Drone Images Expose Unpermitted Gas Generators at New Jersey Data Center

New Jersey has imposed a $1.1 million penalty on DataOne after finding that the company installed and operated gas generators at its Vineland data center without the air permits required by state law. The case is notable not only because of the size of the fine, described by state officials as the largest data-center enforcement action in New Jersey, but also because the equipment was exposed through aerial and thermal imagery rather than voluntary disclosure.

What investigators found

DataOne’s planned complex covers roughly 2.6 million square feet. Investigative reporting and images shared with state officials showed 62 gas generators on the site, with 45 operating when the thermal footage was recorded. The generators reportedly had capacities of 1,982 kilowatts, more than 50 times the state’s 37-kilowatt threshold for the applicable permit requirement.

The New Jersey Department of Environmental Protection said such engines emit carbon dioxide, nitrogen oxides, carbon monoxide, and other combustion-related pollutants. Those pollutants can worsen asthma, contribute to heart attacks, and increase the risk of premature death. The agency’s commissioner said data centers would not be allowed to build or operate “with impunity” in the state.

Why the fine has not reassured neighbors

• DataOne has 45 days to apply for the required air permits. It may continue operating the generators while seeking approval.
• The company says it disagrees with the penalty but will apply for permits for the temporary generators.
• DataOne also says it intends to transition to quieter, lower-emission fuel cells.
• Residents near farmland, homes, and two schools are asking what emissions were released and whether local air quality was ever properly measured.

The enforcement action follows months of concern. Residents raised questions at a public meeting early in the year, while a farmer reportedly noticed unusual equipment while taking images of a neighboring golf course. Inspectors later found tractor-trailer-sized generators that had not been present during an earlier visit. Floodlight’s reporting also said DataOne had sought permits for around 30 generators, withdrew that application after regulators identified deficiencies, and had maintenance records indicating that some engines were already being used.

Noise has added to the distrust. Residents have reported excessive noise and filed a federal lawsuit over public-health violations. Local environmental advocates argue that allowing the generators to keep running during the 45-day permitting window sends the wrong signal: companies may treat penalties as a cost of doing business rather than as a reason to stop unauthorized operations.

A wider AI infrastructure problem

The dispute illustrates a growing tension between AI expansion and local infrastructure rules. DataOne’s facility is associated with Microsoft’s Copilot tools, and rapidly rising demand for AI computing is increasing pressure on data-center operators to secure power quickly. Gas generators can be deployed faster than permanent grid connections, but they introduce emissions, noise, and permitting risks. Fuel cells may reduce some impacts, yet they do not remove the need for environmental review, public participation, and independent monitoring.

More than 80 data centers are planned or under construction in New Jersey. DataOne is notable as the state’s first data-center operator to rely so visibly on gas generators for an off-grid power supply. The state’s recent decision to end data-center tax credits shows that public support for large facilities is no longer guaranteed.

The central question is whether a $1.1 million fine is meaningful at the scale of a multibillion-dollar infrastructure project. Effective oversight may require more than a penalty: regulators could need to suspend unauthorized equipment before permits are issued, publish emissions data, and give nearby communities a meaningful role in decisions about power systems and expansion.

Source: Ars Technica AI