding to $20 billion in public health costs.
An additional issue arises due to the locational dependency of public health impacts; that is, negative health effects from data centers are concentrated in the areas surrounding them. Consequently, the health costs from AI disproportionately affect low income communities, whose financial burden, the UC Riverside report suggests, could amount to 200 times that of less impacted households.
Comparing maps of health costs against locations of data centers.
Above: Map adapted from ‘The Unpaid Toll: Quantifying and Addressing the Public Health Impact of Data Centers’
Below: Data acquired from ‘Data Center Map’
Notably, the research group found that the ten counties with the highest health costs due to data centers in the United States have median household incomes below the national average. In Virginia — the state with the most data centers — 9 of the top 10 most affected counties are also lower income. This uneven distribution of the burdens from AI risks exacerbating existing socioeconomic inequities.
Had U.S. data centers in 2023 been powered by on-site renewables, the resulting public health benefits could have been worth as much as $319 per household, with total nationwide health gains estimated at $7.6 billion. With 50 GW of renewable power waiting to be connected to the electric grid in central United States, AI companies are spoiled for choice.
Now is the time to call for corporate responsibility. If Big Tech wants to do the world a favour by investing in renewables with storage capabilities and improved efficiency, AI has the opportunity to become a leading player in the clean and ethical energy sector.
Associate Science and Technology Editor
