One of the largest planned AI data centers on the U.S. East Coast is facing serious environmental scrutiny after investigators found dozens of gas-powered generators operating at the site without air permits.
The facility is DataOne in Vineland, New Jersey, a massive project being developed for AI infrastructure company Nebius. Microsoft is not the owner of the site, but it has a major commercial connection to it: Nebius agreed in 2025 to provide Microsoft with dedicated GPU capacity from the Vineland facility under a five-year contract worth about $17.4 billion, potentially rising to $19.4 billion with additional services.
The controversy began after nonprofit newsroom Floodlight used thermal drone footage to examine the partially completed site. The footage appeared to show at least 45 of 62 large gas-powered generators operating simultaneously.
The key problem is that New Jersey environmental regulators say those generators do not currently have air permits.
Why the Generators Are Controversial
Large data centers consume enormous amounts of electricity.
AI facilities are particularly demanding because they operate thousands of powerful GPUs around the clock.
When enough grid electricity is not available, developers increasingly look for ways to generate power directly at the site.
DataOne appears to be doing exactly that while construction continues.
The generators involved are not small portable units sitting beside a house during a blackout. Reporting describes them as tractor-trailer-sized gas generators capable of producing substantial electricity and air pollution.
The New Jersey Department of Environmental Protection confirmed that it “has issued no permits, nor are any permit applications under review” for generators at the Vineland facility.
That is what turned the situation from an ordinary data-center controversy into a possible legal issue.
A Former EPA Official Says Federal Law May Have Been Violated
Bruce Buckheit, a former head of air enforcement at the U.S. Environmental Protection Agency, told Floodlight that federal and state law generally require a final permit before generators like these are installed and operated.
He argued that DataOne’s operation of the generators appears to violate federal law and said the equipment should stop operating until the permitting situation is resolved.
That is an expert allegation, not yet a final government ruling.
The distinction matters.
The New Jersey Department of Environmental Protection said it noticed the generators during a July 29 inspection but had not yet completed its compliance determination when the investigation was published.
So DataOne has been accused of violating federal law, but regulators have not yet publicly announced a final enforcement conclusion.
DataOne Says It Intends to Follow Environmental Rules
DataOne did not provide a detailed response about why the generators were operating without permits.
It did say that the company “remains committed to meeting all applicable environmental and permitting requirements.”
Nebius did not respond to the original publication’s request for comment.
Microsoft also declined to comment.
The EPA told Floodlight that New Jersey regulators are responsible for the site’s air permitting.
That means the next major development may come from NJDEP rather than Washington.
Why Microsoft Is Connected to the Project
Calling DataOne a “Microsoft-backed” data center can create the impression that Microsoft owns or operates the facility.
That is not quite accurate.
The commercial structure is more complicated.
DataOne is developing the Vineland facility.
Nebius is the AI infrastructure company using the project to provide computing capacity.
Microsoft has signed a massive long-term contract to use that capacity.
Nebius announced in September 2025 that it would provide Microsoft with dedicated GPU infrastructure from Vineland over five years.
The agreement is worth approximately $17.4 billion through 2031, with the possibility of reaching about $19.4 billion if Microsoft purchases additional capacity.
That contract helped make the Vineland facility economically important long before construction was complete.
Read Nebius’s official Microsoft agreement announcement
AI Data Centers Are Becoming Extremely Power Hungry
The Vineland controversy is part of a much larger problem.
Artificial intelligence requires extraordinary computing power.
Training large models can involve tens of thousands of GPUs.
Running those models for millions of users requires additional computing capacity every day.
That means AI companies are building facilities with electricity requirements comparable to industrial plants.
The proposed DataOne facility is designed around approximately 300 megawatts of power capacity, according to local planning documents and reporting.
For comparison, 300MW is enough electricity to represent the consumption of a significant city-sized load depending on usage patterns.
And Vineland is not necessarily the project’s final limit.
Earlier announcements indicated the facility could potentially expand beyond its initial capacity.
That scale makes reliable electricity supply one of the biggest challenges facing the AI industry.
Data Centers Are Starting to Build Their Own Power Plants
Historically, data centers connected to the electricity grid like most other large commercial customers.
That model is becoming harder.
Grid connection queues can take years.
Utilities need new transmission lines.
Transformers can be difficult to obtain.
Power generation may not expand quickly enough to match AI demand.
So some developers are adopting a strategy sometimes described as “bring your own power.”
Instead of waiting for the grid, the data center builds electricity generation at the site.
Floodlight reports that nearly 60 U.S. data-center projects are now considering their own gas-fired power plants, according to energy analytics company Cleanview.
That represents a major change in how the technology industry interacts with the energy system.
A data center is no longer simply a large electricity customer.
It can start looking like a power-generation facility too.
Elon Musk Helped Popularize the Approach
One of the most famous examples is Elon Musk’s AI infrastructure.
His companies have used large numbers of gas turbines and generators around data-center sites in Tennessee and Mississippi to accelerate access to electricity.
That strategy attracted environmental criticism but demonstrated something important to the rest of the industry:
Waiting years for grid infrastructure can be avoided if a company generates power itself.
Other developers are increasingly following similar approaches.
The problem is that power generation comes with a different regulatory burden.
A data center might receive local building approval.
A gas-burning power source may require additional environmental permits because it produces nitrogen oxides, carbon monoxide, particulate pollution and greenhouse gases.
Speeding up construction does not eliminate those obligations.
DataOne Originally Wanted Large Gas Engines
The Vineland project initially proposed a much larger permanent gas-generation system.
DataOne sought approval for more than 30 Bergen gas-burning engines, equipment commonly associated with marine propulsion.
New Jersey environmental regulators found multiple deficiencies in the air-permit application, according to records reviewed by Floodlight.
The company eventually withdrew that application in May.
Instead, the project shifted toward a different power source:
Bloom Energy fuel cells.
The New Plan Uses 300MW of Fuel Cells
Vineland’s planning board has approved a second phase allowing DataOne to install fuel-cell infrastructure capable of supplying approximately 300 megawatts.
Fuel cells work differently from conventional gas generators.
Rather than burning natural gas through combustion, they generate electricity through an electrochemical process.
That generally produces much lower levels of some local air pollutants.
The system is therefore cleaner in several important respects than dozens of combustion engines.
But it is not emission-free.
The fuel cells still use natural gas.
And according to testimony cited in the reporting, the completed system could produce more than two billion pounds of carbon dioxide every year.
That creates another environmental debate even if the immediate generator-permit issue is resolved.
Why Isn’t the Facility Using the Grid?
One reason is speed.
The AI industry is racing to deploy computing capacity as quickly as possible.
Microsoft, Google, Amazon, Meta, OpenAI and other companies are spending extraordinary amounts on AI infrastructure.
Every month that a GPU cluster sits unused can represent lost revenue.
Nebius’s contract with Microsoft also contains delivery commitments.
Its SEC filing states that if Nebius fails to provide the agreed GPU capacity after applicable grace periods and cannot provide alternative capacity, Microsoft can terminate the affected service.
That gives Nebius an enormous financial incentive to bring Vineland online quickly.
A $17.4 billion contract changes the economics of delay.
That does not excuse environmental violations if regulators determine they occurred.
It does help explain why “speed to market” has become such a powerful force behind modern AI infrastructure construction.
The Generators May Have Been Running for Months
The exact operating history remains uncertain.
Floodlight reports that maintenance records suggest the first temporary generators may have been switched on as early as October 2025.
Separate thermal imagery from late June 2026 appeared to show around 25 generators operating. By August, Floodlight’s own footage appeared to show at least 45 running.
That suggests the generators were not simply activated for a brief emergency test.
However, regulators have not yet publicly stated exactly how long each generator operated or what enforcement standard will ultimately apply.
Those details could be important if penalties are considered.
Residents Say the Noise Is Constant
Air pollution is not the only controversy.
People living near DataOne have complained for months about persistent low-frequency noise coming from the site.
Residents have described a continuous humming or mechanical sound that can be heard inside nearby homes.
A class-action lawsuit has now been filed against the project over alleged noise pollution and nuisance impacts.
One local family told investigators that the sound sometimes wakes them at night and interferes with ordinary family life.
The complaints are significant because the site is not located in an isolated desert or industrial wilderness.
Homes and schools are nearby.
The Guardian reports the facility sits roughly a mile from two schools.
That means generator emissions and noise are not abstract environmental issues for surrounding residents.
Local Authorities Have Already Taken Action on Other Issues
The generator controversy is not DataOne’s first regulatory problem.
Cumberland County health officials previously cited the facility over nighttime noise levels.
Vineland has also issued stop-work orders related to parts of construction at the site.
One recent stop-work order reportedly involved movement of the same fuel-cell equipment that is supposed to replace the temporary generators.
That combination is adding to frustration among nearby residents.
Their concern is not simply whether one particular permit was missing.
It is whether the project is moving so quickly that regulatory compliance is being treated as something to resolve after construction rather than before it.
The Clean Air Act Is the Bigger Legal Issue
The potential federal-law problem centers on air pollution permitting.
The U.S. Clean Air Act creates a national framework for controlling emissions, but much of the day-to-day permitting process is handled by states through EPA-approved programs.
That is why the EPA told investigators that New Jersey is responsible for DataOne’s air permit enforcement.
Operating a sufficiently large pollution source without the required authorization can trigger enforcement actions and financial penalties.
The exact classification of DataOne’s generators, their combined emissions and how regulators treat their operation will therefore matter.
A former EPA enforcement official says the law has already been violated.
NJDEP has not yet publicly gone that far.
That is the current legal status.
Why Portable Generators Can Still Need Permits
Some people may wonder why a temporary generator needs an environmental permit at all.
After all, homeowners use portable generators during blackouts.
Scale changes everything.
One small gasoline generator produces relatively little pollution.
Dozens of enormous gas-fired units running simultaneously can effectively become a temporary power plant.
If 45 industrial-scale generators operate continuously, their combined emissions can become significant.
That is why regulators generally evaluate total pollution rather than simply asking whether a machine is technically portable.
Putting wheels under a power plant does not necessarily remove environmental obligations.
The AI Boom Is Colliding With America’s Electricity System
The bigger story extends far beyond Vineland.
AI infrastructure is growing faster than many power systems were designed to handle.
Utilities are receiving requests for hundreds or even thousands of megawatts from individual projects.
Transmission infrastructure can take years to build.
New power plants can take even longer.
Meanwhile, technology companies want GPU capacity immediately.
This creates a collision between two industries operating at completely different speeds.
Software companies think in months.
Electric utilities and power infrastructure often think in decades.
That mismatch is encouraging increasingly aggressive attempts to build electricity directly beside data centers.
Vineland is an unusually visible example.
It probably will not be the last.
Microsoft Is Spending Heavily on AI Capacity
Microsoft’s involvement also shows how enormous the AI infrastructure race has become.
Instead of relying only on its own Azure data centers, Microsoft is signing long-term contracts with specialized GPU infrastructure providers such as Nebius.
The Vineland agreement gives Microsoft access to dedicated GPU capacity without requiring it to directly build and own every part of the facility.
For Nebius, the deal provides an enormous anchor customer.
For Microsoft, it increases access to scarce AI computing resources.
The arrangement demonstrates how AI infrastructure is becoming an ecosystem involving:
Cloud companies.
GPU suppliers.
Data-center developers.
Power companies.
Utilities.
Natural-gas suppliers.
Specialized AI clouds.
Local governments.
And environmental regulators.
What looks like a technology project is increasingly an energy and infrastructure project too.
The Controversy Could Become a Test Case
If New Jersey ultimately takes enforcement action against DataOne, the case could attract attention far beyond Vineland.
Other AI developers are considering similar on-site generation strategies.
They will be watching what happens.
If regulators allow companies to operate large temporary generation fleets while permanent systems are being approved, developers may treat the approach as a viable shortcut.
If regulators impose major penalties or force immediate shutdowns, the opposite message will be sent.
The question becomes:
Can AI companies move faster than the grid without moving faster than environmental law?
Vineland may help answer that.
Data Centers Still Need Reliable Power
There is no realistic version of the AI boom without more electricity.
Large-language models, image generators, autonomous agents and scientific AI systems all require computing infrastructure.
Computing infrastructure requires electricity.
The solution therefore cannot simply be:
“Do not power data centers.”
The harder challenge is determining how that electricity should be produced.
More grid infrastructure could help.
Renewables paired with batteries could help.
Nuclear power could become part of the solution.
Fuel cells may have a role.
Natural gas may remain important for reliability.
But each choice has costs.
AI’s digital products may feel almost weightless on a laptop screen.
The physical infrastructure underneath them is anything but.
This Is Why Communities Are Pushing Back
Across the United States, opposition to data centers has been growing over concerns about electricity prices, water consumption, noise, land use and pollution.
Several states are now considering restrictions or moratoriums on new large projects as local resistance increases.
Data centers were once relatively invisible pieces of internet infrastructure.
AI has changed their scale.
A facility consuming hundreds of megawatts can affect the same resources used by surrounding communities.
That gives residents a much stronger reason to ask:
Who pays for new power infrastructure?
Who gets the electricity?
Who carries the pollution?
Who benefits economically?
And who decides where these facilities are built?
Those questions are becoming part of the politics of artificial intelligence.
The Microsoft Connection Does Not Make Microsoft the Alleged Violator
This distinction is worth emphasizing.
The allegations reported so far concern DataOne’s operation of generators at the Vineland facility.
Microsoft is a major future customer of the AI computing capacity.
Nebius is Microsoft’s contractual counterparty.
Microsoft itself has not been publicly accused by New Jersey regulators of operating the generators or violating the permit requirements.
So describing the facility as Microsoft-linked or Microsoft-backed is reasonable because of the enormous $17.4 billion contract.
Saying Microsoft itself broke federal environmental law would go beyond the available evidence.
That distinction matters in a story involving several companies and layers of ownership.
What Happens Next?
The most important development will be NJDEP’s compliance determination.
If regulators conclude that the generators required permits before operation, DataOne could potentially face enforcement action.
That could involve orders to stop using the generators, penalties, or requirements to obtain proper approvals before restarting them.
The facility is simultaneously moving toward its permanent fuel-cell plan.
Vineland’s planning board has already approved the second phase involving approximately 300MW of fuel-cell capacity and LNG storage.
Once those systems are operating, the temporary generator issue could become less operationally important.
But any past violations would not necessarily disappear simply because new equipment replaced the generators.
The Bigger Lesson Is About the Physical Cost of AI
AI often appears to exist entirely inside software.
Type a question.
Receive an answer.
Generate an image.
Write some code.
Behind that simple interface sit enormous buildings filled with GPUs.
Those GPUs require tremendous electricity.
The electricity requires generators, transmission lines, substations and cooling.
And those physical systems exist in real communities governed by real environmental laws.
DataOne’s Vineland controversy is therefore about much more than one fleet of generators.
It illustrates what happens when the extraordinary speed of the AI boom reaches the much slower worlds of energy infrastructure, environmental regulation and local government.
Nebius has a $17.4 billion reason to deliver Microsoft’s computing capacity quickly.
Residents have equally legitimate reasons to expect pollution and construction rules to be followed.
Those objectives do not have to be incompatible.
But the current dispute suggests the industry has not yet figured out how to pursue both at the speed AI companies want.
For now, the clearest facts are straightforward:
DataOne’s Vineland facility was observed operating dozens of large gas generators.
New Jersey’s environmental agency says it has issued no permits for those generators.
A former senior EPA enforcement official believes their operation violates federal law.
And state regulators are still determining whether formal enforcement should follow.
That makes Vineland one of the most important current examples of a problem AI companies are likely to encounter again and again:
Building the intelligence may be fast. Building the infrastructure legally is much harder.
