AI news · October 11, 2026
Ukraine strikes two Yandex data centers used for AI services
Ars reports strikes damaged two of Yandex's five data centers, including two of its three YandexGPT training supercomputers; AI products need a capacity and failover plan, not only a model plan.
- Yandex data centers affected
- 2 of 5
- YandexGPT training supercomputers at one site
- 2 of 3
Ars Technica reported on October 9 that Ukrainian strikes knocked out two of Yandex's five data centers in Russia, at Sasovo and Kaluga. The Sasovo site held two of Yandex's three supercomputers used to train YandexGPT. Yandex acknowledged service disruptions, and the report said no employees were hurt.
The event is a hard reminder that model availability depends on physical facilities, power, networks, and the ability to move work. A service with one region can look reliable during normal traffic and still fail when a site disappears. A multi-site design is not automatically safe either: queues, copies of model files, rate limits, and user messages must all behave correctly during the outage.
This is an infrastructure report, not a forecast of future service levels. Builders do not need a military scenario to run the same test. Remove one region in a staging environment, cap the remaining capacity, and verify that the product degrades clearly instead of accepting work it cannot finish. Keep the model plan and the capacity plan together.
What you can do with it
Run a tabletop outage test: remove one region, cap remaining capacity, and measure degraded behavior. Keep model files, queues, and status messages usable without promising normal latency.
Our take
The story makes availability concrete. A strong model does not help a customer when its only usable copy or serving site is down. Failover behavior belongs in the product demo and the budget.
Source: Ars Technica — Made With Models writes the brief; the reporting is theirs.