Ukrainian Strikes Hit Russia's AI Supercomputers

Three Yandex supercomputers used to train artificial intelligence models reportedly suffered damage in Ukrainian attacks on Russian data centres earlier this month. The extent remains unclear, and Yandex hasn't commented. Independent Russian outlet T-Invariant quoted experts in supercomputer infrastructure who said all three machines suffered serious damage and are out of service for an indefinite period.

The attack hit Yandex’s Sasovo data centre on October 8. The site houses the Chervonenkis and Lyapunov supercomputers. Yandex’s third supercomputer, Galushkin, is in Vladimir, where a Ukrainian attack on October 11 shut down its data centre entirely. Satellite images showed a major fire at Sasovo and a burned section of the building, T-Invariant reported. The centres use free cooling, pulling outside air through ventilation systems, so fire, smoke and combustion byproducts could harm equipment far beyond the immediate fire zone.

The Fiscal Picture

Replacing this kind of hardware costs a great deal, and the loss lands on a state that has already invested heavily in its own computing capacity. Russia built its supercomputers around imported chips, so repairs or replacements depend on components that Western export restrictions have made harder to get since 2022. A damaged system needs compatible GPUs, processors, networking equipment and memory. A modern replacement needs the next generation of accelerators, which Russia lacks.

Russia’s other two supercomputers, Christofari Neo and Christofari, were built for SberCloud and sit at Sberbank’s data centre in Skolkovo, outside Moscow. No attack has hit that centre. Ukraine did strike another Sber data centre in Balakovo, Saratov Oblast, on October 14. Taken together, the strikes suggest Russia’s computing infrastructure faces exposure across several regions, not just at one isolated facility.

Competitiveness Gap

Supercomputers can perform quadrillions and quintillions of operations, making them useful for analysing large datasets. Generative artificial intelligence has also made them central to training large neural-network models. In November 2021, seven Russian supercomputers featured in the TOP500 list of the world’s most powerful computing systems. Chervonenkis ranked 19th, Galushkin 36th and Lyapunov 40th. Sberbank’s newly inaugurated Christofari Neo took 43rd place, while Christofari ranked 72nd.

The slide in the rankings is hard to miss. Chervonenkis still runs at about 21.53 petaflops, but it fell from 19th place in 2021 to 101st in June 2026, the latest rating available. Russia now has five systems in the TOP500 and ranks 21st by number of systems. The June 2026 leader, LineShine, reached 2.198 exaflops—more than 100 times Chervonenkis’s performance. Christofari Neo ranked 167th on the June 2026 list, and Christofari remained on it.

The Stanford AI Index 2026 reported that American companies created 50 of the most notable new AI models in 2025, while Chinese companies created 30. By March 2026, the leading US model was ahead of the strongest Chinese model by 2.7% in one of the main rankings. Private AI investment in 2025 reached approximately $286 billion in the United States and $12.4 billion in China; those figures don't capture a significant portion of Chinese government funding. Russia ranked 31st among 83 countries in the Global AI Index 2024. A Sberbank representative estimated at the time that Russian generative models lagged those of the United States and China by roughly six to nine months.

Sanctions and Sovereignty

Russia’s full-scale invasion of Ukraine in February 2022 marked a turning point. Many Western countries imposed broad export restrictions, limiting Russia’s access to semiconductors, equipment and other high-tech components. Meanwhile, countries and major technology companies built more powerful systems with newer generations of accelerators. Russian machines stayed virtually unchanged as newer systems appeared, pushing them down the rankings.

All five Russian TOP500 supercomputers use American AMD or Intel processors and NVIDIA A100 or V100 accelerators. These components belong to an earlier generation, and export restrictions have made even legal supply to Russia significantly more difficult. Russia developed its own infrastructure, software and application solutions, yet its computing power depended exclusively on imported hardware. Yandex said its supercomputers are designed specifically for machine-learning tasks and that it used its own systems to train YandexGPT. Christofari and Christofari Neo helped develop Russian-language and generative models, including the GigaChat family.

Russia can run large computing complexes and build competitive models, but it lacks an industrial counterpart to the NVIDIA A100, along with the more advanced H100 and H200 accelerators. That gap is the real story here. The strikes have simply exposed how thin the foundation was.

Why This Matters: Damage to Yandex’s facilities raises a basic question: how much of Russia’s AI capacity can be restored at all? Repairs require hardware that sanctions have made scarce, while replacement depends on a generation of chips Russia cannot obtain legally or domestically. For a country already ranked far behind the United States and China in AI model development, losing major computing assets threatens to widen that gap rather than narrow it. State-linked firms such as Sberbank and Yandex will bear the fiscal burden of rebuilding, and the wider question is whether Russia’s investment in sovereign technology can deliver results when its supply chain depends so heavily on imports. Any significant, lasting damage to these systems could make it harder for Russia to improve its position in the global AI race—a point that matters to any government weighing how much to rely on imported technology for its strategic capabilities.