We talk a lot about what AI can do, but not enough about where all that AI actually lives.
ChatGPT, Gemini, Claude and the countless other AI tools that have appeared over the past few years, all of which require enormous data centers packed with servers and GPUs. Building them requires land, electricity and, depending on the cooling system, water.
But in communities across the country, residents haven't always known exactly what was coming until surprisingly late in the process.
Data center developers have used nondisclosure agreements, shell companies and code names while negotiating projects with local governments. In some cases, public officials themselves have signed agreements preventing them from talking about projects that could eventually consume huge amounts of their community's electricity.
Now that secrecy is becoming part of the growing backlash against AI data centers.
New rules on data center development
Pennsylvania Gov. Josh Shapiro signed an executive order on August 18 imposing new rules on AI data center development in the state. Among them: state agencies under his authority can no longer enter into nondisclosure agreements with data center developers.
Pennsylvania is also removing AI data centers from its fast-track permitting program and requiring more public disclosure and community involvement before projects move forward.
It's a striking change, particularly because Pennsylvania has been aggressively courting the data center boom. Amazon announced plans last year to invest at least $20 billion in data center campuses in the state.
So how did we get from states fighting to attract AI infrastructure to governments demanding to know who's actually building it?
The secrecy isn't limited to Pennsylvania
University of Mary Washington researchers found that 25 of the 31 Virginia localities with an existing, approved or proposed data center had entered into nondisclosure agreements. The researchers warned that the true number could be higher.
In Texas, developers have used shell companies and NDAs that can make it difficult for residents — and sometimes even state officials — to determine who is behind a proposed project.
Wisconsin communities have faced similar questions over secret agreements surrounding potential data center developments.
The argument for confidentiality isn't completely unreasonable. Companies don't necessarily want competitors knowing where they're considering buying land or building billions of dollars worth of infrastructure before a deal is finalized.
The problem is what happens when those negotiations involve resources public citizens use, too.
A hyperscale AI data center isn't the same thing as a company quietly considering a new office building. It can require enormous amounts of electricity and resources, along with new transmission lines and other infrastructure. Local and state governments may also offer tax incentives to persuade companies to build there.
The lack of information can leave communities debating projects without knowing what they could eventually be asked to support. The World Resources Institute estimates that a modern AI data center can use as much electricity as 100,000 homes, while the largest facilities may consume far more. Large data centers can use as much as 5 million gallons of water a day, and communities may need to build substations, transmission lines, roads and water infrastructure that remain in place for decades.
Without details about a project’s size and resource demands, residents have little opportunity to judge whether the promised investment and jobs are worth those long-term costs.
Even Microsoft is backing away from NDAs
What's particularly interesting is that some of the pressure for transparency is now coming from inside the tech industry.
Microsoft announced this year that it would stop using nondisclosure agreements with local governments when developing data centers.
The company said it wanted communities to know about its plans and acknowledged that greater transparency was necessary to build trust. And while the company is hardly backing away from AI infrastructure with tens of billions of dollars committed to expanding the data centers required to train and run AI, it is an acknowledgment that the old way of building them may no longer work.
Bottom line: Communities are paying attention more than ever
For residents, staying informed may require looking beyond an announcement that might not come until a project is already well underway. Data centers can first appear in planning and zoning agendas, land sales, tax-abatement agreements, utility filings or applications for water and air permits. The project may be listed under an unfamiliar LLC or a name such as “Project Nova,” so searching only for Amazon, Microsoft, Google or Meta may not uncover it.
Residents can ask local officials whether they have signed an NDA, who will ultimately operate the facility and when the public will be able to review its expected electricity and water use. They can also ask who will pay for new substations, transmission lines, roads and water infrastructure — and what happens to those costs if the project is delayed, downsized or abandoned.
By the time construction equipment arrives, many of those decisions have already been made. Pennsylvania’s law is an attempt to bring the public into the conversation sooner.
The AI boom is becoming a local issue. Up until now, for most of us AI has felt remarkably intangible. But the debate over AI data centers is changing that. The physical infrastructures have become hard to ignore.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom's Guide
