ORANGEBURG, New York — Apps and websites are thought of as things that exist purely in cyberspace, but the online world is held up by a lot of physical infrastructure. That infrastructure is increasingly under the spotlight amid the rise of power-hungry artificial intelligence technologies like ChatGPT.
Some of that physical infrastructure comes in the form of large buildings full of computer hardware, known as data centers. Now, about 25 miles north of Manhattan, one of those crucial data centers is being erected at a construction site in Orangeburg, New York.
“So this is a purpose-built data center that’s going to have five suites in it, five data halls,” says Dan Fuentes, the Senior Vice President for Enterprise Sales at DataBank, the company behind the project.
Construction crews work to build a data center for Databank, Aug. 27, 2024, in Orangeburg, New York.
Databank’s Orangeburg location is the size of a warehouse. However, in the industry, data centers aren’t measured in square footage, but rather in megawatts of power.
“It’s gonna have 20 megawatts of critical power for our clients, so 30 megawatts total,” Fuentes tells ABC Audio.
Databank acts similarly to the owner of an office building. Companies as big as Apple and as small as neighborhood banks contract with the company, and those firms handle setting up and running their own computers and servers. Databank, meanwhile, is in charge of making sure those devices get constant security, cooling and power.
“So we have large clients, large hyperscalers, large banks, large enterprise clients, AI clients, who want to put their infrastructure somewhere where they’re positive it never goes down. So these data centers are purpose built to never have failure,” says Fuentes.
This data center is being built next to an electrical substation, from which the center will draw power most of the time, according to Fuentes. But in the event of a power outage or a natural disaster, huge generators that line the outside of the building kick on.
Angel Otero, the operations manager for this data center, says in the event of a major power outage the generators can keep supplying power so long as they have fuel. It’s something he encountered at a different data center during Superstorm Sandy in 2012.
“All the refineries on the East Coast were underwater, damaged,” he says. “We had constant fuel shipments coming in from Philadelphia. And we had — just, we were running on generator power for about a week. No issues, no concerns, no problems. As long as we had fuel we were up.”
Several large gas-powered generators line the outside of a data center being constructed in Orangeburg, New York.
Michael Dobuski/ABC News
Right now, there are more than 8,000 data centers around the world, according to the U.S. International Trade Commission, with more than 2,800 in the US.
“So, globally, there’s about 40 gigawatts of data center infrastructure installed around the world. About half of that is in the United States,” says Databank CEO Raul Martynek.
And he says business is booming.
“All that computation, as this technology has gotten more and more sophisticated, uses more compute cycles, uses more bandwidth, uses more storage. And all that drives demand for data centers,” says Martynek.
McKinsey predicts the data center industry will grow by 10% by the end of the decade. And while companies like Meta, Amazon, and Google have their own data centers, they also rent space from specially designed facilities, like the ones DataBank operates.
Cliff Stein, professor of industrial operations research and computer science at Columbia University, says all this attention on data centers shouldn’t come as a surprise.
“In the world almost everything we do is somehow tied to — either uses them directly or uses something that’s needed on these data centers. And so, in order to keep the economy running, we need this,” says Stein.
Everything from social media companies to streaming services rely on data centers, and recently, generative artificial intelligence has hit the scene.
“Now with the Large Language Models, it’s taken off, and it’s grown tremendously,” says Stein.
Martynek says the explosion of artificial intelligence tools like ChatGPT, Google Gemini and Apple Intelligence has sparked even more growth in his industry.
“It’s kind of like before ChatGPT and after ChatGPT. There’s been a tsunami of demand over the last two years, as everyone wants to adopt this technology and incorporate it into their business. I think it really caught the industry flat-footed,” says Martynek.
That’s why, he says, the sector is now in a race to build new data centers.
At the Orangeburg facility, Fuentes says their clients’ computer hardware is going to live in one of five 20,000-square-foot rooms, known as “data halls.” Floors will be installed about 3 feet off the ground, leaving room for industrial strength air conditioners. Because all this computing can generate a lot of heat, the design allows cold air under rows and rows of computers.
Otero says that means the inside of a data center can get noisy.
“Depending on how much utilization the customer is using at the time, yeah, it can get pretty loud in here. Noise canceling headsets are definitely a must,” he says.
Aside from the noise, air conditioners also require a lot of power. When that’s coupled with the normal power needs a data center like this one demands, it raises some environmental questions.
“The consumption of a lot of energy, depending on where that energy is being sourced from, has an effect on the climate the same way driving more cars would,“ says Jake Bittle, a staff writer for the environmental magazine Grist who covers climate change and energy.
“If you rely on a lot of natural gas or coal power plants to sustain these data centers, that is going to have an effect on the climate because there’s more fossil fuels being burned in order to power them,” he says.
A 20,000 sq. ft. “data hall,” that’s under construction at a data center, Aug. 27, 2024, in Orangeburg, New York.
The AI boom has only thrown fuel on the fire. A report from Goldman Sachs Research finds that asking ChatGPT one question takes up as much energy as ten Google Searches. Training a Large Language Model produces as much CO2 as five gas-powered cars, according to the MIT Technology Review. Bittle says as more data centers pop up to meet that demand, all that energy use could show up on consumers’ electricity bills.
“You’ll probably see some kind of strain on overall power supply. It may not have gotten there yet, but that would lead to some kind of impact for rate payers who, you know, are residential and commercial businesses,” says Bittle.
Martynek says to combat this, the industry is investing in more environmentally friendly power sources.
“We know that our electric grid is kind of outdated, it hasn’t been modernized. And we need more renewable energy connected to the grid,” he says.
But both Stein and Bittle are skeptical that the data center industry can fully rely on renewables, at least in the near term.
“It seems to me like it can make a dent, but it’s not going to — it’s not going to solve the problem completely,” says Stein.
“Even though the companies are doing I think, in many cases, a lot of good things to reduce the energy and water consumption of the data centers themselves, they’re just energy hogs and that has consequences,” says Bittle.
Fuentes says that AI companies account for just a tiny sliver of energy capacity Databank has sold overall.
“AI is a component of it, but it’s miniscule compared to the spectrum of verticals that we sell to.”
But the technology will be present in Orangeberg. Databank says when the location holds its grand opening later this year, 80% of the facility will be leased to Coreweave, an AI infrastructure company.
