The Hidden Cost of AI Data Centers: Who Pays When Towns Say Yes?
AI data centers have become national security infrastructure, fast-tracked by federal policy, but the electricity costs and grid strain are distributed across entire regions, meaning residents far from the construction sites bear the financial burden. While individual towns negotiate local tax benefits and job promises, the broader power grid that serves 67 million people across 13 states faces potential reliability risks as early as 2027, according to analysis of the PJM Interconnection, which operates the wholesale electricity market for the largest concentration of data centers in the world.
Why Are Data Centers Suddenly Treated as National Security?
The classification shift happened quietly but with significant consequences. Data centers are no longer treated as ordinary commercial real estate; they are now reclassified as critical national security infrastructure, which accelerates their construction through executive order and changes how opposition is monitored. This reclassification means that federal agencies track opponents of data center projects using the same surveillance frameworks designed to monitor terrorism threats.
The Joliet, Illinois case illustrates how this plays out in practice. On March 16, the Joliet City Council chamber filled hours before a hearing on a proposed data center, with hundreds of residents opposing the project and union trades supporting it. The council voted 8 to 1 to approve a $20 billion, 795-acre campus on farmland beside the Chicagoland Speedway, with only Councilwoman Suzanna Ibarra voting no, citing her district's history as a "dumping ground" for infrastructure other areas reject.
What Do Towns Actually Get From Approving These Projects?
The financial math looks appealing at first glance. Developers project thousands of construction jobs and roughly $2.1 billion in local tax revenue over thirty years. The building trades turned out in force because construction work is real, well-paid, and unionized. A single hyperscale campus can become one of a county's largest taxpayers; in DeKalb, Meta's data center property taxes flow directly to schools.
But the permanent job creation tells a different story. The Joliet campus is designed to draw up to 1.8 gigawatts of electricity and as much as 150,000 gallons of water per day, yet it promises only about 700 permanent jobs. That works out to roughly $28 million in investment per permanent job, making clear these are not jobs programs in any traditional sense. The construction crews increasingly travel from project to project rather than being hired locally, and once a facility is energized, it runs on automation with a small, specialized staff usually imported from elsewhere.
How to Understand the Real Cost Structure of Data Center Expansion
- Local Benefits: Towns receive capped, predictable benefits including a few hundred permanent jobs and roughly $70 million per year in tax revenue spread across thirty years, making the annual local benefit modest relative to the infrastructure footprint.
- Regional Costs: The electricity demand is distributed across a regional grid serving 67 million people, meaning power prices rise for all consumers across 13 states, not just the host town, creating open-ended costs that are genuinely disputed among economists.
- Grid Reliability Risk: PJM has no authority to refuse a data center connection even when supply is short, and the region risks falling below its reliability standard as soon as 2027, creating potential brownout scenarios that affect entire regions.
The power grid operator PJM Interconnection cannot refuse data center connections even during periods of supply shortage. This structural limitation means that as data centers proliferate, the risk of grid instability increases. The region is at risk of falling below its reliability standard as soon as 2027, according to analysis of PJM's capacity auctions.
The cost debate is genuinely contested among experts. The capacity-auction spike in electricity prices is a forecast, not a meter reading. The research firm SemiAnalysis argues the increase is largely a simulation that does not reflect actual conditions, driven by the grid operator's own demand projections, and notes that Texas, with comparable data center buildout but different market design, saw power futures rise only a few percent. By that reading, the real increase to PJM bills is closer to 15 percent, with market design a bigger factor than data-center demand alone. However, PJM's own market monitor reaches the opposite conclusion, calling data-center load the primary driver and noting that price impacts are large and irreversible.
Why Does the Approval Process Look So Scattered Across Different Towns?
Each jurisdiction votes on a national phenomenon using its own zoning code and local exhaustion levels, producing scattered answers with no one summing the total impact. A few exits west of Joliet, Naperville voted 6 to 1 to deny a Karis Critical data center after residents warned about power load, noise, and diesel generators. The same developer took the same pitch north to Hoffman Estates, where the plan commission rejected rezoning 4 to 2 before a packed room, even with two data centers already under construction in the village. Lisle postponed a hearing the room could not hold. Blue Island residents booed the developers. Rockford drew petitions against a roughly 1,300-acre proposal.
Some places say yes. In Sangamon County, the County Board approved a $500 million CyrusOne campus after turning down a moratorium. Illinois already hosts 139 operating data centers with more than 120 planned, fourth in the country. The scattered approvals and denials create a patchwork where developers simply move projects one exit down the interstate when a town votes no, rather than stopping the expansion.
Data Center Watch counted roughly $64 billion in projects blocked or delayed amid local opposition over two years, then another $98 billion in a single quarter of 2025. The era of rubber-stamping these as quiet tax generators is over, yet the national demand for data center capacity continues unabated because these campuses are built against contracted enterprise workloads. Banks, insurers, and manufacturers now run data and AI systems as core operations, plus training and inference loads stacked on top. The demand is as durable as the businesses underneath it.
The siting logic has only intensified since the colocation era: go where power and land are cheap, and the fiber already runs, and move faster than the local process can adapt. Joliet checks every box, and so does much of the Midwest, which is why a no vote in Naperville does not stop the machine. It simply moves it one exit down the interstate.
The broader cloud computing market is accelerating this demand. The US cloud computing market is expected to grow from $485.54 billion in 2025 to $721.30 billion by 2030, at a compound annual growth rate of 8.2 percent. This growth is largely supported by ongoing IT modernization efforts across enterprises, with companies shifting ERP systems, banking platforms, and large data environments to cloud services. Morgan Stanley estimates global cloud capital spending at $582 billion, significantly above earlier levels and expectations.
The increased use of AI and data analytics raises pressure on cloud service providers, particularly those relying on GPU-based infrastructure for fraud detection, recommendation engines, and medical image analysis. Adoption is also affected by government policies; cloud-first and zero-trust policies are encouraging public-sector agencies to migrate workloads to secure clouds, with federal governments adopting AWS GovCloud and other government cloud environments for healthcare data systems, defense analytics, and digital public services.
The fundamental tension remains unresolved. The benefit a town like Joliet is promised is local and capped: a few hundred permanent jobs and about $70 million per year in tax revenue. The cost is regional and open-ended: the data-center share of the grid's capacity bill rises across 67 million people, with no mechanism for those distant residents to vote on whether they want to bear that burden.