Ukraine’s recent drone strikes showed how fast an opponent can throw a high-tech system off balance by going after the power, cooling, and networks behind it. That lesson hits hard for the United States, where AI and other critical digital systems depend on physical sites that can be disrupted far more easily than most people assume.
Three Yandex computing facilities were knocked offline in just four days, including sites tied to the company’s most important AI training work. The damage may not have destroyed every piece of hardware, but it exposed something bigger: when the infrastructure underneath intelligence and computing is vulnerable, the whole machine starts to wobble.
The point is simple and uncomfortable. A system can have cutting-edge software and still fold if the building, power supply, or communications links fail at the wrong moment. That’s not a software problem, it’s a survival problem.
For national security planners, that should trigger a fresh look at America’s own data center backbone. The country’s most sensitive missions increasingly lean on private computing facilities, and that means the security of those sites is no longer just a business issue. If a site matters to defense or intelligence, it matters to the country.
The word “cloud” makes everything sound weightless and clean, but there is nothing abstract about the hardware underneath it. Servers need buildings, electricity, cooling, and communication lines to keep humming. Take away any one of those and the whole promise of constant availability starts to crack.
That is why a strike on infrastructure can do more damage than a strike on code. If intelligence arrives too late, or a mission can’t be supported when it is needed, the outage becomes a battlefield event even if no one ever breaches the software itself. Deadlines matter, and recovery that comes after the decision point is just failure in slow motion.
America also has to think about where these facilities are concentrated. Places like Loudoun County and nearby Northern Virginia hold dense clusters of data centers that help power everything from commercial services to government work. Fences and cameras help with ground threats, but they do not stop a drone overhead.
That is where policy starts catching up with reality. The government does not need to own every critical site, but it does need to know which ones are essential, what they support, and how long the nation could function if they were knocked out. Mission importance should drive protection levels, not just size, cost, or prestige.
There is already a foundation to build on. A June national security memorandum pushed agencies to work with private companies on advanced AI protection, including threat sharing, red-team testing, and stronger physical and cyber defenses for key data centers. The question now is whether those partnerships actually exist in practice and whether they are good enough to survive a real attack.
One major gap is responsibility. Private operators can harden their facilities and build redundancy, but the airspace above them is a government problem. Federal agencies and local law enforcement need a real counter-drone plan for the most sensitive sites, plus a clear chain of command when seconds count.
There also has to be a tougher standard for continuity. Critical systems should be tested the way emergency plans are tested, not just described in a binder and forgotten. If one facility goes dark, another site should be able to pick up the load, and if shared power or communications fail, the backup plan should already be proven.
Congress should want answers in classified form, with a public version that still keeps the pressure on. That kind of oversight does not weaken security, it forces it to mature. The country already knows how embarrassing it is when everyone assumes the backup works and then discovers it never really did.
Private ownership does not rule out strong security rules, either. The nuclear sector has lived with armed guards, barriers, intrusion detection, and force-on-force testing for years, even though most plants are not government-owned. AI infrastructure is different, but the basic idea holds: if the mission is vital enough, the protection has to be real.
The deeper concern sits in the background: a major fight with China would put America’s computing power under immediate pressure. If an enemy can disable the systems that support command, intelligence, or logistics at the start, it gains a huge advantage before the broader contest even settles in. That is why the country cannot afford to treat digital infrastructure like a soft target with a nice name and a shiny press release.
Every new AI breakthrough rests on a pile of very ordinary physical dependencies. That’s the part Ukraine just put on display, and it is the part Washington needs to take seriously before someone else gets a chance to test it.
