Yesterday Yandex’s message about the drone strike on its Sasovo data centre was, broadly, “Search is fine, Yandex Cloud has a problem.” By Thursday evening it had changed. Yandex now says it can’t confirm the equipment can be saved, and the outage turned out to reach a long way past one cloud zone.
When we wrote up the Sasovo fire on Thursday morning, the company was insisting its core services were running normally. Here’s what has changed since.
“We cannot confirm” the hardware can be restored
At 18:26 Moscow time on Thursday, the press service told RIA Novosti: “The scale of damage to the infrastructure is still being assessed. For now we cannot confirm the possibility of restoring the equipment of the data centre in Sasovo” (our translation). AFP got the same evening statement in English: “At this stage, we cannot confirm whether the equipment at our Sasovo data center can be restored.” 3DNews says Yandex gave Forbes the same line.
Vedomosti points out the shift directly. Yandex warned that some services might be unavailable, “although earlier in the morning it had said the main services were working normally.”
The blast radius was the Runet, not just ru-central1-b
Thursday’s list of affected sites is long. According to Vedomosti, the sites of RBC, Interfax, RIA Novosti and Vedomosti itself wouldn’t open. Users also reported problems with Avito, T2, Ivi, Profi.ru and even the outage tracker Sboi.rf, plus the T-Bank and Raiffeisenbank apps. Cian blamed an incident at an unnamed “infrastructure partner”. Yandex 360 said there had been temporary problems with Mail that were later fixed. 3DNews adds Kommersant, Vesti, Gosuslugi in some regions, VK and Russian Railways, and says things were working again by Thursday evening. Tom’s Hardware cites Detector404 logging 58,268 complaints in 24 hours.
The SearchEngines.guru thread on the strike now runs to six pages, and the webmaster reports match the press reports. Early on Thursday one member posted simply “Metrika has stopped”, and later that morning another reported Yandex Mail had dropped out. By lunchtime members were asking whether Habr and VC.ru were down for everyone. One reply blamed the anti-DDoS provider Curator, which both sites use, saying it was down too. Another member said both opened fine a few hours later. One poster also claimed that Yandex Advertising Network (RSYa) payouts had been held back. I haven’t seen anything from Yandex confirming that, so treat it as forum talk for now.
Vedomosti’s experts explain why the damage spread so far. Timofey Voronin of Moscow State University argues that many businesses put their servers and databases in a single data centre to save money, so when it lost power they had no automatic failover. RUVDS chief executive Nikita Tsaplin says rerouting traffic between sites doesn’t happen instantly, so users see gaps while it does.
The AI angle: two supercomputers in the building
Sasovo houses two of Yandex’s three supercomputers, Chervonenkis and Lyapunov. Tom’s Hardware says both run on Nvidia A100s and have been used to train YandexGPT. Chervonenkis ranked 19th in the 2021 Top500 and, at 21.53 PFLOPS, would now sit around 101st. The third machine, Galushkin, is in Vladimir. Yandex hasn’t said whether either Sasovo machine was damaged. In Vedomosti, Sergei Plyusnin of E-flops says rebuilding them would be hard given restrictions on A100 supplies, and warns that losing them could slow work on YandexGPT and Alice AI. He estimates about 18 months to build a comparable facility and two to three years to fit it out with IT equipment. ComNews’s Leonid Konik puts the loss at 500–800 million roubles if the site is a write-off.
Not everyone thinks the consumer impact will last. Euler analyst Vladimir Bespalov told Vedomosti that Yandex builds in enough spare capacity for incidents like this, and none of the reports we’ve read say Search itself went down.
What I’d do this week if I worked on Russian sites
- Annotate 8 October in Metrika now. If your counters stalled, Thursday’s data has a hole in it. That hole will look like a traffic drop in every month-on-month report until the end of the year.
- Check Webmaster and your logs for 5xx. This applies even if you aren’t on Yandex Cloud. If your CDN, anti-DDoS layer, payment widget or API partner was, crawlers may still have hit errors on your URLs.
- Map third-party dependencies. Habr and VC.ru apparently went down because of a protection layer, not their own hosting. List every outside service your pages need in order to render.
- Don’t blame the algorithm yet. Any ranking or visibility wobble from 8–9 October is more likely to be infrastructure than an update. Wait a week before you call it.
- Watch for Alice changes. If training compute really is lost, expect fewer or slower Alice AI model updates. That matters for anyone tracking how to get cited in Alice answers.
My take: Yandex was right to stop repeating “everything is normal”. The more useful lesson is the one the forum got to first. Most of what broke on Thursday wasn’t Yandex’s own product. It was businesses that had quietly put all their eggs in one Yandex basket. We’ll keep following this in Yandex News.