When Data Centers Became Targets

Ukraine's drones struck a Yandex data center this week housing supercomputers that train artificial intelligence models — and in doing so, moved AI from the hypothetical threat board into actual wartime targeting. The implication is quieter than it should be: if this matters enough to bomb, it will reshape how states build critical infrastructure for the next decade. The strike proves the vulnerability; the response builds the superiority.

We are watching what happened to telecommunications in the 1940s happen to machine learning in real time. When Allied bombers hit German telephone exchanges in 1943 and 1944, military planners believed a few precise strikes would cripple communication networks across occupied Europe. For weeks it looked like they had won. Then Germany learned to route calls through dispersed, redundant systems with no single point of failure — and recovery accelerated past what centralized targets could ever do.

Yandex, like most cloud companies, concentrated computational power because efficiency demands it. Redundancy costs money. Distributed systems are slower, more expensive to maintain, harder to optimize. That logic makes sense until someone decides your data center is a military objective. The moment it becomes one, efficiency stops mattering. Only recovery matters.

Russia will respond to this strike by doing what Germany did: fragmenting its AI infrastructure into geographically scattered nodes, hardened against targeting, designed for rapid failover. AWS, Google Cloud, Microsoft Azure. The companies that have won the last decade of cloud dominance. Are still built on the centralized model because it was never supposed to be under attack. They optimized for a peacetime internet. When state-level actors stop assuming peacetime, centralization becomes a liability. A distributed Russian AI infrastructure would be harder to cripple, slower to recover, more resilient to what comes next. The strike was tactically brilliant. It may be strategically catastrophic.

You see this pattern in any domain where the rules shift from efficiency to redundancy. A jazz musician who practices one version of a song owns nothing when a key gets stuck. A startup with one critical supplier owns nothing when supply chains break. The Yandex strike didn't degrade Russian AI capacity. It clarified what matters now. Fragmentation wins in a world where infrastructure is a target. That's a different game than the one Western tech built to dominate.