North American data centre vacancy sits at 1%, but little of it can host high-density AI racks
Empty floor space is not the same as usable capacity when racks need 100 kW or more and liquid cooling that legacy halls were never built for.
Data Center Knowledge reported on 18 August 2026 that North American data centre vacancy is holding at 1 percent for the third consecutive year, yet according to JLL only a very small share of that capacity can support high-density AI deployments. Andrew Batson of JLL said most vacancies are in legacy products built for earlier generations of servers.
The gap between available and usable comes down to rack density. Facilities designed over the past decade average roughly 20 kW per rack. An Uptime Institute survey cited in the article found that 82 percent of facilities have a maximum rack density below 30 kW, only 9 percent report their highest-density racks at 50 kW or above, and fewer than 1 percent of operators run 100 kW racks as standard. Fewer than 4 percent of US data centres can support that level at all. Daniel Bizio of Uptime Intelligence said that above 50 kW it becomes increasingly impractical to cool with air because of space limits.
The article describes a tenant seeking 5 MW at 140 to 160 kW per rack that approached more than ten North American operators. Nearly all had space; few could provide the liquid cooling and electrical capacity required. Carl Beardsley of JLL Capital Markets added that tenant creditworthiness affects how AI projects are structured.
More than 66 GW is under construction across North America, 77 percent of it in frontier markets outside the traditional hubs, raising the question of whether new build is specified for the densities AI tenants now request.
Why it matters
Vacancy rate has been the industry's headline metric for decades, and this report suggests it no longer tells buyers what they need to know. Power delivery per rack and the presence of liquid cooling are becoming the real constraints, and both are expensive to retrofit. Expect leasing and site selection to be described in kilowatts per rack rather than square metres.
Vacancy rate has been the industry's headline metric for decades, and this report suggests it no longer tells buyers what they need to know.
- Facilities whose maximum rack density is below 30 kW82%
- Facilities reporting highest-density racks at 50 kW or above9%
- US data centres able to support 100 kW racks<4%
- Operators running 100 kW+ racks as standard<1%
Figures: Uptime Institute survey figures as reported by Data Center Knowledge
What you can learn from this
- Rack density is a power figure, and power turns into heat. A rack rated at 20 kW draws up to 20 kilowatts of electricity, and almost all of it leaves as heat that must be removed from the same small footprint. Doubling density does not just need thicker cables; it needs the cooling system to carry twice the thermal load through the same volume of air or fluid. Electrical distribution, floor loading and fire suppression all scale with that figure too. This is why density, not floor area, has become the binding constraint for AI hardware.
- Air cooling has a physical ceiling. Air is a poor carrier of heat compared with water, so removing more heat with air means moving more of it, faster, through larger ducts and fans. Past a certain point the fans, plenums and containment aisles take up more room than the servers, which is the space limitation Bizio describes. Direct liquid cooling puts a cold plate on the chip and pipes coolant to it, moving far more heat per litre than air ever could, at the cost of pumps, manifolds and leak detection that an air-cooled hall never needed.
- Legacy facilities were designed for a different load profile. A data hall built around a 20 kW average has switchgear, busways, floor loading and cooling sized for that figure. Raising it to 100 kW or more means new electrical distribution, reinforced floors for heavier racks and a plumbing network that never existed. Many operators find the cost approaches a new build, which is why vacant legacy space stays vacant even in a market with 1 percent availability.
- Capacity under construction is a pipeline, not an inventory. A gigawatt announced today may not be energised for several years, and it can be delayed by grid connection queues, transformer lead times and permitting. Buyers with immediate needs are therefore competing for a much smaller pool than the headline pipeline implies. Reading construction numbers alongside energisation dates, and noting which projects are in established hubs versus new markets, gives a more honest picture of supply.
- Tenant credit shapes what gets built. Because a high-density facility is expensive and specialised, lenders and developers want confidence the tenant will pay for the full lease term. A tenant with strong credit can secure purpose-built capacity; a weaker one may only be offered existing space. This links the physical constraints of cooling and power to the financial structure of the deal, and it explains why the same request can get very different answers from the same operator.
We teach this
How to use this in practice
- Calculate the thermal load of a rack you can see. Pick any server rack you have access to, or a spec sheet online, and add up the rated power of its components in watts. Treat that figure as the heat the rack emits at full load, then convert it to the airflow required using the airflow-per-kilowatt guidance in any major cooling vendor's published design guide. Done is a single page showing the rack's kW, the air volume it demands and how many such racks a typical room air handler could serve, which makes the 50 kW ceiling tangible.
- Draw the two cooling paths side by side. Sketch an air-cooled rack with hot and cold aisles, fans and a room air handler, then a liquid-cooled rack with cold plates, a manifold, a coolant distribution unit and a heat exchanger. Label where heat enters and leaves each system and which components would have to be added to an existing hall. Done is a diagram that shows why liquid cooling needs plumbing that legacy halls lack.
- Build a small capacity checklist. Write a one-page questionnaire you would send to a colocation provider that asks for maximum kW per rack, cooling type, available power at the distribution unit, floor loading and energisation date rather than square footage. Add a column for the answer you would consider disqualifying for a 100 kW deployment. Done is a checklist you could use to filter a shortlist in an afternoon.
- Model the density gap in a spreadsheet. Put the Uptime survey shares reported in the article into a sheet, add a column for a hypothetical tenant needing 140 kW racks, and calculate how many of 100 typical facilities could host it. Then add a second scenario at 50 kW to see how much the answer changes. Done is a chart showing the mismatch between where capacity sits and where demand is heading.
Sources
- For High-Density AI, Available Data Center Space May Not Be Usable — Data Center Knowledge
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report