How AI’s Growth Is Reshaping the Industrial Market

This is the third in a four-part series of reports by CBRE Research on the impact of artificial intelligence on commercial real estate. This report explores how the AI boom is driving industrial demand for space near data centers and transforming how warehouses are designed and used.

• AI is reshaping industrial demand in two ways. The proliferation of data centers and the rise of AI warehouse automation are changing what occupiers need. Although building quality and location remain the top considerations for leasing decisions, power capacity also ranks high.
• Companies supporting data centers are playing a location game. Companies that build, supply and service data centers want to be close to data center operating clusters (11 or more facilities within about two miles). Within these clusters, tech tenants sign nearly twice as many leases as they do elsewhere (13.9% vs. 7.1%) and specialized R&D tenants nearly double their contribution (35.0% vs. 19.0%).
• The cluster premium is stable, not a temporary trend. Industrial space near 11-facility data center clusters commands rents of about 6.8% more than comparable space elsewhere, rising to 9.8% near clusters of 21 or more data centers. Almost no new supply is being built inside these areas, so this rent premium is durable. Industrial leasing volume inside data center clusters grew by 17.0% year-over-year in the first half of 2026, nearly the same pace as outside these clusters (17.4%). This shows that while a cluster location commands a rent premium for industrial space, it does not result in higher leasing volume.
• Cluster rent premiums aren’t consistent. Rent premiums are concentrated in specialized R&D buildings in established hubs for the servicing of existing data centers. For owners and investors, the opportunity is concentrated in a handful of markets. Northern Virginia and Silicon Valley have the most industrial space inside clusters, while Phoenix, Boston and Milwaukee are the likeliest places for new clusters to form.
• AI automation is raising warehouse roofs. Automation, robotics and AI-guided operations need high ceiling heights, robust electrical capacity and layouts built for machines. Warehouses with ceilings of 40 feet or more have led leasing volume growth every year since 2022, adding 67.9 million sq. ft. in 2025. Buildings with ceiling heights of less than 32 feet had more than 14.1 million sq. ft. of leased space returned over the same year.

A cluster is defined as 11 or more operating data centers within about two miles of one another.

Figure 2: Silicon Valley Data Center Rent Premium for Warehouse Space

Industrial space inside clusters commands rent that is about 6.8% higher than that for identical space outside them. Space near clusters of 21 or more data centers rents for 9.8% more, while space near 10 or fewer does not command any rent premium. Leasing volume inside clusters in the first half of 2026 grew at nearly the same rate as overall leasing volume within a specific market (17.0% vs. 17.4%).

Figure 3: Rent Premiums for Warehouse Space Within Data Center Clusters

The premium also varies by the type of data center—everything from a server room in an existing industrial building to a gigawatt campus. Approximately 3,100 colocation facilities operate across the U.S. These are multi-tenant buildings that rent space, power and cooling to enterprise customers, typically located in established industrial districts close to fiber and their customer base. There also are about 240 enterprise facilities, each of which are owned and operated by a single company for its own use, and 200 cloud facilities that house the servers, data storage systems and networking equipment to run cloud computing services. Hyperscale campuses, purpose-built for a single cloud or AI operator, number 119 and are usually located on large metro-edge parcels near bulk transmission lines. By facility count, small and mid-sized data centers outnumber hyperscale by about 30 to one.

Figure 4: Operating Data Centers by Type

Clusters containing only colocation facilities command a rent premium of around 6.0% because they usually are located within industrial districts. Conversely, hyperscale campuses are generally located on metro edges with few industrial buildings around them to measure a premium.

The premium also varies by property type within a cluster. It is highest for specialized R&D space at about 14% above comparable space outside clusters, and lowest for distribution space at about 5%. Power is part of the reason. According to the U.S. Energy Information Administration, specialized R&D buildings use about six times as much electricity per sq. ft. as non-refrigerated warehouses, while buildings used by computer and electronics manufacturers use about 12 times as much.

Figure 5: Rent Premium Inside Data Center Clusters by Space Type

The occupier base inside clusters differs from the one outside. Manufacturers lease 34% of the space inside clusters, compared with 13% outside of them, while third-party logistics providers run the other way (19% vs. 36%). Within manufacturing, half of this sector’s in-cluster space is used for tech and data center components, compared with 20.5% of space elsewhere. These occupiers include semiconductor equipment, photonics, chip cooling and server assembly firms. Logistics automation accounts for 9.0% of in-cluster space vs. 3.2% elsewhere, while pharmaceuticals account for 5.9% vs. 1.7%

Figure 6: Share of Leased Manufacturing Space Inside & Outside of Data Center Clusters

Why it matters: A 6.8% rent premium is consequential enough to matter financially and plan around. Because it is a stable price level, tenants can budget for rent and owners can underwrite it with confidence. But the premium is not spread evenly. The best opportunity is not a parcel next to the newest hyperscale campus. It is for specialized R&D space in older, established data center hubs, where a deep pool of existing buildings, proven power capacity and almost no new competing supply keep the premium durable.

Most of today's data center clusters are made up of colocation facilities in established industrial neighborhoods.

However, a wave of new hyperscale campuses currently under construction could change that pattern. Nearly half of all data centers being built are hyperscale campuses, compared with just 3.2% of existing data centers. As a result, the number of data center clusters nationally could grow by roughly 40%. But most of these new hyperscale center clusters will have little to no existing warehouse space nearby.

Phoenix, Boston and Milwaukee are the exceptions. The new clusters forming there have a sizable amount of existing warehouse space, making them the clearest emerging opportunities to watch. More conservative estimates, based only on sites that developers have already secured, point to a much smaller and nearer-term number of new clusters.

Figure 7: Operating & Planned Data Centers by Type

Figure 8: Industrial Completions in 2024 & 2025 as a Share of Existing Inventory

Geographically, opportunities for industrial space owners remain concentrated. Northern Virginia stands out with nearly 44% of its total industrial space located inside of data center clusters. Silicon Valley follows with about 34%, compared with less than 3% nationally. Salt Lake City and Orange County, CA have smaller but still notable concentrations. Some markets with many data centers, such as Los Angeles and Seattle, have very little of their total industrial space inventory located within data center clusters.

Why it matters: Because this rent premium depends on proven power capacity and because new supply inside these areas remains minimal, the safer, more durable play is existing space in established hubs, not land next to a single newly announced project. Northern Virginia and Silicon Valley provide most of these opportunities. Phoenix, Boston and Milwaukee are the most likely markets for new clusters, while Atlanta, Chicago, Dallas-Fort Worth and Columbus also have enough project activity to be worth considering as well.

Beyond generating rent premiums for industrial space near data centers, AI is also changing what happens inside of a warehouse.

Warehouse automation, AI-guided inventory picking, autonomous forklifts and mobile robots need a very specific kind of building: high ceilings for racking and robotics to operate, ample electrical capacity for charging equipment and running computing systems, floor layouts designed for machines to move efficiently and sufficient power capacity.

Buildings with ceiling heights of 40 feet or more have led all annual warehouse demand growth since 2022, totaling 67.9 million sq. ft. in 2025 alone. Meanwhile, the number of warehouses with ceilings below 32 feet has been falling since 2023—down by 14.1 million sq. ft. in 2025—due to demolitions, conversions to other uses or simply failing to attract new tenants.

Figure 9: Net Absorption by Clear Height of Industrial Buildings

Approximately 134,000 of the 233,000 industrial buildings we track are more than 40 years old. When one of them has space available, it has roughly a 46% chance of going three years without a new lease, compared with 26% for a building completed in the past five years. Buildings with clear heights of less than 24 feet are nearly twice as likely as 32- to 36-foot clear-height buildings to wait three years for a tenant. However, there is one contradiction: Although buildings inside data center clusters tend to be older and smaller than the national average, they still command a rent premium. This shows that an old, low-ceiling building in the right location, with power and fiber already in place, can still perform better than its physical characteristics alone would suggest.

Why it matters: AI is limiting the viability of older, lower-quality warehouse space in two ways: bidding up demand for modern, power-rich buildings, while making the oldest and lowest-ceiling buildings progressively harder to lease. The exception is where location, particularly proximity to a proven power source, offsets an older building’s disadvantages. The widening gap between these two groups of buildings is where the sector’s biggest risks and opportunities are now concentrated.

Occupiers

• Recognize that a nearby data center cluster is an important site-selection factor. Although warehouse space near a cluster generally costs more, the rent premium indicates something valuable: proof that the power is actually available there, rather than planned for what could be many years ahead.
• Expect the highest cost for specialized R&D space at roughly 14% above market. For standard distribution or logistics space, the added cost is much smaller (about 5%).
• If automation, robotics or AI-driven operations are planned, start taking space now in buildings with high ceilings (40 feet or more) and strong electrical capacity. This kind of space is tightening every year, so don’t wait and risk paying more or settling for less.
• Try to lock in space early, ideally while nearby data center projects are still in the planning stages. Once a center is operational, landlords gain more of an advantage in lease negotiations.

Investors

• Underwriting of warehouse facilities near data centers should include the rent premiums outlined in this report.
• When evaluating a deal, price the rent premium at an average of 6.8%, a one-time step up in what space is worth and not a compound growth rate. Focus on established data center clusters where enough leasing activity exists to trust the numbers, rather than newer or thinner clusters where the numbers are less reliable.
• Prioritize specialized flex and R&D buildings in older, established data center hubs over land near a single, newly announced hyperscale campus. These established hubs combine a deep supply of existing buildings with almost no new competing construction, which is exactly what makes the pricing premium durable.
• Over the long term, look for opportunities in markets with the most new data center project announcements. Give greater weight to projects where developers already have formal site control.
• Favor newer buildings with tall ceilings and proven power capacity. Underwrite the risk for older, low-ceiling buildings that are harder to lease unless that building is within an established data center cluster.

The findings in this report are based on taking rents of CBRE-tracked industrial leases across the top 25 U.S. data center markets, 23 of which have leases inside of data center clusters.

The rent averages are based on 7,372 leases signed from January 2025 through mid-July 2026, 436 of them inside a cluster, across 174 submarkets.

Rent premiums are modeled on a building and lease constant: subtype, class, size, age, clear height, dock doors, site area, leased area and lease type, together with distance to fiber, grid and substation capacity, the interconnection queue, industrial electricity prices, local income and employment, freight access and site characteristics, and comparing buildings within the same submarket in the same year.

A cluster is defined as 11 or more data center facilities operating within two miles of one another in the group. Data center inventory and pipeline figures reflect LandGate data as tracked by CBRE Research.

External sources cited for market context: U.S. Department of Energy and Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report; Lawrence Berkeley National Laboratory, Queued Up: 2026 Edition; U.S. Energy Information Administration, Short-Term Energy Outlook and Electricity Monthly Update, mid-2026; U.S. Census Bureau, Monthly Construction Spending, June 2026 release.