As AI data centers scale, investigating their impact becomes its own beat

Data centers have long underpinned the internet as we know it, but the generative AI boom has ushered in a new era of rapid, largely unchecked development. Microsoft, Google, Amazon, and Meta alone are expected to spend over $700 billion on capital expenses in 2026, an increase of 60% from 2025. Much of that money will be funneled toward stockpiling specialized chips and breaking ground on massive new data center facilities across the U.S. and around the world.

Many of the data centers of old pale in comparison to new AI-focused facilities, some of which are the size of university campuses. The rapid development has inspired a new term for the tech giants that build them: hyperscalers.

For journalists, this moment of ballooning investment and aggressive rural land acquisition by tech companies presents an opportunity — and a responsibility — to investigate. Data centers drive climate change by burning fossil fuels, using large amounts of electricity, and requiring up to five million gallons of water a day to fuel cooling systems. Research has shown these facilities can harm the health of local residents through air and noise pollution, while providing minimal long-term job stimulus. Despite subsidies from national and local governments, many proposed data centers have been criticized for hiding the projected impacts on local communities under the guise of “trade secrets.”

“If you’re a tech journalist, you can go in. If you’re a climate journalist, you can also go in. If you cover business or energy or if you’re a very local journalist — there’s a story for you,” said Laís Martins, an investigative journalist at Intercept Brasil who published a series of major stories on data centers in Brazil over the past year. (Martins and Nieman Lab’s Andrew Deck previously worked together as reporters at the nonprofit publication Rest of World.)

First-time data center reporters may find the topic intimidatingly technical and challenging to humanize. From the outside, facilities may not look like anything more than windowless warehouses stocked with whirring machines. But major investigations over the past year have shown how many grounded stories and novel reporting strategies are emerging on the data center beat.

“Trying to get readers to grasp the scale is really important”

When Hannah Beckler and her team at Business Insider started looking into data centers for what would turn out to be a George Polk Award–winning series about their impacts on power grids, water supplies, and communities across the country, the centers were still relatively obscure. “[We were] trying to get our heads around what are essentially giant buzzing buildings,” Beckler, a correspondent at Business Insider, said. “How do we make people excited about that and understand why these are so important for us to be tracking from an accountability and regulatory standpoint?”

To start, they needed a framework for mapping and tracking data centers, and the team got its first big break when a source in Virginia told reporter Ellen Thomas to look into air permits, which data centers need to use backup generators. Similarly, water permits — which allow data centers to siphon off water to cool their extremely hot server racks — provide an idea of how much water they’re using. But getting those permits took hours of work.

“We submitted, I want to say, 103 different public records requests [for air permits],” Beckler said. Some states have air districts with their own small agencies that oversee air permits, and the team had to submit requests to each of those districts rather than state environmental agencies. “Water [permitting] is even more piecemeal,” she said. It’s especially tricky in Western states, where water is scarce but land for development is plentiful. States like Arizona, for example, have seen massive data center development. “You have all these teeny, teeny regulators and operators. So it was hundreds of individual requests to these little tiny agencies, and a number of times they had been privatized, so then all of a sudden, you had no public insight into their operation.”

The reporters received massive pushback on their records requests. Two water utilities in Colorado asked a judge to deny the requests — the first time Beckler had seen an agency take legal action of that kind — and Business Insider’s legal team was forced to respond to avoid a default judgment that could have set harmful case precedent for future requests. Once all the requests were fulfilled (sometimes the team had to send additional requests to double-check their information), the journalists had to figure out which big tech companies were behind which data centers; the permits were usually made out to murky shell companies with names like “Greater Kudu LLC” or “Magellan Enterprises LLC.”

Along the way, Beckler and her team realized that federal data, like an EPA dataset about environmental justice, was disappearing as the Trump administration pulled records from public sites. “I pulled that data 24 hours before it disappeared,” Beckler said.

Using the data from the records requests, Beckler and her team built a map of data centers around the country. They sent reporters to Virginia, Georgia, Arizona, and Ohio to talk to people who lived in communities that were seeing explosive data center development. As national reporters, their focus was finding places that were emblematic of the larger problem; in Arizona’s Maricopa county, for example, they found data centers being built in one of the most water-stressed regions in the country — a place where homebuilders and farmers had already halted projects because of water scarcity.

“Trying to get readers to grasp the scale is really important,” Beckler said. “It’s one thing to use sweeping generalizations, which can be impactful, but having something specific to be able to benchmark against [is important].” They found that data centers around the country are using an amount of electricity equivalent to what the entire state of Ohio used in 2023, and that 322 of them are responsible for 80% of the share of energy use. “That’s a generational transformation,” she said.

When the Business Insider team finally published their series last summer, they didn’t initially publish their map of data centers. But an accompanying video documentary about the investigation blew up (it has about 5.4 million views on YouTube as of publication time) and Beckler started hearing from readers asking for the map. So she convinced her editors to make an interactive map as a piece of service journalism so that readers could find their nearest data center. “It really immediately caught the imagination of a lot of readers,” Beckler said. “People care because it’s impacting their communities.”

“It’s important that we change our visual vocabulary around data centers”

Data centers aren’t very interesting to look at. Many of them are hidden away behind fences and trees, making them hard to see from the ground, but they also tend to be nondescript warehouses that give away little of what’s happening inside.

Evan Simon wanted to change that. So he bought a thermal drone.

“It’s important that we change our visual vocabulary around data centers,” Simon, a producer at the investigative climate nonprofit Floodlight, told us. “Thermal imagery provides a unique way of looking at these facilities. They’re not just these empty warehouses with no windows. Once you turn on that thermal, you can see the energy pulsing through them, and I think that helps us see these things for what they are. People don’t really realize that they are literally becoming fossil fuel plants.”

Simon specifically looked at a data center project in Mississippi, where Elon Musk’s xAI set up a data center with its own gas power plant — what’s known as a “behind the meter” project, which doesn’t hook up to the regular power grid. These kinds of projects are becoming more common around the country as data centers scale faster than local power companies can keep up, and 75% of them are powered by natural gas.

From local news reports, Simon knew that area residents complained about loud sounds and pollution from the gas plant, which didn’t have a permit and was operating illegally. But because of fences and treelines, no one had actually seen the plant.

“I’ve always been interested in using new reporting tools, especially those that can help visualize environmental threats in compelling and unique ways,” Simon said. Inspired by a Southern Environmental Law Center investigation into a similar plant in South Memphis, he settled on thermal imaging, which could prove whether or not the plant’s turbines were running — if they were, they would give off an intense heat signature.

Simon spent months preparing. For one thing, he had to get his drone operator’s license. But the drone was just one part of his toolbox, which also included standard reporting methods like talking to residents who had been documenting the gas turbines. He also had to figure out a place to fly the drone from; while he could have legally parked on the road outside the turbine complex and flown his drone from there, he didn’t want to risk running into security. Instead, he connected with a private property owner whose house was so close to the turbines that she had moved to a different part of town.

What Simon saw was striking. “As soon as I lifted the thermal drone above the tree line, I could see multiple massive stacks emitting very strong heat signatures that looked like these giant burning candles,” he said. Even though the residents had been hearing the turbines for months, Simon’s footage was the first time they saw them.

“A lot of the public had no idea that this was going on in the first place,” Simon said. “So when those images started circulating in the community, there was a real, palpable reaction, like ‘oh, this is what’s in our backyard.’”

The resulting investigation, which Floodlight co-published with The Guardian, was one of the first stories to visualize the energy use of data centers beyond maps and charts. While thermal images can’t provide a full picture — they show heat, which proves that turbines are running, but not the actual pollutants that are being emitted — the response has been overwhelmingly positive, and Simon intends to keep looking into new ways to visualize their impact.

“A story that rewards stubbornness”

While many individual stories about data centers are hyperlocal, the beat itself is undeniably global. American and Chinese tech giants are turning to markets outside North America and Western Europe in search of cheaper land, lower operating costs, and higher tax subsidies. Laís Martins, the reporter at Intercept Brasil, is taking inspiration from the work of journalists in other regions to bring accountability coverage to her Brazilian readers.

In particular, Martins credits the Pulitzer Center’s AI Accountability Network as a resource. The initiative encourages knowledge sharing among reporters covering the societal impacts of AI. Early stories by Pablo Jiménez Arandia, who first began investigating data center development in 2024, were a model for Martins. Jiménez recently co-reported a series on “mega” data centers built by Amazon Web Services, Microsoft, and Google in Spain, Chile, and Mexico.

“I learned different ways to try to work around the walls that governments and private companies put up,” said Martins. “It mostly confirmed my feeling from my initial reporting that in our context — the Global South — big tech companies are not using their own names. They’re either using shell companies or intermediaries, which makes the reporting even harder.”

Earlier this month, Martins co-reported a story detailing how big tech companies are masking their data center development projects in Brazil by partnering with less conspicuous local players. TikTok, for instance, secretly routed a data center project in Northeast Brazil through a wind power company. In the state of São Paulo, a sports management company was the face of a data center project proposal secretly backed by Microsoft.

Even when projects touch down in communities hundreds of miles away from Brasília, sourcing inside the federal government has been instrumental. For her initial TikTok story, Martins sought out public meeting logs that showed the company’s head of infrastructure in Brazil sitting in with high-level federal officials. She then reached out to government sources in the attendance logs to confirm that they had been discussing a data center proposal she’d already identified in the Northeast state of Ceará.

In December, several months after Intercept Brasil broke its story, TikTok publicly confirmed its involvement in the project. By that time, an indigenous group in Ceará had rallied to oppose the project’s development on their lands and filed a formal complaint with federal authorities. Some local community leaders said they didn’t know what data centers were until news about the TikTok project broke.

“When you’re able to pinpoint the company behind [the data center], the reporting is so much more impactful,” Martins said, comparing the response to other stories she’s published that only named relatively unknown developers. Some readers told her they deleted TikTok from their phones after her investigation. “It makes the story more relatable to people.”

The data center beat is fertile ground, but Martins has learned firsthand that these stories are rarely straightforward.

“[Data center] reporting is a story that rewards stubbornness. The breakthroughs have come from reporters…who haven’t stopped at the first no,” she said. “At first you’ll think I have no idea what I’m doing — like you’re walking in a dark room. But eventually things will start shaping up and the more you learn, the better you get.”