Ukrainian Drones Disable Yandex AI Data Centers in Russia

Ukrainian drones strike Yandex data centers

Ukrainian drones took out two of Yandex’s five Russian data centers on October 8 and 9, hitting facilities that house the supercomputers driving the company’s AI ambitions. The strikes landed in the Sasovo and Kaluga regions, per Reuters, and look like a deliberate step up in what’s becoming a clear pattern of data center warfare. Satellite shots from Planet Labs PBC back this up, showing structural damage at the Sasovo site-which, critically, held two of the three supercomputer clusters Yandex uses to train its YandexGPT large language model.

The hits were a direct answer to a Russian drone campaign that kicked off in late September targeting Ukrainian data infrastructure, knocking out mobile and internet service for thousands of civilians. President Zelenskyy called it a symmetrical response on social media, framing it as a measured counter to attacks on Ukraine’s digital backbone. The Ukrainian outfit Firepoint confirmed its FP-1 one-way attack drones took part in the Kaluga operation, which struck Yandex’s biggest data center facility.

Yandex-basically Russia’s Google, given its grip on search and adjacent services-admitted to service disruptions but said no staff were hurt. The fallout spread way beyond their own platforms, though. Russian users lost access to the Ivi streaming service, T-Bank’s digital banking, and the Russian Railways site, according to Meduza. A dating app, a real estate aggregator, a self-publishing platform, and the website of the Russian Premier League’s most successful soccer club also went dark, Reuters noted.

Computing infrastructure military targets Russia Ukraine

This exchange is just the latest phase in a conflict where computing infrastructure has become a legitimate military target. Back in February, Iranian drones disabled multiple Amazon data centers in Bahrain and the UAE following U.S. and Israeli strikes on Iran. Russia has hammered Ukrainian power grids, transport hubs, and residential areas since the full-scale invasion started in February 2022, but Ukraine’s recent drone campaign has widened from military bases and energy facilities to include commercial logistics warehouses run by Wildberries and Ozon-Russia’s versions of Amazon. The sinking of two commercial ships in the exclusive economic zones of NATO members Romania and Bulgaria on October 5 and 6, blamed on Russia by Zelenskyy, signals the conflict is spreading geographically, too.

AI compute physical vulnerability geopolitical risk

For anyone building or deploying AI systems, these events drive home a hard reality: the physical iron behind model training and inference is increasingly exposed to geopolitical violence. Yandex concentrating its supercomputing capacity in just three clusters-with two now damaged in a single facility-shows just how fragile centralized compute really is. That same logic applies to any outfit relying on a narrow set of data center locations, owned or leased. Redundancy across politically stable jurisdictions isn't just a reliability best practice anymore; it’s a strategic necessity.

The attacks also highlight how deeply AI services are woven into critical civilian infrastructure. When a data center goes dark, the ripple effects cascade through banking, transport, media, and everyday digital life. Builders need to evaluate not just their own infrastructure resilience, but the dependency chains of the models and APIs they plug into. A provider’s geographic footprint matters as much as its benchmark scores.

Audit compute dependencies AI infrastructure

Readers should audit their compute dependencies immediately. Map every data center hosting training runs, inference endpoints, or model weights. Flag single points of failure, especially in regions with active conflict or rising tensions. Negotiate contractual guarantees for geographic failover. Invest in distributed training frameworks that can tolerate node loss. And keep an eye on the evolving norm around data center targeting-what happened to Yandex and AWS in the Gulf may become routine rather than exceptional. The era of treating compute as an abstract, placeless resource is over.

Source: Ars Technica