Data centres as targets
Civilian data centres are being recast as potential military targets as AI workloads grow in strategic importance. Security analysis warns that this shifts placement decisions from pure ops and cost tradeoffs toward resilience questions about site control, telecom/power corridor concentration and the blast radius of regional disruptions (spiceworks.com).
For years, a data center was sold as boring infrastructure. It was a warehouse full of servers, tucked near cheap land, fat fiber routes, and reliable power. That framing is breaking down fast. As AI systems move into intelligence work, targeting, logistics, and command software, the buildings that train and run those models are starting to look less like back-office real estate and more like strategic assets in their own right (congress.gov, spiceworks.com). That shift became harder to ignore after recent reporting tied military operations in the Middle East to strikes involving commercial cloud and data-center infrastructure. The details of those episodes are still filtered through security reporting and official silence, so not every claim can be independently verified. But the larger pattern is clear enough: once civilian facilities are entangled with military AI, cloud hosting, and communications, they stop looking civilian in the old sense of the word (theintercept.com, datacenterdynamics.com). The reason is simple. Modern military systems do not just run on weapons. They run on compute. AI models need training clusters, inference capacity, storage, and constant network access. The Congressional Research Service put it plainly in February 2025: AI development and AI-enabled services depend on data centers, cloud computing, and energy resources, which raises policy questions around infrastructure access, security, and reliability (congress.gov). Once that dependency exists, geography starts to matter in a different way. A site is no longer just close to customers or tax breaks. It sits inside a web of substations, transmission lines, fiber routes, and cable landing stations. Those links are chokepoints. NATO has spent the past two years building out a formal Critical Undersea Infrastructure Network because undersea cables and pipelines are now treated as security problems, not just maintenance problems (nato.int, nato.int). The concern is obvious when you remember how much digital traffic depends on those routes. By early 2026, UN reporting on submarine-cable resilience was citing the International Telecommunication Union’s estimate that about 99 percent of international internet traffic travels through submarine cables. Damage to those lines can hit finance, emergency services, government systems, and cloud access all at once (news.un.org). A data center can have redundant generators and still fail as a strategic asset if its fiber corridors or power feeds are concentrated in the same vulnerable region. Power concentration is becoming its own kind of military problem. The Department of Energy said in December 2024 that data-center load growth had tripled over the prior decade and was projected to double or triple again by 2028, driven in part by AI applications (energy.gov, eta-publications.lbl.gov). That means more compute packed into fewer giant campuses, each one leaning on the same grid infrastructure. In strategic terms, that is not efficiency. It is target density. The grid is already showing what that density means. In a July 10, 2024 disturbance in northern Virginia, a transmission fault triggered the simultaneous loss of about 1,500 megawatts of data-center load. NERC’s incident review said the reduction was exclusively “data center-type load,” and the sudden drop forced operators to manage frequency and voltage swings the system was not built to expect (nerc.com). A Belfer Center analysis later noted that roughly 60 data centers disconnected in the event (belfercenter.org). That was not an attack. It was a warning. If one ordinary fault can create a regional reliability scare, planners have to ask what a deliberate strike, sabotage campaign, or cable cut would do. The old data-center checklist favored latency, tax incentives, and cheap megawatts. The new one has to ask who controls the land, how many other critical facilities share the same corridor, and what happens if an entire region goes dark at once. Northern Virginia, the most concentrated data-center market on Earth, already supplied the concrete detail. One fault. Sixty facilities. Fifteen hundred megawatts gone in minutes (datacenterdynamics.com, nerc.com).