Whether you deposit a check on your bank’s app, stream a movie, or fly to meet a client, you rely on data centers every day. Even when you’re offline, your digital life awaits inside them — their interiors filled with rows and rows of networked computer servers — poised for your smartphone tap. But while internet use and the number of data centers have both increased since the 1990s, artificial intelligence is supercharging development of the latter.
Training and running AI models requires serious computing power, so data centers have supersized. Insiders refer to these ballooning facilities as “hyperscale” data centers, and to their owners, such as Google, Amazon, and Microsoft, as hyperscalers. Yesterday’s data centers were 250,000 square feet in size; today’s AI-ready buildings measure in the millions. But it’s their energy footprint that’s increasingly significant, given the vast amount of power they consume. A decade ago, 30 megawatts was considered large. Now, 200-MW data centers are common. Meta is building a 5-gigawatt facility in Louisiana.
Given these data centers’ need for energy, cooling water (to keep all that hardware from overheating), and land use, many communities in the United States are protesting plans for them. Further, AI skeptics are questioning whether the technology will deliver returns to companies and society that justify the huge investment needed to build them in the first place. The country’s biggest tech companies expect to spend around a combined $650 billion on AI infrastructure and data centers this year alone, and a McKinsey report projects that by 2030, the industry will require $7 trillion in capital expenditures to satisfy demand.
Developers say that given demand, we’re nowhere near overbuilding, pointing to the near-zero rate of vacancies in data centers. Meanwhile, hyperscalers promise they’re working on sustainability. And though the debate over whether AI will do more harm or good for society continues, some research shows that adopting the technology is worthwhile for corporations. A recent report from Wharton Human-AI Research and GBK Collective found that “nearly three-quarters already see positive ROI, and four in five expect positive returns within two to three years.”
The data-center sector is evolving blisteringly fast, and forecasts about negatives and positives are tough to make in such an unprecedented space. For timely insights, we asked Wharton alumni working across the sector — along with Wharton scholars studying it — for their perspectives on development, impact, and what’s ahead for the engines powering AI.
Data Center Pioneering
Marc Ganzi W93
CEO, DigitalBridge
Marc Ganzi was inside a data center years before most of us had even heard of them. His digital infrastructure roots date back to 1994, when he started a company that built cell towers. By 1998, he had a fiber-optic cable business piping high-speed internet into the Northeast Corridor. The owner of a defunct Westin hotel building in Seattle invited him to consider his rooftop for cell towers. The interior was a server farm.
“On each floor, they were generating millions of dollars leasing racks to AOL, Excite, Yahoo. The cables looked like spaghetti,” Ganzi recalls. “And I said to myself, ‘Wow, someone’s going to figure this out, and there’s going to be a real business.’”
In 2013, Ganzi began carving his path into that business. He formed DigitalBridge and was soon buying cloud computing data centers. “Fast-forward to today: We own over 220 data centers globally,” he says. “We manage about $55 billion-plus of data-center assets around the globe.” Customers include Amazon, SpaceX, Microsoft, Google, and Meta, and the company operates in Latin America, Africa, Europe, the Gulf, Asia, Canada, and the U.S.: “We’re going to deploy close to $29 billion of new capital this year to continue building out for the industrialization of AI.”
At the end of 2025, Japanese tech giant SoftBank announced a deal to buy DigitalBridge for $4 billion, citing the value of the latter’s data centers, 500,000 cell sites (owned and master-leased), and 200,000-plus route miles of the fiber vital to networking. Ganzi will continue to lead what will be a separately managed entity.
From his perspective, there’s no current danger of overbuilding data centers: “We’re massively behind.” Indeed, amid the expansion race, power and material supply constraints are stymieing many developers.
“Power is the gatekeeper,” Ganzi says. Many developers are grappling with long wait times to hook up to the U.S. electric grid, but thanks in part to his early entry into the sector, DigitalBridge secured 24 GW through power suppliers over the last decade. “We’ve got to use it correctly, but we’re really fortunate to be ahead in that regard.”
He’s optimistic about AI returns, pointing to the concept of “inference” — AI lingo for when a trained model starts to yield real-world results. “We’re at the inception of inferencing, he says, “and once you get to inferencing, you really start to turn on AI.” Ganzi paints this future vividly: “How do you make drones land perfectly to deliver that package or a Domino’s pizza? How do you take a piece of luggage from an airplane to the carousel without it touching a human being and make sure that it arrives 40 percent quicker? Within four years, autonomous trucks will be picking up your trash.”
As for data centers’ environmental impacts, Ganzi says he’s sticking by the company’s pledge to be carbon-neutral by 2030, noting that solar, wind, and battery storage are among the “arrows in my quiver.” He’s also keeping an eye on two companies that are currently testing the capture of carbon emissions for conversion back into power: “Nothing is commercially viable yet for data centers. But I can say with a lot of integrity that within the next 10 years, we’re going to be able to recapture CO2 emissions. We’re going to be able to harness that and either sell it back into the grid or put it back into the data center.”
Essential Energy
Andrea Hepp WG26
Vice President of Power Development, Shell
Downtime costs data centers money, with some outage losses running over $1 million. Such financial hits, along with today’s impatient user expectations, drive operators to aim for the industry gold standard of “five nines.” The term refers to a 99.999 percent uptime — fewer than six minutes of disruption a year. Andrea Hepp admits that she didn’t know about the “five nines” until she started spending her days at Shell thinking about how to fuel these power-hungry sites. Energy-intensive AI has only heightened the need for a highly reliable energy source. “To do this, your buildout has to have a good grid connection and really good backup,” she says.
Hepp’s focus is the hyperscalers and co-location providers (where multiple companies rent space in a data center). She coordinates utilities, regulators, equipment manufacturers, capital providers, and engineering partners.
Due to the current state of infrastructure and regulatory hurdles, wait times to connect to the nation’s electricity grid can stretch to more than 10 years. For a quicker alternative, in the past year, more and more large data-center developers and owners started exploring “behind-the-meter” (BTM) solutions — independent, dedicated onsite power plants. Some are eyeing options like nuclear, wind, and solar, but the go-to is natural gas turbines. BTM means less reliance on the grid even when a connection finally happens. Still, almost as soon as natural gas turbines were embraced as a speedier path to power, turbine supply shortages arose. And yet another challenge for data-center customers like Microsoft is their public commitments to carbon neutrality. Burning natural gas emits carbon dioxide.
“It’s about building repeatable, scalable energy solutions,” Hepp says. “In certain regions, grid interconnection timelines don’t always align with the pace of AI expansion, so we evaluate complementary solutions alongside traditional grid supply, balancing speed, reliability, capital efficiency, and long-term decarbonization goals.”
As big tech companies pursue sustainability while delivering AI services to customers, Hepp foresees renewable and gas energy in the mix. She also believes innovation can reduce environmental impacts, citing the potential of battery energy storage systems, which could allow data centers to, say, ease off the grid when overall demand is high but rely less on gas generator backup. Battery technology continues to advance, and while such systems are costly, some data-center operators are investing in them. In February, alongside announcing plans for a new data center in Minnesota, Google publicized a deal in that state to build what would be the world’s largest battery system.
“People want to get their data centers ready as soon as possible,” Hepp says, “but now, what they’re realizing is that if they want to keep growing in the future, they have to take a quick pause and look at what actually works and what doesn’t work.”
Expanding Physical Footprints
Jeff Blau WG92
CEO, Related Companies
What inspires a prolific developer and manager of $70 billion-plus in real estate assets including multifamily homes, commercial office space, and retail to get into the business of building data centers? For Jeff Blau, the answer is renewable energy.
Though Related Companies is best known for projects like the 28-acre Manhattan neighborhood transformation dubbed Hudson Yards, it launched a renewable-energy business in 2020. Since then, energyRe has been building utility-scale solar and battery storage projects throughout the U.S. and selling the power to utilities and corporate partners. But then something changed.
“About three years ago, we started getting a whole bunch of inbounds from the big tech companies looking to buy the power,” Blau says. “At first, we didn’t really understand what they were doing with all this power. This was before the current hysteria around data centers. And we realized that they were building these data centers and were looking to have renewable energy to power them and meet their renewable energy commitments.”
According to Blau, his team soon realized they could bring their expertise in large-scale construction, complex infrastructure, and energy to meet the hunger for data centers. So they began incubating a business to address the demand, hired several data-center veterans, and officially kicked off Related Digital in early 2025, with plans to raise $8 billion and start breaking ground on a $45 billion development pipeline that initially includes sites in Michigan, Missouri, Wyoming, and Illinois.
The swelling of data centers gives Related an edge, Blau says: “The history of data centers has really been much smaller. What we’re talking about building now are really complex, large developments. There are very few people that have experience building these things at scale, because they haven’t existed at that level until the last couple of years. And the good news is that the advancements in the chips and technology mean that we can be really thoughtful about their size and power versus how they fit into the surrounding area responsibly.”
To steer clear of overbuilding, “We’re not building any of these buildings on spec. We’re partners with many of the big tech companies, and we’re really building for them.” Further, when talking to those partners, Blau hears that demand far exceeds supply: “I don’t know if it lasts forever, but I think it’s going to be a long time before the demand abates.”
Tech firms are also leaning on Related’s decades of experience developing housing across the U.S. to address community concerns. “We operate in many of these cities and counties,” Blau notes. “You can’t just come into these towns and do whatever you want.” In Cheyenne, Wyoming, for example, Related Digital committed $3.5 million to support an affordable housing project. “We spend an awful lot of time working with communities, finding out what they want, what the needs are,” Blau says.
The company’s partners must pay for all the energy and infrastructure upgrades that local utilities have to make to power the data centers, Blau says. He adds that they’ve also embraced the newer “closed-loop air cooling system” that many data centers are starting to adopt. According to Blau, with this method, a data center uses “less water than the farmer used on the property before we put our buildings up.”
Ultimately, Blau says, data centers serve national interests: “This is a huge opportunity for the United States to really be the leader in this space and be the AI superpower. Right now, we have a tremendous lead, but we can’t stop. We have to really keep going.”
Connecting the Networks
Kate Johnson WG94
CEO, Lumen Technologies
While data-center construction is grabbing headlines, an integral part of the AI boom is tunneling, literally, under the radar: the miles upon miles of fiber cable along which data speeds. Among fiber network owners, Louisiana-headquartered Lumen Technologies is a giant. When Kate Johnson became its CEO in 2022, Lumen had about 12 million miles of fiber running between cities across the U.S. Under previous names (including CenturyLink), the company laid this network, beginning in the 1990s, for the internet. But speculative installation ran headlong into the dot-com bust, and telecom firms were left with stranded assets.
The data-center boom has put that infrastructure back in play. “This legacy telecom company, like many others, was struggling for a while with a declining business,” Johnson says. “Now, we’ve pivoted our physical assets to become hugely valuable as the trusted network for AI.”
Under her leadership, Lumen’s fiber network grew to 17 million miles by the end of 2025. The goal for 2031: 58 million intercity miles of high-capacity, low-latency fiber. “We’ve done $13 billion of supply-side AI economy construction deals, with more in the pipeline, connecting the data centers for the biggest technology companies in the world so they can provide AI services to businesses,” Johnson says.
Lumen is establishing brand-new fiber routes but also upgrading existing fiber or putting new fiber (referred to as “overpulling” in the industry) into PVC conduits that were installed during the dot-com boom and left fallow. “We are capitalizing on a once-in-a-generation rise in demand for data-transport capabilities and leveraging that to drive a return on assets that were paid for a couple decades ago,” Johnson says. A partnership with Corning has led to improvements so that more strands of fiber fit into the same sheathing — quadrupling data capacity in what was laid as far back as the late 1990s, according to Johnson.
Lumen is also helping the businesses that consume AI services — retail, banking, health care, manufacturing — to upgrade for today’s speed and security requirements. “AI is driving up the amount of data,” Johnson says. “It’s proliferating, and it’s not like five times or 10 times. It’s often a thousand or more times the amount of data they’ve had in the past. Existing networks are insufficient for the AI economy. And that is the opportunity for Lumen.”
Out-of-This-World Opportunities
Philip Johnston WG18
CEO, Starcloud
During an October 2025 TED Talk about his Seattle-based startup’s efforts to put data centers in space, Philip Johnston pointed out that the idea had been “roundly pilloried from all corners.” He gamely read out a few scathing reactions posted to X, including, “I thought you nerds were supposed to be good at science.”
Johnston readily admits there are challenges to overcome before Starcloud establishes data centers in low Earth orbit (LEO). But he firmly believes that a decade from now, we’ll be building more data centers in space than we are on Earth. Investors are bullish, too: Nvidia is already a backer, and along with $21 million in seed funding, Johnson says, a successful Series A round was just completed.
Meanwhile, the hurdles are technical and logistical. While the sun essentially provides unlimited cheap power, keeping the hardware cool is tricky. Starcloud is designing a radiator to handle the task. Moving all the pieces from Earth into space is another heavy lift. Making the financials work depends on frequent flights of reusable rockets like SpaceX’s Starship, which is still in the testing phase. Still, Johnston says the concept is a “complete no-brainer. … Data centers on Earth consume enormous amounts of power, mainly, but also water. And we are growing the number of data centers at an extremely high rate. There are some challenges with cooling in a vacuum, but it’s definitely doable.”
Last November, Starcloud launched a demonstration in the form of one satellite, about the size of a small refrigerator, with a powerful Nvidia AI chip aboard. (Besides the cooling, designing shielding to protect chips from space’s harsh environment is another challenge.) The company became the first to train a large language model in space — using Shakespeare’s works so that replies to queries are delivered in the playwright’s manner.
Within the next five to 10 years, Starcloud hopes to have a constellation of up to 88,000 small satellites, each weighing around 6,000 pounds, handling AI workloads such as ChatGPT queries and customer-service chatbots. According to Johnston, this would make up around 20 GW of computing capacity. The data center’s solar panels will equal the surface size of about four tennis courts; the radiator will measure about one tennis court.
Meanwhile, just by reaching for space, Johnston says, Starcloud is yielding benefits that will ripple beyond AI, much as NASA’s exploration has led to breakthroughs for the Earth-bound. He cites examples like mining asteroids for minerals and building habitats in LEO: “The core technology we’re developing is a very large, low-cost, low-mass deployable radiator that is extremely useful for basically every other space application.”
Still, he remains focused on the end goal: data centers to power the AI revolution without Earth’s resources. “The exponential curve we’re about to be on is absolutely mind-blowing,” Johnston says.
Janine White is a freelance writer and editor based in Philadelphia.
Published as “Inside the Data Center Boom” in the Spring/Summer 2026 issue of Wharton Magazine.
