The AI Infrastructure Building Boom, Part II: A Commercial Real Estate Reboot

The first article in this series looked at the players behind the AI infrastructure boom. This one looks at the simpler, less technical question underneath the buildout: why did data centers become so attractive so quickly?

Put the AI story aside for a moment. Forget model benchmarks, frontier labs, Nvidia chips, and whether the next generation of AI will justify the spending. Before any of that can be answered, there is a more basic real estate question: what happens when a large part of the building economy needs a new growth category?

That is the useful frame for Part II. Data centers are not only being built because AI companies need compute. They are also being built because developers, investors, lenders, contractors, utilities, equipment suppliers, and local governments all had reasons to want the next large physical project category to arrive.

For years, many familiar development stories had become harder to tell. Office space was damaged by remote and hybrid work. Retail remained important, but new development became more selective and expensive. Multifamily housing faced pressure from interest rates, insurance costs, construction costs, and local opposition. Industrial and logistics real estate remained active, but the pandemic-era warehouse surge cooled. Even film and television soundstages had their own version of the problem after the streaming boom encouraged a wave of capacity that later ran into slower production.

Then data centers became the new bright spot.

They were large. They were expensive. They required land, electricity, cooling, fiber, construction crews, equipment, financing, and government approvals. They could be connected to major technology customers. And most importantly, they came with a story powerful enough to make all that building feel inevitable: artificial intelligence needed a new physical layer.

That does not make the boom fake. It means AI demand arrived at a moment when the real estate and construction industries were looking for something new to finance and build.

The Old Growth Stories Got Harder

The easiest example is office space. Moody’s data reported by Axios showed U.S. office vacancy across 79 markets reaching 21% in the first quarter of 2026, up from 17% in 2020. That is not just a normal cyclical slowdown. It reflects a lasting change in how many companies use office space after remote and hybrid work became part of ordinary business life. (axios.com)

But the broader point is not just “office space is weak.” Colliers’ 2026 commercial real estate outlook described retail as resilient but supply-constrained, with 2026 construction expected to drop 37%, and noted that data centers were surging even as power constraints and community opposition slowed new projects. Invesco’s commercial real estate outlook also pointed to high interest rates, labor constraints, higher construction costs, and reduced construction debt as forces curbing new building activity across traditional property types. (colliers.com)

That matters because real estate is not only a set of buildings. It is an ecosystem. Developers need projects. Lenders need loans. Contractors need work. Architects and engineers need assignments. Cities need tax base. Investors need places to put capital. Utilities need customers. When older growth stories become harder, the machine does not disappear. It looks for the next thing.

Data centers became one of those next things.

Data Centers Offered a Reboot

Traditional real estate is usually organized around people: people at desks, people in stores, people in apartments, people moving goods through warehouses. Data centers are different. They are organized around machines, electricity, cooling, and network access.

That difference made them attractive. A data center could be described less like another commercial real estate bet and more like infrastructure for the AI economy. The building was still real estate, but the pitch was bigger than real estate. This was not just a box on land. It was powered capacity. It was a place where electricity could be turned into compute.

That shift changes the way the asset is understood. In a conventional building, the question is often how much space can be leased and at what rent. In a data center, the question is whether the site can secure enough power, cooling, fiber, permits, and customer demand to support high-density computing. The valuable thing is not merely square footage. It is powered square footage.

That is why the category has become so compelling. Data centers give the building economy a way to attach itself to AI without having to be in the AI business. A contractor does not need to build a model. A developer does not need to design a chip. A county government does not need to understand inference costs. They only need to believe that the AI economy will require more physical infrastructure, and that their project can be part of it.

The Numbers Are No Longer Small

The scale is now visible in national construction data. Data Center Knowledge, using newly broken-out U.S. Census Bureau construction spending data, reported that private data center construction reached a seasonally adjusted annual rate of about $50.7 billion in April 2026. That made data centers the largest segment within the private office construction category, accounting for roughly 52% of that activity. (datacenterknowledge.com)

That classification is worth pausing on. This is not a revival of traditional office towers. It is almost the opposite. A Census category once associated with people sitting at desks is now being carried in large part by buildings designed for servers.

For the construction economy, that is not an abstraction. Data centers require high-voltage electrical work, cooling systems, substations, backup power, security, fiber, steel, concrete, switchgear, transformers, and specialized engineering. JLL’s 2026 Global Data Centre Outlook projected roughly 100 gigawatts of new capacity between 2026 and 2030, equivalent to about $1.2 trillion in real estate asset value creation, with average global construction costs forecast at $11.3 million per megawatt in 2026. (jll.com)

Even if those projections prove too optimistic, they explain why data centers are so attractive to the physical economy. They are not just software infrastructure. They are projects, contracts, materials, labor, equipment orders, and years of work.

Private Capital Followed the Story

The money followed because data centers could be turned into something investors already understand: a large physical asset with long-term customers, scarce inputs, and potentially durable cash flows.

S&P Global Market Intelligence reported that private equity investment in U.S. data centers reached $45.7 billion in 2025, the highest total in at least five years. S&P also reported that private equity accounted for about 72% of the $63.35 billion invested in the U.S. data center space that year. (spglobal.com)

That does not mean every deal is speculative or every investor is overreaching. It means the AI infrastructure boom is not simply a story of Big Tech spending on its own facilities. It is also a story of private capital helping convert AI demand into financeable real estate.

The logic is straightforward enough: secure land, secure power, obtain permits, attach a strong customer, finance the facility, and move to the next project. That model has existed in other forms of infrastructure for a long time. AI gave it a new target.

This is the core of the Part II argument. AI created the demand signal, but the real estate and finance world helped turn that signal into a new category for investment.

The Soundstage Cautionary Tale

The best cautionary tale may come from outside data centers altogether.

Film and television soundstages are not the same as AI infrastructure, but they show how a real demand story can still lead to overbuilding. Streaming was real. Demand for content was real. Production companies did need more space for a time. But the streaming wars also encouraged a wave of investment in studio capacity that later ran into a more disciplined market.

World Property Journal, summarizing FilmLA’s 2026 soundstage report, reported that major Los Angeles-area soundstage occupancy averaged 62% in the first half of 2025, with streaming companies reducing output after years of heavy spending and other markets adding capacity.

That does not prove data centers are headed for the same outcome. The assets are different, the customers are different, and the demand drivers are different. But the warning is useful: a technology or media trend can be real and still produce too much physical capacity if too many participants build around the most optimistic version of the future.

That is the narrow point here. AI can be real, cloud demand can be real, and data center demand can be real, while the physical buildout still risks getting too large, too fast, or too concentrated in the wrong places.

Why This Matters

The AI infrastructure boom is powerful because it gives the real estate and construction industries a new story at a moment when several old stories were harder to sell. Office space had a demand problem. Retail had a growth problem. Multifamily had a cost and politics problem. Industrial and logistics had a normalization problem. Soundstages had an overcapacity problem. Data centers arrived with a future-facing narrative and a customer base that looked capable of supporting enormous investment.

That is why this boom moved so quickly. It was not just that new AI companies needed more compute. It was that many other old institutions needed a new place for money, labor, land, and political attention to go.

The risk is sizing. A real technology can still produce an overbuilt physical layer. The internet was real during the dot-com bubble. Streaming was real when soundstages were overbuilt. Railroads were real when investors financed too many routes ahead of durable demand.

The question is whether today’s AI infrastructure buildout is being sized for long-term need, or whether part of it is being pulled forward because the story works so well for everyone around it.

AI created the demand.

Real estate turned it into a development category.

Capital turned it into an investment category.

And now the physical economy is building around it.

Coming Next

Part III of the series will focus on the timing mismatch at the center of the boom. AI chips, model architectures, and compute economics move quickly. Data centers, substations, transmission upgrades, tax incentives, and land-use decisions move slowly.

The chip runs on one clock.

The concrete runs on another.

That mismatch may be where the AI infrastructure story takes a turn.