JLL reports that while North American data center vacancy remains at 1%, only a minimal fraction of existing inventory can support high-density AI deployments requiring heavy power and cooling upgrades.

North America’s data center vacancy rate remains at 1%, yet JLL notes that only a minimal fraction of this available space can accommodate high-density AI deployments.

A buyer seeking 5 MW of capacity recently contacted over 10 North American operators for space capable of supporting 140–160 kW per compute rack. While nearly every operator had vacant space, almost none could meet the required density using direct liquid cooling, according to James Mercer, principal at Metro Colo Advisory, who led the search on behalf of a client. Mercer agreed to be quoted on the record but declined to name the operators, citing his ongoing business relationships with them.

This search underscores a critical distinction between nominal data center vacancy and capacity actually suited for the highest-density AI workloads. JLL’s regional brokerage teams track available capacity by rack density, according to Andrew Batson, head of the firm’s data center research. Only a “very small percent” of the 1% vacancy can support high-density deployments like the one Mercer was sourcing, Batson told Data Center Knowledge. He emphasized that JLL considers liquid-cooling infrastructure essential for AI deployments at these power levels.

According to JLL’s Midyear 2026 North America Data Center Report, the regional vacancy rate has held steady at 1% for three consecutive years. Available inventory consists of small, fragmented blocks, and most new leases are securing space scheduled for delivery in 2028. JLL reported that more than 66 GW of capacity is currently under construction across North America, with 77% located in frontier markets. For an AI buyer targeting 140–160 kW per rack, however, the viable market is far narrower than the headline vacancy rate implies.

“Most vacancies are in legacy products,” Batson said. While JLL anticipates additional AI-ready capacity coming online over a multiyear horizon, he noted that the gap between nominal vacancy and AI-ready space remains unchanged.

Uptime Institute’s 2025 Global Data Center Survey revealed that 82% of respondents operate facilities with a maximum rack density below 30 kW, while only 9% reported racks reaching 50 kW or higher. Although Uptime identified a handful of cabinets exceeding 100 kW, it noted that such configurations remain rare. “Probably a low single digit (under 4%) of data centers in the US can currently support these racks,” said Daniel Bizo, research director at Uptime Intelligence, the market research arm of the Uptime Institute. Bizo clarified to Data Center Knowledge that this figure is an inference drawn from survey data and should be treated as indicative given sample size limitations. He added that Uptime’s surveys indicate fewer than 1% of operators run 100 kW+ racks as a standard configuration, and fewer than one in ten US operators could likely support even a limited number of such racks, such as a single row.

“With AI computing for training still proliferating, 100 kW+ racks are becoming more common,” Bizo said. This places Mercer’s 140–160 kW requirement at the extreme upper end of current market capabilities.

The 30-kW threshold reflects both the statistical distribution of existing rack densities and the evolving hardware and rack configurations required for denser deployments, Bizo explained. Most facilities designed over the past decade were engineered around an average rack power of approximately 20 kW, even if that capacity was not fully utilized upon occupancy. Such sites can typically accommodate 30–40 kW racks in limited quantities through relatively straightforward upgrades, he noted. “Above 50 kW, it becomes increasingly impractical to cool with air due to space limitations in the IT chassis or rack, mandating direct liquid cooling,” Bizo said.

At scale, higher-density racks introduce significant power-distribution challenges, requiring larger and heavier busways, PDUs, power cables, breakers, and potentially additional circuits, Bizo noted. These demands can constrain electrical room size and layout. Furthermore, high demand has pushed component lead times to six to twelve months, while the added weight of supplementary power-distribution equipment can strain the structural capacity of multistory buildings. “100 kW+ racks pose a bigger electrical challenge than thermal,” Bizo said.

Bizo added that a 140–160 kW specification does not imply every rack in a broader deployment will run continuously at that peak. A facility optimized for such density might average approximately 80 kW per 19-inch rack position, accounting for network, storage, and auxiliary racks within the same footprint. Additionally, compute racks rarely operate at maximum power simultaneously, though training workloads can generate recurring power spikes that electrical systems must be designed to handle.

Uptime lacks a direct cross-tabulation comparing liquid-cooling capability with ultra-high-density rack support. Bizo noted that liquid-cooling readiness is more widespread than 100 kW+ rack compatibility, meaning the presence of liquid cooling alone is not a reliable indicator of extreme rack-density capability. More than half of the US operators surveyed by Uptime commissioned some form of high-density capacity during the twelve months leading up to April 2026, Bizo observed. That proportion increases to roughly two-thirds for the current twelve-month window. However, because Uptime did not strictly define high density as 100 kW+ racks, these figures do not directly quantify the rollout of ultra-high-density capacity.

A buyer procuring 5 MW at 140–160 kW per rack is not seeking generic vacant space. Such a deployment demands electrical infrastructure capable of delivering precise power loads at the rack, cooling systems engineered to dissipate the resulting heat, native direct liquid-cooling capability, and suitably contiguous floor space. Mercer’s published analysis indicates that rack-scale GPU systems operating at these densities rely on direct-to-chip liquid cooling, utilizing cold plates to cool processors while a coolant loop extends into the white space via a coolant distribution unit. His report also draws a critical distinction between a provider claiming general liquid-cooling support and a specific data hall that currently has a liquid-cooling loop actively serving the white space—a distinction that proves vital for buyers aiming for immediate deployment.

Mercer’s search confirmed that many operators with available space lacked the necessary density and cooling configurations. His analysis identifies Nvidia’s GB300 NVL72 platform as requiring approximately 140–160 kW per rack. While JLL’s 1% vacancy rate reflects broad market availability and Uptime’s data illustrate how rare extreme rack densities remain, neither metric quantifies the actual vacant capacity capable of supporting 140–160 kW racks.

Physical infrastructure limitations were not the sole hurdle Mercer faced. Of the operators he contacted, only two purpose-built, liquid-cooled facilities engaged substantively rather than issuing outright rejections. Both proactively asked whether his client held investment-grade status. Mercer again declined to name these providers, citing his ongoing business relationships. One operator confirmed it had no available capacity for the requested timeline but referenced a future project that might align, subsequently asking whether the client was investment grade. Mercer noted that further discussions were explicitly contingent on that answer. While he has observed this screening criterion specifically for high-density AI requirements, he does not characterize it as a broader industry-wide practice.

This observation builds on an April 2026 Data Center Knowledge report detailing stalled neocloud transactions, which found that operators were rigorously evaluating customers for credit quality, long-term viability, and balance-sheet durability. That reporting indicated operators were demanding stronger financial guarantees, such as letters of credit or parent-company backing, and were turning away certain neocloud deals despite highly competitive commercial terms. This development raises a more focused question: As high-density capacity grows increasingly scarce, does tenant credit quality function as an additional filtering mechanism for facilities qualified to host these deployments?

JLL’s financing analysis focuses on project-level lending rather than individual operators’ customer-selection policies, and it does not establish that colocation providers universally mandate investment-grade tenants. Nevertheless, tenant credit can significantly influence how developers structure AI and high-density projects, according to Carl Beardsley, senior managing director and data centers leader at JLL Capital Markets. When asked whether tenant credit influences developer strategies toward AI and neocloud customers, particularly for specialized high-density deployments, Beardsley responded: “The simple answer is, yes, it does.” “It is important for developers to engage an advisor or lender early in the process while structuring the lease to ensure it remains financeable,” Beardsley said.

JLL’s latest report provides crucial context regarding the financing dimensions of this issue. Construction lending remains robust across all credit tiers, encompassing AI firms and neoclouds, JLL noted. However, deals lacking established tenant credit are evaluated on a case-by-case basis, with debt metrics and risk parameters ultimately determining deal viability.