As Ukraine strikes Russian data centres, the country's supercomputers that are largely used to train AI models are under threat, which, in turn, jeopardises Russia's participation in the global AI race.
Reportedly, three of Russia's top supercomputers, all operated by the online giant Yandex and used to train AI models, were damaged in Ukrainian attacks earlier this month.
Yandex' data centre in Sasovo, which was attacked on October 8, houses the Chervonenkis and Lyapunov supercomputers. The company's third supercomputer, Galushkin, is located in Vladimir. That data centre also ceased operations entirely following a Ukrainian attack on October 11.
The exact extent of the damage to the supercomputer is unclear, and Yandex has not provided any comments. However, independent Russian media outlet T-Invariant quoted experts on supercomputer infrastructure as saying that all the three supercomputers were seriously damaged and are currently out of service for an indefinite period.
Satellite images of Sasovo show a major fire and a burned-out section of the building, T-Invariant reported, adding that due to the free-cooling system used at the data centres, server equipment receives outside air through the ventilation system, and, as a result, fire, smoke and combustion byproducts could damage equipment even outside the immediate fire zone.
The other two Russian supercomputers, Christofari Neo and Christofari, built for the SberCloud infrastructure, are installed at Sberbank's data centre in Skolkovo, outside Moscow. That data centre has not been attacked, but, on October 14, Ukraine struck Sber's other data centrer in Balakovo, Saratov Oblast.
Russia's supercomputers: once a big promise
Just five years ago, Russia's position in the global supercomputer race looked promising. Supercomputers - computing systems capable of performing quadrillions and quintillions of operations - are used for various purposes, including the analysis of large datasets. Most recently, with the development of generative artificial intelligence, they have taken on another strategic function: training large neural network models.
In November 2021, Russia made a significant leap in the TOP500 ranking, the international list of the 500 most powerful computing systems. Back then, seven Russian supercomputers made the list. Yandex' Chervonenkis came in 19th, Galushkin 36th, and Lyapunov 40th. Sberbank's newly inaugurated Christofari Neo took 43rd place, while the original Christofari ranked 72nd.
At that time, the Russian industry was developing as part of the global technology ecosystem. Chervonenkis and Galushkin were built on AMD EPYC processors and NVIDIA A100 graphics accelerators, while Lyapunov was based on AMD EPYC and A100.
Christofari Neo used AMD EPYC and NVIDIA A100, while Christofari used Intel Xeon and NVIDIA Tesla V100. Basically, Russia was creating its own infrastructure, software, and application solutions, but its computing power was based exclusively on imported hardware components.
Hit by sanctions
Russia's full-scale invasion of Ukraine in February 2022 was a turning point for this model. Many Western countries imposed sweeping export restrictions, significantly curtailing Russia’s access to semiconductors, equipment and other high-tech components.
Simultaneously, the global race itself accelerated sharply. Countries and major technology companies began building significantly more powerful systems based on new generations of accelerators. Meanwhile, Russian machines remained virtually unchanged, and gradually slipped in the rankings as new supercomputers emerged.
Take the example of Chervonenkis. Its performance remains the same, at about 21.53 pflops, but it fell from 19th place in 2021 to 101st in June 2026 – the most recent rating available.
Russia now has a total of five systems in the TOP500 and ranks 21st in terms of the number of systems. Exaflop-class machines already occupy the top spots in the rankings: the June 2026 leader, LineShine, achieved 2.198 exaflops - more than 100 times the performance of Chervonenkis.
A vital role in AI model training
Incidentally, the importance of Russian supercomputers has increased in recent years, as they have become one of the key resources in Russia's race toward generative AI. Yandex stressed that its supercomputers are specifically designed for machine learning tasks. The company trained YandexGPT on its own supercomputers.
Christofari and Christofari Neo, originally built within Sber’s infrastructure, play a similar role: their computing power was used to develop Russian language and generative models, including the GigaChat family and earlier projects. In the June 2026 TOP500 list, Christofari Neo ranked 167th. The original Christofari also remains on the list.
A major throwback
In the wake of Ukraine's strikes on Russian data centres, the technological dependence of the Russian supercomputers takes on a very practical significance. All five Russian top supercomputers use American AMD or Intel processors and NVIDIA A100 or V100 accelerators.
This is previous-generation equipment, but even its legal supply to Russia after 2022 is significantly complicated by export restrictions. At the same time, leading international AI clusters are transitioning to much newer generations of accelerators.
Therefore, potential damage to Russian supercomputers means more than just the need to purchase new servers. Restoring such a system requires compatible GPUs, processors, networking equipment and memory. Meanwhile, building a modern replacement would require access to the next generation of accelerators. Russia is capable of operating large computing complexes and building competitive models based on them, but it lacks its own industrial counterpart to the NVIDIA A100, let alone more advanced H100/H200-class accelerators.
That is why the current situation with data centres has implications far beyond ordinary IT infrastructure. In the era of generative AI, computing power is becoming a strategic resource. In 2021, Russian companies were still able to expand their computing power by purchasing the best equipment available on the global market. After 2022, this avenue narrowed sharply. Now, Russia risks losing a resource that was much easier to acquire five years ago than to replace today.
Lagging far behind
Despite Russian officials' claims that Russia is still running strong in the global AI race, the actual situation is far from rosy. This gap is particularly noticeable when looking not only at supercomputers but also at the AI ecosystem as a whole.
Today, the global race is effectively centred on the US and China. According to the Stanford AI Index 2026, American companies created the 50 most notable new AI models in 2025, while Chinese companies created 30.
At the same time, the gap in the quality of the two countries' top models has virtually disappeared: by March 2026, the leading US model was ahead of the strongest Chinese model in one of the main rankings by just 2.7%. The US continues to hold a massive advantage in investment: in 2025, private investment in AI totalled approximately $286bn, compared to $12.4bn in China, although these statistics do not account for a significant portion of Chinese government funding.
Russia is already outside this top group. In the Global AI Index 2024, it ranked 31st out of 83 countries. A Sberbank representative at the time estimated that Russian generative models lagged behind those of the US and China by approximately six to nine months.
Another indicator of the gap is the scale of computing infrastructure: while Russia's TOP500 currently includes only five systems, the US and China have significantly larger and rapidly updated infrastructure, and it is precisely in these two countries that the bulk of cutting-edge model development is concentrated.
And in this situation, any significant damage to Russian supercomputers makes the country's chances of improving its position in the global AI race even more problematic.