
There is no single, permanent answer to “What is the world’s largest data center?” A campus can lead by land area, total building area, data-hall space, power capacity, rack count, or a future full-build target. Those measures are not interchangeable, and published rankings often mix operating facilities with projects that are still planned.
This comparison was last verified on 24 August 2026. Every figure below keeps the operator’s or public authority’s original scope and states whether it describes an operating site, a completed phase, or a plan. Missing values are left missing rather than inferred.
Four distinctions matter before any facility can be called “largest”:
The following examples are useful because their operators or public authorities disclose a scale figure. They still do not form a perfect league table.
These are among the largest first-party campus figures we found, but they should not be presented as 2.62 million m² of live data halls. The operator separates total construction area from planned professional data-center-room area.
The often-repeated 10.7 million ft² claim describes a broad information-park figure and is not supported as operational data-hall area in the operator’s current disclosure. It is therefore not enough to name this site the unqualified world’s largest data center.
The previous 663,000 m² description treated an older proposal as completed dedicated data-center space. The official record supports a smaller planned total and a much smaller completed area at the 2022 phase boundary.
This is the Las Vegas campus. It should not be combined with Switch’s much larger Tahoe Reno full-build target.
Tahoe Reno is a good example of why rankings must show status. Its full-build target is much larger than the capacity disclosed when the first building opened.
Kolos was promoted as a 600,000 m², 1 GW data-center campus, but it was never built. HIVE Blockchain Technologies reported that it sold the Kolos Norway subsidiary back to the local community on 10 May 2021 after development conditions were not met. The project belongs in the historical record, not in a list of operating data centers. See the company’s SEC-filed disclosure.
A simple “top 10” can look precise while comparing different things. Runze publishes total park and building area alongside planned data-center-room space. China Telecom currently describes Inner Mongolia through a 100 MW-scale portfolio category rather than a like-for-like floor-area figure. Switch publishes operating announcements and larger full-build targets for separate campuses.
Power numbers need the same discipline. MW is a capacity rating, not annual consumption. A site can also disclose utility capacity, contracted power, critical IT load, or energized capacity, and those values can differ. Annual energy use should be stated in MWh or TWh with a defined period.
Project status can change the answer entirely. Announced campuses may be delayed, reduced, sold, or canceled. Kolos shows why a proposed headline number cannot remain in an operating ranking after the underlying project ends.
Size alone does not determine environmental impact. Utilization, hardware efficiency, cooling design, climate, grid mix, redundancy, and workload all matter. The International Energy Agency estimates that data centers used about 415 TWh of electricity in 2024, around 1.5% of global electricity consumption. Its 2026 update estimates 485 TWh in 2025 and a central projection near 950 TWh in 2030.
Water figures also depend on the accounting boundary. On-site cooling water and water used to generate electricity are different measures, and local climate changes cooling demand. The Lawrence Berkeley National Laboratory’s 2024 U.S. Data Center Energy Usage Report provides the methodology and national scenarios. A single daily water figure should not be applied to every hyperscale facility. For the broader demand and technology context, see our overview of data-center growth trends.
Hivenet uses product-specific infrastructure paths rather than one universal operating model. Its current architecture pages describe services backed by Policloud and other infrastructure appropriate to each product. Environmental performance should be evaluated by product, workload, region, baseline, redundancy, electricity mix, and stated exclusions. See how Hivenet works and the Hivenet sustainability methodology for the current scope and assumptions.
There is no defensible single answer unless “area” is defined. Among the first-party figures reviewed here, Runze publishes exceptionally large campus and building totals, but part of the data-center-room figure is planned. Campus land, total building area, and operating data-hall area should be ranked separately.
Power comparisons must separate operating IT load from utility, contracted, and full-build capacity. Switch currently advertises up to 850 MW for the Tahoe Reno campus at full build, but that target is not the same as operating capacity today.
China Telecom includes Inner Mongolia in its current portfolio of 100 MW-scale AIDC hub parks. Its March 2026 presentation does not support the old 10.7 million ft² figure as live data-hall area or name the park the world’s largest by a comparable current metric.
Northern Virginia combines dense fiber connectivity, established cloud and colocation markets, proximity to major network routes, and a large concentration of power-intensive facilities. The IEA reports that nearly half of U.S. data-center capacity is concentrated in five regional clusters. Rankings by city still depend on geographic boundaries and whether the measure is facility count, floor area, or power.
The sources and status labels were checked on 24 August 2026. Facility plans and disclosed capacity change quickly, so the source date and project status should always travel with the number.
Pick one AI, compute, or storage workload and see the difference for yourself. Spin it up in minutes, or let our team map your fastest path to production.