Ukraine Strikes Expose Russia’s AI Hardware Dependence

Three Yandex supercomputers used to train artificial intelligence models reportedly suffered damage in Ukrainian attacks, exposing Russia’s dependence on foreign-made components that are now difficult to replace. The damage remains unclear, and Yandex hasn’t commented. Independent Russian outlet T-Invariant cited infrastructure experts who said all three machines sustained serious damage and remain out of service indefinitely.

Damage to Russian Systems

The Yandex data centre in Sasovo houses the Chervonenkis and Lyapunov supercomputers. It was hit on October 8. Satellite images showed a major fire and a burned section of the building, T-Invariant reported. The centre uses free cooling, drawing outside air through ventilation equipment. Smoke and combustion byproducts could harm equipment beyond the area reached by the flames.

Yandex’s third supercomputer, Galushkin, sits in Vladimir. A Ukrainian attack on October 11 shut down its data centre entirely. The three machines trained artificial intelligence models, including YandexGPT, which Yandex says it trained on its own systems.

Ukraine struck another data centre, belonging to Sber in Balakovo, Saratov Oblast, on October 14. Sber’s other two supercomputers, Christofari Neo and Christofari, sit at Sberbank’s data centre in Skolkovo, outside Moscow. That centre hasn’t been attacked.

Imported Hardware Limits

Russia built its supercomputing infrastructure domestically in areas such as software and applications, but imported hardware supplied all of its computing power. Chervonenkis, Galushkin and Lyapunov used AMD EPYC processors and NVIDIA A100 graphics accelerators. Christofari Neo used AMD EPYC processors and NVIDIA A100 accelerators; Christofari used Intel Xeon processors and NVIDIA Tesla V100 accelerators.

Russia’s full-scale invasion of Ukraine in February 2022 marked a turning point for access to those supplies. Many Western countries imposed broad export restrictions, limiting Russia’s access to semiconductors, equipment and other high-tech components. Even legal supply of the previous-generation processors and accelerators in Russian systems has become significantly harder.

Repairing a damaged supercomputer takes compatible graphics processors, CPUs, networking equipment and memory. A modern replacement needs newer accelerators. Russia can run large computing complexes and build competitive models, but it lacks an industrial counterpart to the NVIDIA A100 and the more advanced H100 and H200 accelerators.

Falling Behind in the AI Race

Seven Russian supercomputers appeared on the TOP500 list in November 2021. Chervonenkis ranked 19th, Galushkin 36th and Lyapunov 40th. Christofari Neo ranked 43rd and Christofari 72nd. Russia now has five systems on the list and ranks 21st by number of systems.

Chervonenkis still delivers about 21.53 petaflops. Yet it ranked 101st in the June 2026 TOP500 rating, the latest available. The list’s leader, LineShine, achieved 2.198 exaflops—more than 100 times Chervonenkis’s performance. Christofari Neo ranked 167th, while Christofari also remains on the list.

The Stanford AI Index 2026 said American companies created 50 of the most notable new AI models in 2025, compared with 30 from Chinese companies. Private AI investment that year totaled approximately $286 billion in the United States and $12.4 billion in China; those figures don’t include a significant portion of Chinese government funding. Russia ranked 31st among 83 countries in the Global AI Index 2024. A Sberbank representative estimated that Russian generative models lagged those of the United States and China by approximately six to nine months.