Data center vacancy in the United States remained exceptionally tight through the first half of 2026, even as developers delivered a large wave of new capacity. That combination matters for AI infrastructure planning because it shows that added megawatts alone have not restored slack to the market. For security and engineering teams, scarce commissioned capacity can narrow site-selection options, compress migration timelines, and increase dependence on facilities whose power, cooling, and network designs may not match every workload requirement.
The available evidence points to a market where demand has continued to absorb supply quickly. CBRE reported that vacancy across primary U.S. data center markets fell to 1.4% in H1 2026, down from 1.6% in H1 2025, while supply increased 33.7% year over year CBRE H1 2026 data center trends. AFIRE cited an even lower national vacancy reading of about 1.1% as of Q1 2026, reflecting very limited commissioned capacity available for use AFIRE data center market analysis.
Why Data Center Vacancy Stayed Low Despite New Supply
Data Center Vacancy At 1.4% Is Not A Normal Buffer
The data center vacancy figure reported for primary U.S. markets in H1 2026 was not merely low by historical commercial-real-estate standards; it signaled a shortage of immediately usable capacity. In ordinary property markets, a rising construction pipeline would often be expected to lift vacancy. The 2026 data showed a different pattern: new capacity entered the market, but much of it had already been claimed or was absorbed quickly after delivery.
CBRE also reported that the primary U.S. markets had about 7,481 MW under construction in H1 2026. That scale is large, but construction capacity is not the same as commissioned capacity. AI training clusters, cloud regions, and high-density colocation deployments need usable power, cooling, network access, and operational readiness. A shell building or partially complete hall does not help a customer that needs racks energized on a specific date.
Preleasing Reduces The Capacity That Reaches The Open Market
Preleasing is one reason vacancy stayed compressed. The research notes show that preleasing in primary U.S. markets had historically been around 40% to 50%, but had moved into the mid-70% range by late 2025. That means a growing share of capacity was committed before customers could evaluate it as open inventory.
This is a rational response to scarcity, but it changes market behavior. Large buyers with predictable demand can reserve future space, while smaller operators, late-stage AI firms, enterprise migration teams, and security-sensitive workloads may face fewer practical choices. The effect is not only financial. Capacity scarcity can influence architecture, resilience planning, and incident recovery options because workloads may be placed where space exists rather than where the technical fit is ideal.
Market Details Behind The Scarcity Signal
Northern Virginia Shows The Extreme Case
Northern Virginia remained the clearest example of constrained supply. The research shows that vacancy in that market dropped to 0.2% during H1 2026. In Q1 2026, it had been about 0.3%, while other major markets were also tight: Atlanta at 1.0%, Dallas-Fort Worth at 1.8%, and Chicago near 2.2%.
Those figures matter because Northern Virginia is not a small edge market. It is the largest U.S. data center market in the research set, so a 0.2% vacancy rate implies very little room for unplanned demand, emergency expansion, or rapid relocation. From a cybersecurity perspective, that can affect resilience planning. If an organization depends on a particular region for latency, cloud adjacency, or network density, scarce capacity can limit practical failover options.
| Market Or Measure | Reported Period | Vacancy | Operational Meaning |
|---|---|---|---|
| Primary U.S. markets | H1 2026 | 1.4% | Record-low availability despite rapid supply growth |
| U.S. national market | Q1 2026 | About 1.1% | Very limited commissioned capacity |
| Northern Virginia | H1 2026 | 0.2% | Minimal open capacity in the largest market |
| Atlanta | Q1 2026 | 1.0% | Tight availability in a growing market |
Supply Growth Did Not Translate Into Slack
In 2025, total vacancy across eight primary U.S. markets was also reported at 1.4%, even as capacity expanded to 9,432 MW. The persistence of that figure into H1 2026 suggests that the market did not receive enough available, uncommitted capacity to create a meaningful cushion. The issue was not a lack of construction activity. It was that demand and precommitments continued to meet or exceed the pace of delivery.
For buyers, the practical distinction is significant. A market can look active on paper because many projects are under construction, while still being difficult to enter because most near-term capacity has already been reserved. This is especially relevant for AI infrastructure, where deployment schedules can be tied to accelerator procurement, model training windows, and cloud commitments.
What Low Data Center Vacancy Means For AI Infrastructure
Capacity Scarcity Can Become A Security Constraint
Low data center vacancy also changes the risk profile of AI deployments. Security teams often prefer time to assess physical controls, network segmentation, logging architecture, remote-hands processes, supply-chain exposure, and incident response procedures before placing sensitive systems in a facility. Tight availability can shorten those review cycles or push workloads into locations that satisfy power needs but require compensating controls.
The risk is not that scarce capacity automatically makes a facility unsafe. The risk is that scarcity weakens buyer choice. If only a few halls or regions can support high-density AI racks, procurement teams may have less room to reject a site because of audit gaps, maintenance constraints, limited carrier options, or operational practices that require extra oversight.
Power Availability Remains The Hard Limit
The research identifies power availability, permitting, and grid issues as major constraints that prevented vacancy from rising. This matches what operators have been reporting across AI infrastructure planning: power is not just an operating cost, but a gating factor for capacity. A related analysis of data center energy forecasts explains why power risk has become central to site planning.
Power constraints also affect resilience. Redundant electrical design, generator capacity, utility interconnection timelines, and cooling plant capacity all influence whether a data center can support dense AI equipment safely. In a constrained market, the strongest site may not be the one with the most attractive lease terms. It may be the one with credible power delivery dates and enough operational evidence to support the workload’s risk profile.
- Organizations should separate announced capacity from commissioned, available capacity.
- Security reviews should include power, cooling, access control, and operational dependency checks.
- AI capacity planning should account for preleasing, not just public construction totals.
- Regional concentration should be tested against failover and recovery requirements.
How Buyers Should Read 2026 Vacancy Data
Vacancy Numbers Need Operational Context
A low vacancy percentage is useful, but it does not explain whether the remaining space can support a specific deployment. Buyers need to ask what power density is available, when capacity can be energized, whether cooling design matches the equipment profile, and what network providers serve the site. For AI clusters, those details can decide whether a nominally available hall is usable.
There is also a timing issue. A facility under construction may be marketed as future capacity, but delays in grid interconnection or permitting can change delivery dates. That uncertainty should be treated as a project risk, not as a minor scheduling detail. Capacity that arrives six months late can affect hardware utilization, cloud commitments, customer obligations, and security validation work.
Comparing Sources Requires Care
The research notes include different vacancy readings because they measure different scopes and periods. Primary-market figures, national-market figures, colocation-only measures, and North America-wide readings are not identical datasets. A 1.4% primary-market figure and an approximately 1.1% national figure can both be directionally consistent: commissioned capacity was scarce across major U.S. data center markets in early 2026.
Readers comparing infrastructure coverage across related sites in the same network may find Natewin to be a valuable resource for adjacent technology and operations topics. The key analytical caution is to avoid treating every vacancy percentage as interchangeable. The definition of market, facility type, and measurement date matters.
Data Center Vacancy In A Constrained Expansion Cycle
The 2026 evidence shows a U.S. market expanding rapidly without creating much open capacity. Data center vacancy stayed near record lows because demand, preleasing, and infrastructure bottlenecks absorbed new supply before it could form a meaningful buffer. That is a technical infrastructure signal, not just a real-estate statistic.
For AI operators, cloud buyers, and enterprise security teams, the practical lesson is direct: capacity planning should begin earlier, include power-delivery evidence, and treat facility choice as part of the security architecture. Low vacancy does not make careful review less necessary. It makes disciplined review harder, because fewer acceptable options may be available when the deployment window opens.
The market may continue to add megawatts, but the 2026 data showed that expansion by itself did not guarantee availability. Until commissioned capacity, power delivery, permitting, and operational readiness catch up with demand, the U.S. data center market will remain tight in the places where AI workloads most want to land.
