All three Yandex supercomputers that made the global TOP500 list were housed in data centers hit by drones between October 8 and 11, 2026: Chervonenkis and Lyapunov in Sasovo, Ryazan region, and Galushkin near Vladimir. Yandex has not said whether the machines themselves were damaged.
Machines like these train the neural networks behind search, the Alice assistant and translation. App outages get fixed in weeks; replacing AI compute can take years.
What we know about the machines
In November 2021 Yandex announced three TOP500 entries: Chervonenkis at 21.53 petaflops (19th in the world), Galushkin at 16.02 (36th) and Lyapunov at 12.81 (40th). All three use NVIDIA A100 accelerators and AMD EPYC processors, and Yandex said they train models with billions of parameters, including the YaLM language models used in search and Alice.
Two independent sources place the machines. Kommersant reported on October 11 that the Ryazan site housed two supercomputers and the Vladimir site housed Galushkin, and The Insider reached the same conclusion from contractor documents and a site-visit report by the Joint Institute for Nuclear Research.
The scale
Together the three deliver 50.36 petaflops. The TOP500 leader in November 2025, El Capitan in the US, measured 1.809 exaflops, or 1,809 petaflops. All of Yandex's listed supercomputers combined come to under 3% of that single machine.
Age matters too. Galushkin ranked 36th in 2021 and only 134th in the June 2026 list, according to The Insider. The machine did not slow down; the world moved on. The A100 launched in 2020 and NVIDIA has shipped two newer generations since. Yandex does not disclose how many newer GPU clusters it runs or where.
Why replacement is hard
Experts quoted by Rossiyskaya Gazeta on October 12 estimate three to four months to buy ordinary servers and up to two years for GPU servers amid the global shortage; building a comparable data center takes about a year and a half. Server maker Fplus told Kommersant that getting large volumes of hardware before year-end is a "utopia". For Russia, where sanctions restrict NVIDIA accelerator supplies, the wait may be longer still. Yandex has not commented on any damage to the supercomputers or on the impact on its AI work.
