Yandex Won't Guarantee Full Server Recovery After Attack on Sasovo Data Center

Yandex has declined to guarantee that the server equipment at its Sasovo data center in Ryazan Oblast will be fully restored. The trigger was a major fire at one of the company's key infrastructure sites. Khightek, citing Forbes.ru, names a drone attack as the cause. The retelling contains no quantitative data: it is unknown how many racks were damaged, how much computing capacity went offline, or how the company values the damage. The wording "we cannot guarantee" says a lot in itself: some of the hardware has probably been lost for good.

What is known so far

The available picture is thin, so we separate what is confirmed from what is unknown.

• Site: Yandex's data center in Sasovo, Ryazan Oblast, described as key to the company's infrastructure.
• Event: a major fire; according to the outlet's headline, it followed a drone attack.
• Company position: full recovery of the server equipment is not guaranteed.
• Source: Forbes.ru, cited via Khightek, so this is a secondhand retelling.
• Not disclosed: the extent of the damage, the number of servers and racks, the share of the site's capacity, timelines, a monetary estimate.

Without an official report, any loss figures are guesswork. They should come from the company itself or from its financial reporting.

What is actually lost in a data center fire

Burned racks are only part of the damage. Smoke, combustion products, fire-suppression agents and water also damage equipment that looks intact: boards corrode, and drives and power supplies fail later, after they have been powered back on. That is why the "repair or replace" decision is made on the basis of diagnostics, which take weeks. This uncertainty is probably what lies behind the company's cautious wording.

There is also a procurement side. In Russia, servers, networking gear and accelerators have to be bought under sanctions restrictions and through long supply chains. Replacing lost capacity may take noticeably longer than in a typical region of the world. The retelling does not say which workloads ran in Sasovo (search, cloud, AI services), and we have no confirmed data on outages of specific products.

Large operators build infrastructure with headroom: load can be redistributed across sites and availability zones. But that headroom is sized for the failure of one zone, not the loss of several large facilities. If the damage turns out to be serious, the company will have to choose which services get the remaining resources first. To keep core products stable, experimental and resource-hungry workloads usually suffer, including model training and fine-tuning.

What clients and engineering teams should do

If you build products on Yandex's cloud or on other Russian providers, it is worth running through a short checklist:

• Check which availability zones host your virtual machines, databases and storage, and whether you have a single point of failure in one zone.
• Clarify your target RPO and RTO and compare them with the real ones: when did you last restore from a backup?
• Store backups away from the site of your primary workload, preferably with a different provider.
• For AI inference, keep a ready-made image and configuration so you can move the model to another zone without a manual build. Keep training checkpoints in a separate storage.
• Reread the SLA: which incidents it covers, how compensation is calculated, and whether force majeure is excluded from it.
• Follow the provider's official statements, not retellings. For now, the only thing in the public domain is the company's cautious position.

The main question for the coming weeks is how quickly Yandex will state the scale of the losses and the timeline for replacing capacity. The answer will determine whether the incident remains a local problem at one site or affects service availability and plans for developing AI infrastructure.