Yandex announced the complete closure of its data center located in Russia’s Vladimir region after a drone attack on October 11. No injuries were reported, but some group services were no longer available. The new incident draws special attention due to its repetition: it is the Yandex IT’s third major site arrived in a few daysafter Sasovo and Kaluga, in a campaign that now places data centers in the same strategic range as energy infrastructure.
Three sites interrupted in a few days
The Russian technology group said equipment was damaged in Vladimir and the center had to be closed completely. The company did not detail the site’s capacity, the number of servers affected or the expected duration of the outage. This lack of immediate assessment is common after an attack: teams must first secure the facility, verify the power supply, and examine each bay before turning power back on.
Two days earlier, drones had already attacked the Kaluga site, southwest of Moscow, causing the partial closure of several modules. The previous day the Sasovo center had been closed after a fire. The latter hosts tens of thousands of servers and two of the three supercomputers used by Yandex to train its artificial intelligence models. The company then indicated that it did not yet know if the equipment could be repaired.
A data center outage rarely spreads evenly. Consumer services can continue to operate through replication, while cloud clients, lazy processing, or regional functions become unavailable. This is what makes the impact difficult to measure from the outside. Previous incidents at Cloudflare and Amazon Web Services have shown that a localized failure can affect many services that do not carry the operator’s name.
AI infrastructure becomes a physical target
Yandex occupies a central place in the Russian digital ecosystem: search engine, mapping, messaging, commerce, payments, advertising, cloud and artificial intelligence. Its computing capabilities are used in particular to develop YandexGPT. Therefore, affecting the buildings that house these resources can slow down model training, disrupt client companies, and force the group to move loads to sites that already have demand.
This vulnerability distinguishes AI from distributed and compute-intensive software. Training a large model requires GPU clusters, fast networking, stable power supply, and properly sized cooling for high density. A supercomputer is not reconstituted simply by transferring files to another region, especially when sanctions make it difficult to purchase recent accelerators.
Drones also make the attack less expensive than building the target infrastructure. Introducing the K3 Scout on-board radar, an example of the race for detection means. For a data center, the defense must cover the building, power lines, backup pools, cooling systems, and the fibers that connect it to the network.
An alleged escalation in the infrastructure war
Ukrainian authorities have not commented in detail on each attack, but Volodymyr Zelensky said his country was responding symmetrically to Russian attacks on Ukrainian data centers and networks. This logic extends retaliation beyond refineries, warehouses, and power grids. It targets digital assets upon which civil economy, communications and analytics capabilities depend.
Russia itself has tightened its control over critical infrastructure. A provision adopted in August allows the government to take over when they are considered vulnerable. At the same time, Ukraine is moving some equipment underground to reduce its exposure. Therefore, both countries now consider server rooms as strategic facilities that must be physically protected.
This series comes at a time when data center power needs are increasing rapidly. The 3.59 GW contract signed by Google illustrates the energy scale of modern AI infrastructure. In a conflict zone, this concentration of power, cooling, and calculation also creates an obvious point of weakness.
After the attacks, Yandex removed the locations of several data centers in four Russian regions from its map service. The measure obviously does not displace the buildings, but it shows that the group now intends to reduce public information that makes it easier to identify its facilities.
