C harlotte, North Carolina, had at least 26 data centers by the end of last year. Northwest of the Queen City sits Hickory, a town of 46,000 that made its name manufacturing furniture and hosiery and later became a hub for making the fiber-optic cable that carries the internet. Hickory has no data centers of its own.

Since 2022, Hickory’s general obligation bonds have traded at an average yield spread of about 44 basis points. Charlotte’s have averaged 38. A yield spread is the extra interest investors demand over a safer benchmark, in this case a maturity-matched Treasury yield, so a wider spread means higher borrowing costs. The six-basis-point gap could simply reflect differences between the two cities. Charlotte carries the highest possible credit ratings, while Hickory sits a notch or two below.

But researchers at Georgia Tech found a similar pattern across the country. The higher borrowing costs showed up in communities near data centers, not the communities that actually host them, and were concentrated in states that don’t offer data-center tax incentives. Each additional data center within 25 miles was associated with an average 10-basis-point increase in neighboring municipalities’ bond spreads, according to “Who Pays for Data Centers? Evidence from Municipal Financing,” a new paper from Alex Hsu and Hala Moussawi of Georgia Tech’s Scheller College of Business. On $100 million of debt, 10 basis points works out to roughly $100,000 in additional interest each year. Host communities saw no comparable increase.

Data centers have become such a political football that almost any new fact about them can be turned into ammunition for either side. Supporters see billions of dollars of investment and new tax revenue. In Virginia’s Loudoun County, data centers occupy 4% of commercial properties but generate 39% of the county’s general fund revenue . Opponents see bigger power bills, strained water supplies and tax breaks for some of the richest companies on Earth. This study gives them something new to fight over. A 10-basis-point increase in borrowing costs isn’t going to bankrupt a town, but the town with the data center isn’t the one whose borrowing costs went up. Tim Chubb, chief investment officer at Girard , a Pennsylvania-based wealth manager with $5.9 billion in assets under management that also advises municipal pension funds, thinks of data centers as “local assets with a regional bill.”

The host gets the property and equipment tax base. Nearby communities likely share higher power costs, construction demand, water needs and competition for workers without collecting the same revenue. “Costs just don’t stop at the county line, even when the tax does,” Chubb says.

The researchers combined municipal bond trades with a database tracking the locations of thousands of data centers from 2014 through 2025. The harder problem was figuring out whether the data centers were actually responsible for the higher borrowing costs. Data centers don’t land at random. Places that attract them may differ from other communities in ways that also affect borrowing costs. So Hsu and Moussawi looked at high-voltage transmission lines built before the AI boom. Large data centers have traditionally needed access to that kind of infrastructure, and in the researchers’ data, proximity to those lines strongly predicted which regions saw more data-center growth after 2022. That gave the researchers a way to separate the effect of the data centers from the places that attracted them in the first place.

You might expect all this to show up in the bonds that pay for power, water and infrastructure. It doesn’t. The researchers found the strongest increase in general-purpose bonds issued by nearby municipalities, the debt cities use to pay for everything from public buildings to parks. They found little evidence of higher spreads on utility or transportation bonds.

Moussawi doesn’t claim to know why. One possibility is that much of America’s power infrastructure isn’t owned by the municipality issuing the bonds. An investor-owned utility may pay for a new transmission line, while the city still absorbs higher costs elsewhere.

Data centers may be the issue of the moment, but towns have been sharing the costs of their neighbors’ big development projects for a long time. Casinos have done it for years. One recent study found that for every job created inside a new casino, another 1.4 jobs popped up close by . But when researchers zoomed out to the broader area, the job gains disappeared. Big projects can be good for the place that lands them without being equally good for everyone next door.

State policy can move the bill again. Hsu and Moussawi found the clearest increase in general municipal borrowing costs in states without data-center tax incentives. In states with incentives, the researchers found a different pattern in school finances. State transfers to school districts fell about $673 per student after incentives were adopted, while spreads on education bonds rose about 12 basis points. The higher borrowing costs didn’t appear to come from a local boom. The researchers found no corresponding increase in employment, school enrollment or property values.

North Carolina has offered sales-tax breaks to qualifying data centers, although lawmakers repealed an exemption for the electricity they consume this year. That makes Charlotte and Hickory an imperfect example of the broader 10-basis-point finding. It also illustrates the paper’s larger argument. State policy can change where the costs appear.

The paper was released in July and hasn’t been peer reviewed. Tripp Kaiser, the executive director of the Center on Municipal Capital Markets at the LBJ School of Public Affairs at the University of Texas at Austin, said the researchers deserve credit for looking at an emerging issue in municipal finance. But he thinks it’s still too early to draw firm conclusions from a 10-basis-point difference. AI data centers are still a new source of demand for local power and infrastructure, and questions remain about how much debt communities will take on to accommodate them and how dependent they could become on a handful of enormous customers.

Chubb also sees value in the research, but puts the 10-basis-point finding in perspective. He said that amount is “really not moving the needle” for municipal investors today. Nor should the paper’s roughly linear relationship be read to mean that 20 new data centers will automatically add 200 basis points to a town’s borrowing costs.

But more than 40% of U.S. municipalities are already within 25 miles of a data center.

The midterm elections are less than two months away, and data centers are already on the ballot. In June, voters in Monterey Park, California , overwhelmingly approved a citywide ban. In Ashville, Ohio, residents have put two proposed data centers and a gas power plant before voters this fall. Candidates elsewhere are debating what data centers mean for power bills, tax revenue and jobs.

This study gives them something else to fight over. Opponents can argue that data centers push some of their costs onto communities that never agreed to host them. Supporters can argue that if those costs spread across a region anyway, more towns should want a piece of the investment and tax revenue that comes with them. The decision to welcome or reject a data center may happen town by town. Its economic effects don’t respect those boundaries. Saying “not in my backyard” may keep the building out. It may not keep out the costs.