How Inference and AI Could Rewrite the Entire Site Selection Playbook for Commercial Real Estate

The old site selection checklist (land price, labor, tax incentives, highway access) is being shoved aside by a single question almost no broker asked five years ago: can this parcel actually get power, and can it move data fast enough to matter? Artificial intelligence, and specifically the shift from training models to running them at scale through inference, is quietly rewriting how the most valuable real estate in America gets chosen. For Chicagoland and southern Wisconsin owners, investors, and developers, the implications reach well beyond the data center pad itself.

Why Inference Changes the Math

For most of the AI boom, the headline demand came from training, the enormously power-hungry process of building models in massive clusters that could sit almost anywhere a transmission line and cheap land happened to meet. Inference, the act of running those trained models to answer real user requests, behaves differently. It has to be close to the people and applications using it, because latency is the product. A response that takes 40 milliseconds feels instant, while one that takes 300 milliseconds feels broken.

The industry consensus is that inference is about to dominate, with analysts anticipating that inference workloads could overtake training as the dominant AI requirement as early as 2027. S&P Global notes that while model training can be carried out a long distance from end users, inference requiring rapid connectivity needs to sit near urban centers. That single change pulls compute back toward population centers, and Chicago is one of the largest population and connectivity hubs in the country. We have written before about how this reshapes asset values in Data Centers vs. Moore’s Law and explored the underground fiber and power layer that decides which facilities get built in our conversation with Bruce Garrison on RFP 104.

The Two Constraints That Now Decide Everything

Site selection in 2026 has narrowed to two gating factors that sit ahead of every traditional real estate metric.

1. Deliverable Power, Not Promised Power

Power has moved from a due diligence item to the primary investment constraint, and across major U.S. markets grid interconnection timelines now stretch from four to ten years. The critical distinction for any owner evaluating a sale or development play is that a “will-serve” letter from a utility does not equal powered land. What matters is a contract for transmission capacity by a date certain. In the ComEd footprint specifically, one study projects power shortfalls beginning by 2029, and ComEd is among the PJM zones with the largest data center pipelines, with its annual energy demand projected to double by 2046.

2. Latency and Fiber Proximity

Inference workloads requiring rapid connectivity need to sit near urban centers, and power alone is no longer enough: sites must also have fiber access and proximity to primary markets. This is where Chicago’s legacy infrastructure becomes a genuine competitive asset. The region is one of the most fiber-dense interconnection points in North America, which means a parcel here can serve low-latency inference to a huge share of the central United States in a way that a remote, power-rich site in another state simply cannot. This same power-as-gating-factor dynamic was the heart of our discussion with Whitaker Irvin Jr. on hydrogen, data centers, and the new math of industrial development (RFP 100).

What This Means for Chicagoland and Southern Wisconsin

The practical effect is a reshuffling of which submarkets matter and why. Powered land near existing substations and fiber routes is becoming the scarce asset. We are watching demand concentrate along the I-88 corridor and DuPage County, the O’Hare and I-90 corridor, Will County’s logistics base, and increasingly across the southern Wisconsin line into Pleasant Prairie, Kenosha, and Racine, where land economics and proximity to both Chicago and Milwaukee make hybrid power strategies viable.

Several forces are converging here at once:

• Roughly one-third of data centers in 2030 are expected to use 100% onsite power, and 73% of operators are actively evaluating or selecting onsite power providers, which makes parcels suited to behind-the-meter generation or battery storage far more valuable.
• PJM has created an expedited “Bring Your Own Generation” track, rewarding sites that can pair load with new firm generation or storage, a dynamic we unpacked in The Battery Real Estate Play with Aaron Shavel (RFP 95).
• MISO forecasts data center power usage in its territory rising from roughly 3% of total load in 2026 to 13% by 2035, a structural shift that will ripple into industrial and commercial rates region-wide.

The Risk Hiding in the Boom

There is a real cost-allocation fight underway. PJM data center demand has already driven sharp household bill increases in some states, and regulators are now exploring special rate structures and cost-sharing mechanisms for large loads. For owners of conventional industrial and office assets, this matters: rising regional power costs and politically sensitive large-load growth can change the economics of an entire submarket. A market with technical capacity but unstable political support may rank lower than a smaller market offering regulatory clarity. Site selection is now as much a regulatory and community-relations exercise as a real estate one.

What Comes Next

The owners and investors who win the next cycle will treat power and latency as the first screen, not the last. Five moves are worth making now:

1. Audit your holdings for proximity to existing substations, high-voltage transmission, and major fiber routes, because those attributes now carry a premium that may not yet be reflected in your basis.
2. Engage utilities early and insist on the distinction between a will-serve letter and a date-certain capacity commitment before underwriting any power-dependent use.
3. Evaluate whether your parcels are candidates for behind-the-meter generation or battery storage, which can unlock value through expedited interconnection paths.
4. Track the regulatory and rate environment in your submarket, because cost-allocation decisions will directly affect operating costs across all asset classes, not just data centers.
5. Reassess older industrial and flex assets near power and fiber, since conversion or assemblage potential may now exceed their current-use value. If you are actively hunting parcels, our commercial space search tools and brokerage, development, and management services are built for exactly this kind of power- and connectivity-driven evaluation.

The site selection playbook is being rewritten in real time, and Chicagoland’s combination of fiber density, transmission infrastructure, and proximity to a massive end-user base positions it to capture an outsized share of the inference economy, provided owners move with the right information. To understand how power, latency, and AI-driven demand affect the value and positioning of your specific property, contact