Yandex has finally answered the question the Russian tech press has been asking for nearly three weeks: what happened to its supercomputers? The company says all three, Lyapunov, Chervonenkis and Galushkin, came through the October attacks on its data centres undamaged. They have been moved out and are being redeployed at new sites. Meanwhile, webmasters commenting on Yandex’s own Webmaster blog say that adding new sites and sitemaps is still stuck.
“They have been taken out of the locations where they were”
The statement first reached RBC on 26 October and was picked up by SEOnews and Vedomosti. Yandex says the three machines “were not damaged by the attacks on the data centres”. They “have been taken out of the locations where they were and are currently being deployed on new computing sites”, and all three will keep serving as infrastructure for Yandex services and for training foundation language models.
For context, here is what was where:
- Lyapunov, in service since December 2020, and Chervonenkis, in service since June 2021, were both in Sasovo, Ryazan region. That is the site that caught fire on 8 October.
- Galushkin, in service since June 2021, was in the Vladimir data centre, which was hit on 11 October. Yandex said at the time that the site’s operations were fully halted.
All three are built on AMD EPYC processors and NVIDIA A100 GPUs. Each links more than a hundred nodes over InfiniBand HDR, and Yandex designed its own nodes for Chervonenkis and Galushkin with optimised heat removal. In the November 2021 Top500 list they took first, second and third place among Russian systems, and Chervonenkis made the global top 20.
Not everyone is convinced
A Gordon report, citing Meduza, adds two useful caveats. The first is that the hardware is old. Chervonenkis was 19th in the world Top500 in 2021 but had fallen to 101st by June 2026. The second is that experts at the Russian outlet T-invariant said after the strikes they were confident all three machines must have suffered, if not from the strikes and fires themselves, then from smoke. Their argument was that soot could get into the electronics through the direct free-cooling systems, which draw in outside air.
Yandex hasn’t addressed the soot argument publicly, and nobody outside the company can check the hardware. So treat the statement for what it is: Yandex’s word that the clusters are intact and moving.
The recovery so far
As a refresher: the Sasovo fire was on 8 October (our coverage), part of the Kaluga infrastructure was damaged on 9 October, and Vladimir was hit on 11 October (our coverage). Vedomosti notes that Direct’s main functions came back on 20 October. On 23 October Yandex Cloud engineers had restored access to 90.47% of customer data, with part of it moved to a new site that opened the same day. Yesterday we also covered Yandex telling advertisers it saw no churn.
Webmaster: new sites and sitemaps still waiting
Search-side tools are less clear. Yandex’s last formal status update is the “Yandex Webmaster is recovering” post from 16 October. It lists robots.txt analysis, page indexing checks, sitemap analysis, the structured data validator, IP checks and the XML validator as working. Reviews and ratings, Feeds and Errors, Billing, page recrawl and page removal are view-only. Other tools may show delayed data “due to slow crawling by the robot”.
The comments under that post tell the story since then:
- 22 October: “My site isn’t being indexed either. I’ve been waiting a week.”
- 24 October: “Why can’t new sitemaps be added? When will you open uploads?”
- 26 October: “Data is only shown up to 8 October. Is there any way to influence this?” Another user is “also waiting to add a new sitemap”.
- 27 October: “Is there more precise information on when indexing of new sites will start working? I’ve got two hanging too.”
The only reply from a Yandex account, posted as “platon”, tells one of those users that indexing information is confidential and to contact support through the Webmaster chat while signed into the verified account. It gives no timeline. Another commenter asked on 21 October when Metrika’s Webvisor would be back, and got no public reply.
That matches what we saw in our Yandex forum recap on 23 October: fresh pages not indexed, recrawl blocked and IndexNow as the main workaround.
Also today: MTGBench
In a small sign that the AI teams are still shipping, Yandex and the School of Data Analysis today open-sourced MTGBench. It is a set of seven multi-step text games, including Wordle and the Russian “Field of Miracles”, for testing whether models can reason over dozens of steps without contradicting themselves. Sixteen models were tested, from a small Qwen up to Gemini and GPT, and the benchmark is on GitHub.
My take
The supercomputer news matters more for Alice AI and YandexGPT training than for search rankings. Moving and redeploying three A100-era clusters isn’t a quick job, and for a few weeks I would expect less, not more, from Yandex’s model training. For SEOs the urgent issue is still Webmaster. If you are launching a new site in Russia this month, don’t wait for Yandex to accept the sitemap. Use IndexNow, link the new URLs from pages Yandex already crawls, and keep a log of what you submitted and when, so you can tell later whether delays were down to Yandex or to you. From what users are reporting, new sites are where the outage still hurts most.