Can a Warehouse or Old Factory Become an AI Data Center?

Short answer: sometimes. A warehouse already has two things a developer cannot manufacture quickly — a large, enclosed, single-storey industrial building, and an established industrial location. What it usually lacks is the power capacity, structural loading, or cooling infrastructure an AI facility needs. Whether conversion makes sense is decided by a short list of testable criteria, and the tests are cheaper than the mistake of skipping them.

Context matters. With North American vacancy at a historic low of 1.4% at the end of 2025 (CBRE) and new construction timelines extended by permitting, zoning approvals and power sourcing, existing buildings are an obvious place to look for capacity. But an existing building is only an advantage if it passes the same screens a greenfield site must pass — plus several of its own.

Screen 1: Structure and floor loading

AI hardware is heavy and getting denser. Berkeley Lab's 2024 energy usage report documents that U.S. data center power demand more than doubled between 2017 and 2023, largely driven by GPU-accelerated AI servers — and that density arrives in the building as physical weight: racks, battery systems, cooling plant and electrical gear, much of it concentrated rather than spread like warehouse racking and pallets.

The screening questions are basic and non-negotiable. What is the slab's design load, and what does a structural engineer — not a brochure — say it can carry in the configuration the design requires? Can the roof structure carry new cooling or electrical equipment, or is there ground space for it? Are column spacings workable for data hall layouts and airflow? DLR Group's adaptive reuse guidance makes the general point plainly: data centers require structural integrity beyond typical office or warehouse space, and a structural engineer's assessment is what establishes whether the shell can support the operational intensity of the use. Buildings that fail here are not necessarily worthless — slabs can be reinforced and loads redistributed — but reinforcement cost belongs in the first estimate, not the third revision.

One category deserves a note: heavy legacy industrial buildings — former manufacturing plants and power-adjacent facilities — often start ahead of modern logistics warehouses on this screen, because they were built for heavy machinery and frequently retain substantial electrical infrastructure. That is a reason to screen them early, not to skip the engineering.

Screen 2: Clear height

Data halls need vertical room: for the racks themselves, for overhead power and network distribution, and for air management or liquid-cooling distribution above or below the equipment. Warehouses are often assumed to pass this test automatically because logistics buildings are tall — but what matters is clear height after existing services, roof structure and any new containment are accounted for, which can be much less than the advertised eave height, particularly in older factories with low trusses or mezzanines.

Practitioner guidance gives a sense of scale. DLR Group's conversion checklist treats about 15 feet of ceiling height as ideal and 12 feet as a minimum, noting that water-cooled (liquid-cooled) racks can allow lower ceilings. Treat those figures as one design firm's working guidance rather than a code requirement — the actual requirement comes from the specific equipment and cooling design — but the principle is firm: measure the real clear height, test-fit a data hall layout against it, and do it before valuing the building.

Screen 3: Power proximity and capacity

This is the screen that decides most conversions, and it is the same constraint that governs greenfield development. An ordinary warehouse typically operates with electrical service sized for lighting, HVAC, forklifts and light machinery. A data center needs service an order of magnitude larger, delivered with redundancy. DLR Group's checklist starts, correctly, with power availability: whether the existing utility infrastructure can support the megawatt capacity the project requires, whether dual feeds are possible, and how backup generation will be accommodated.

So the conversion question is never "does the building have power?" It is "what capacity can the serving utility deliver to this address, in what timeframe, and under what classification — planned, applied, committed or operational?" A building next to a substation with documented spare capacity and existing high-voltage service — as some former industrial sites genuinely have — can be years ahead of a greenfield parcel. A building whose utility must build miles of new line to serve it has a warehouse's problems and a greenfield's timeline, and the conversion premium evaporates.

Screen 4: Cooling and water

Every watt of IT load becomes heat that must leave the building. AI densities push designs toward liquid cooling, and the conversion building must physically accommodate the chosen system: space for cooling plant, routes for piping or air handling, and roof or yard capacity for heat rejection equipment. Gensler's analysis of building conversions flags exactly this: older buildings were not designed for high-density IT loads and often require major electrical and HVAC replacement, and their power and cooling limits are the primary constraint on what they can host.

Water is the second half of the screen. Depending on the design, a facility may require substantial make-up water for evaporative systems, or very little with closed-loop and air-cooled designs. The screen checks the site's actual supply position against the chosen design — municipal capacity, allocation limits, discharge arrangements — because a conversion that works thermally on paper but cannot secure water in practice does not work.

Screen 5: Fiber

A data center without diverse, high-capacity fiber is a building full of computers that cannot do their job. The screen establishes which carriers serve or pass the site, whether physically diverse routes exist or can be built, and what that build would cost and take. Industrial locations are mixed on this test: sites near rail and highway corridors often sit on long-haul fiber routes, while otherwise attractive plants can be surprisingly isolated. This must be verified with carriers, not inferred from a map.

Screen 6: Zoning and permits

Industrial zoning is a genuine head start: a warehouse or factory usually sits in a district where heavy, 24-hour, high-utility use is contemplated. But "industrial" is not the same as "data center permitted." The screen checks whether the local code names data centers, whether the use is by-right or conditional, what conditions (noise from generators and cooling plant, water use, height, screening) attach, and whether the jurisdiction has adopted data-center-specific rules. Gensler adds a caution for older and historic buildings: preservation rules can restrict exactly the structural and rooftop modifications a conversion requires.

When conversion wins — and when greenfield wins

Conversion tends to win when most of the following hold: the structure is heavy industrial stock with verified loading capacity; clear height passes with the intended cooling design; the site already has, or can verifiably obtain on schedule, utility capacity a greenfield competitor cannot match; zoning is industrial with a documented data center path; and speed matters enough that an existing shell is worth a design compromise. Gensler's analysis adds a scale caution: the power and cooling available in most existing buildings suits smaller facilities far better than hyperscale ones — it contrasts edge-scale facilities of roughly 1–20 MW with hyperscale campuses above 100 MW. Many realistic conversions are therefore mid-scale projects, and should be underwritten as such.

Greenfield tends to win when the project needs very high density at large scale with a layout optimised from a blank sheet; when the structure cannot carry the loads and reinforcement approaches rebuild cost; when power is not actually available at the building's address, so the conversion inherits the full utility timeline anyway; or when constraints force so much bespoke engineering that, as Gensler puts it, retrofitting ends up more expensive than new construction. The honest comparison is never "conversion is cheaper" — it is total cost and time to operational capacity, at the required scale, with evidence at each step.

The sequence

Screen the building the way you would screen any site — power evidence first, classified by state (planned / applied / committed / operational); then structure and clear height with an engineer's input; cooling and water against a defined design; fiber verified with carriers; zoning path documented. A warehouse or old factory that passes those screens is a real asset in a market where operational capacity is scarce. One that fails them is a warehouse — and the correct outcome of the screen is to price it as one.

Sources

1. DLR Group, "Data Center Adaptive Reuse: 5 Strategies for Existing Buildings," 2025. Practitioner checklist: power availability first (can existing utility infrastructure support the required MW; dual feeds; backup generation); structural load capacity and ceiling height (about 15 ft ideal, 12 ft minimum; liquid-cooled racks may allow lower ceilings); cooling planning; code, fire protection, security and zoning compliance. https://www.dlrgroup.com/idea/data-center-adaptive-reuse-strategies/ — Evidence grade: B (professional design-firm guidance).
2. Gensler, "Historic Buildings: A New Solution for Modern Data Centers," 2025. Older buildings' power and cooling limitations; edge facilities (~1–20 MW) versus hyperscale (100+ MW); retrofitting can be more expensive than new construction due to floor load limitations, ceiling height constraints and seismic reinforcement; historic preservation constraints. https://WWW.GENSLER.COM/blog/historic-buildings-a-solution-for-modern-data-centers — Evidence grade: B (professional design-firm analysis).
3. CBRE, North America Data Center Trends (H2 2025), press release "Fast-Growing North American Data Center Market Set Records in 2025," 2026. Vacancy 1.4% (historic low) at end of 2025; construction activity declined for the first time since 2020 due to extended timelines tied to permitting, zoning approvals and sourcing adequate power. https://www.cbre.com/press-releases/fast-growing-north-american-data-center-market-set-records-in-2025 — Evidence grade: B.
4. Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report, December 2024. U.S. data center power demand more than doubled between 2017 and 2023, largely due to GPU-accelerated AI servers; AI requires increasingly powerful chips and intense cooling systems. Berkeley Lab News Center, January 15, 2025. https://newscenter.lbl.gov/2025/01/15/berkeley-lab-report-evaluates-increase-in-electricity-demand-from-data-centers/ — Evidence grade: A.