Old Factories, Empty Offices Help Fuel Midwest Data Center Building Boom

The world's largest tech companies have led a surge of data center construction across the Midwest with huge campuses for artificial intelligence. But the region's next wave of build-out is expected to involve smaller facilities, with developers seeing opportunities to repurpose defunct industrial sites and languishing office space as AI processing centers. 

From the suburbs of Chicago and Columbus, Ohio, to rural counties in Wisconsin and Iowa, tech giants like Amazon, Microsoft and Meta are building gigawatt-scale data center campuses to train the AI models on which Big Tech has pinned its future. 

While public attention has largely concentrated on hyperscale projects, prominent voices in the data center industry say there is an emerging wave of development across the region for inference: smaller facilities housing the computing power necessary for individuals and businesses to utilize AI.

As AI data center demand shifts towards inference, developers are taking advantage of infrastructure built in the Midwest’s industrial past. The region has many former manufacturing facilities with access to large blocks of electricity they no longer use, as well as power infrastructure built by utilities for factories that never materialized as U.S. manufacturing declined. 

These blocks of “stranded power” — both in former industrial hubs and in urban cores in the form of unused office buildings — are often the right size for inference data centers. This trend will represent a larger share of development across the region in the months ahead, industry leaders said last month at Bisnow’s DICE: Midwest event at the Renaissance Chicago O’Hare Suites Hotel.

“You better believe that those types of opportunities are going to be emerging in the Chicago area and places around the Rust Belt,” said Bill Kleyman, CEO and co-founder of AI cloud platform Apolo.us. 

“I have partners right now that are going after factories, stranded power. Some slight modernization, and all of a sudden they have a facility that can support 40 to 60 megawatts — too small for a hyperscaler but just right to do that next level of inference compute.”  

Some hyperscalers have been taking advantage of stranded power and existing industrial infrastructure for their large-scale projects in the Midwest. 

Last month, Microsoft announced it is beginning the second phase of a 900 MW data center complex on the site of a failed Foxconn manufacturing facility in Mount Pleasant, Wisconsin, using power and other infrastructure developed for the Chinese firm. In Lordstown, Ohio, SoftBank acquired an existing Foxconn factory that it plans to convert into an AI training data center through its Stargate partnership with OpenAI.  

“A lot of factories moved out, and that power then was left there, and data centers are able to take advantage, to take that power and put it back to use,” Verrus CEO Nelson Abramson said. 

The region's next phase of data center growth is expected to revolve around inference AI computing, as executives say that segment is in the early stage of its growth. While it is only a fraction of the AI computing deployed today, inference is expected to account for the majority of AI demand by the end of the decade. 

Instead of the massive, centralized clusters of computing power that have been developed for AI training, inference will often be deployed in smaller facilities — with capacity measured in dozens of megawatts instead of hundreds — that are near major metro areas or specific end users.

With an anticipated flood of demand for these “edge” data centers on the horizon, Kleyman said he expects industrial sites with stranded power will emerge as prime opportunities for developers to deliver capacity quickly. While a shuttered factory with 40 MW of power wouldn’t necessarily be viable as a hyperscale training site, it would be in the range of what end users are seeking for inference deployments. 

“The focus certainly is on the big hyperscalers. … That's what is going to make the headlines, but what I'm seeing are those acquisitions of facilities that are perfect for edge AI, edge inferencing,” Kleyman said. 

These sites may also present fewer hurdles for developers compared to greenfield sites. 

Local opposition to data center projects has emerged as a major source of risk for developers as communities balk at the build-out of large-scale industrial assets nearby. But residents and officials in communities with shuttered industrial properties are often eager to return those properties to active use, both to eliminate blight and to restore tax revenue. 

“They want these facilities there,” Kleyman said. “We’ve found a more amicable community in some of those Midwest locations, because they have a deeper connection with infrastructure and how these facilities are being deployed.”

Former industrial facilities aren't the only sites where developers are turning stranded power into data center capacity, said Daniel English, president and co-founder of Legacy Investing.

Downtown financial districts in cities like Chicago and Minneapolis are increasingly offering opportunities to turn sparsely occupied office buildings with unused pockets of power into inference processing centers that can serve specialized workloads for nearby corporate customers.

English said his firm converted a downtown office tower in Minneapolis into an AI inference data center and is pursuing a similar project in Chicago. Last month, startup developer HydraVault announced what it is calling Chicago’s first downtown AI data center — a 20 MW office conversion expected to break ground later this year. 

Local governments are embracing these projects, English said, pointing to his firm’s Minneapolis office conversion, a project that received support from a city eager to counteract declining tax revenue from downtown office real estate.

“The city loved it because we dramatically increased the tax base,” English said. “We used power that was stranded, we took a building that was mostly vacant and reenergized it in the urban core.”