The data center development boom is headed downtown

Perched atop a famed Art Deco skyscraper in Chicago’s South Loop, a statue of Ceres, the Roman goddess of agriculture, looks upon downtown clutching a sheaf of wheat and a bag of corn. It’s a calling card for the building’s occupant, the Chicago Board of Trade, and a reminder of the importance of these commodities when the building was erected in the early 20th century. 

Across the street, in the former Chicago Board Options Exchange office, a symbol of a new type of commodity is under construction. Legacy Investing is transforming the former trading floor into a 33-megawatt data center, set to open later this year, underscoring the crucial nature of data in the 21st century and the coming boom in new data center development in and around urban America.

The physical manifestation of the economy’s AI frenzy has been gargantuan, gigawatt data center campuses across the rural corners of the nation. Multibillion-dollar developments for Big Tech players like Meta, Amazon, Microsoft, and Alphabet are taking shape in Mississippi, Louisiana, Wyoming, and other places with cheap land and available power. More than 4,000 data center projects are in the pipeline, according to industry policy shop the American Edge Project.

But industry experts note the playing field has shifted and is rapidly growing in urban and suburban parts of the country. Over the next few years, they expect exponential growth in both edge and inference data centers; these are smaller sites located near users in population centers, on the so-called “edge” of the network, that locally process data requests or process and infer for AI apps, helping apps and programs respond faster and with less delay, or latency. These will become a massive real estate play in downtowns, urban industrial areas, and suburban industrial sites.  

A big reason for the focus on population centers is the evolution of computing demand for artificial intelligence. The gigantic data centers currently grabbing headlines — and provoking community pushback — are meant for training large language models and developing the technology. When these systems mature, and when these models get used more frequently in consumer-facing apps and programs, tech platforms will need compute power as close to users as possible to deal with latency issues. Placing data centers on the edge of where consumers are means the digital tools and apps they rely on work faster and more reliably. Chris Sharp, CTO of Digital Realty, a large data center developer, says 2026 will be the year of inversion for AI, where there’s more activity around inference than training. 

Forecasts by industry analysts at the Proptech Connection predict AI inference traffic will grow 25% between 2024 and 2027, just below the 30% growth in AI training. As soon as the industry figures out the next new thing, it’ll require a new build-out. 

Daniel English, managing investor of Legacy Investing, predicts 10x growth of AI inference and edge data centers between now and 2030, with new sites sprouting up in every major city with an NFL team. New York City could be a flagship for data centers given its proximity to financial firms, but the phenomenon is widespread.

Thomas Bailey, vice president of energy for Flexential, a data center operator and developer with 40-plus sites nationwide, predicts 12.5 gigawatts’ worth of new edge sites, a measure of computing power that roughly translates to building about 300 new 50-acre sites across the country. Ivo Van Breukelen, managing partner at Chicago-based advisory firm The Proptech Connection, sees the edge market growing from roughly $20 billion a year to more than $100 billion by 2030.

“Technology doesn’t go backwards, it goes forward,” English said. “AI is not going backwards.”

Having more inference centers in place will make it easier for ChatGPT to produce and deliver photos more quickly, or help chatbots, often teased for their long pauses, respond more rapidly when asked questions. 

“When you think about the trillions of dollars that are being built on these training models, the edge compute that incorporates inference AI is the way we’re going to pay for all of it,” said Pat Lynch, executive managing director of data center solutions for CBRE, a massive international real estate services firm. “That’s going to create the commercial applications.”

Factor in increased demand from streaming and medical imaging, as well as growing computing need from everything from big banks and research labs to autonomous vehicles, automated drone delivery, and robots, and that adds up to a new generation of smaller sites.

“As these applications are becoming more and more data intensive, latency becomes a bigger issue and network congestion becomes a bigger issue,” said Christopher Brown, chief technical officer of the Uptime Institute, a data center business group. “These companies have been moving some of their IT assets closer and closer to their end users for a while.”

These urban and suburban sites can come in a variety of configurations, typically within a 60-mile radius of an urban center and potentially located near research parks, hospitals, and universities. Bailey said Flexential is working on a series of edge and inference sites taking up 15 to 30 acres — likely in formerly used industrial areas, called brownfield sites — that can quickly be turned into mini data center campuses and already have the power hookups to feed rows and rows of servers. 

Alternately, smaller and more efficient chips, servers, and racks mean that small data centers can be set up in office buildings of 80,000 square feet; there’s even an opportunity to create what’s called a mesh network of small servers installed in, say, a series of parking garages. Equipment providers like Schneider Electric and Vertiv Holdings have been working for years to design new equipment for these smaller sites, Brown said.

These developments could have significant impacts on downtown real estate value, said Van Breukelen. Owners with vacant or obsolete office stock could reposition them into hybrid tech campuses, with inference hubs as anchor tenants. Not only would these inference centers command comparable rents on a per-square-foot basis in some metro markets, but their presence could attract a cluster of AI companies, life science companies, and other high-value tenants seeking compute and proximity. 

“This trend is likely to accelerate as we see further movement in the reemergence of enterprise data centers in sensitive and regulated businesses such as government and finance, as they pull sensitive data and AI tools into proprietary setups in-house,” Van Breukelen added.

This expansion will likely meet community pushback amid rising power costs attributed to data center demand. Climate change publication Heatmap discovered that 25 data center developments had been canceled in 2025, 4x the rate of cancellations the year before, and a patchwork of state and local regulations have popped up to regulate or even block data center construction.

Industry leaders are already working with community leaders, looking to tout the economic benefits of data centers, and selling the idea that these sites can, with proper noise abatement, blend into the urban landscape. Uptime’s Brown thinks the battle over regulations, zoning, and approvals will be fought on state and local levels, and “it’s going to be different in every place.” 

“There’s certainly a narrative we have to overcome,” Flexential’s Bailey added. “It’s a belief that every data center is equal and homogenous, which isn’t right.” 

Legacy’s English takes a different tack. He argues the pushback against rural data centers comes in part from the scale as well as the aesthetics: pristine farmland ruined with concrete boxes and power lines. He sees these smaller urban sites — utilizing existing building shells and power connections — as having a much smaller impact. He’d argue another Legacy project, a data center within a glass office building in downtown Minneapolis, blends in with the (man-made) landscape.

“This is major transformational infrastructure that we are putting into cities for the first time,” he said. “We’re trying to figure out how to make that happen. I firmly believe there’s an opportunity to do data centers that people like.”

Update (March 4, 2026): Corrected misspelling of Ivo Van Breukelen’s name.