A stout man with a graying beard, wearing a bright-blue Yandex-branded hoodie, shouted over the din of the cooling fans.
“The servers installed in our data centers number in the tens of thousands, and they are necessary to ensure that Yandex services run without interruption,” he said, smiling widely and pointing to rows and stacks of black-and-silver boxes, glimmering with green lights and interconnected by cables.
The video, from a tour last year of what Russia’s biggest IT giant called its “iron heart” — a massive data center about 250 miles east of Moscow — resurfaced on social media this month after Ukrainian drones set the facility ablaze.
In the span of four days, the building in Sasovo and two other key Yandex data centers in Kaluga and Vladimir were struck. Attacks on facilities of VK, Russia’s biggest social network, and an under-construction center belonging to the country’s largest bank followed.
The attacks disrupted Russians’ digital lives — hailing a taxi, ordering dinner, making payments, working or learning remotely. Most services went back online within a few days, Yandex said, using other data centers.
But the Sasovo and Vladimir facilities also housed something far more complex: the country’s three most powerful supercomputers, tasked with powering Moscow’s AI aspirations.
‘Sovereign AI’
Chervonenkis, Lyapunov and Galushkin, named after renowned Soviet and Russian scientists, were built before the full-scale invasion of Ukraine using AMD processors and Nvidia accelerators — the types of hardware that have become the new gold in the global AI race and that are now almost impossible for Russia to replace on a large scale while under heavy sanctions.
Among many key tasks for government and commercial clients, the supercomputers train Yandex’s neural networks, including YandexGPT, one of the country’s leading large language models. Taking them out of commission, even temporarily, sets back Russia’s aspirations to catch up with the world’s top AI models and President Vladimir Putin’s push for “sovereign AI,” built entirely in Russia from exclusively Russian components.
At the AI Journey conference in Moscow last November, Putin described homegrown generative AI as “a matter of national, technological and value-based sovereignty,” and said only Russian models should be used for national security and intelligence-gathering.
“We cannot allow a critical dependence on foreign systems,” he said.
On paper, Russia is a perfect place for the widespread construction of AI infrastructure — an abundance of cheap electricity and vast, cold, sparsely populated landscapes, ideal for data centers, where two of the biggest headaches are energy bills and the heat produced by the hardware.
This is the pitch Russia’s business envoy Kirill Dmitriev made to President Donald Trump’s friend and special envoy Steve Witkoff. Following Witkoff’s trip to Moscow last week, Dmitriev told reporters joint data centers “could become one of the areas of technological cooperation between the U.S. and Russia.”
Before the 2022 invasion of Ukraine, Russia had a shot at being a serious contender in the AI race. Though it could hardly outrun much-richer American IT giants, it had two big domestic players with the data, money and infrastructure to back ambitious projects: Sber had already trained its ruGPT-2 and ruGPT-3 models, and Yandex’s Chervonenkis ranked 19th on the Top500 list of the world’s most powerful supercomputers.
The invasion unraveled much of this within months. Western sanctions cut off access to the chips needed to train frontier models, and talent fled the country in droves, fearing military draft. Yandex split in two; its Russian businesses were sold to local investors in 2024, while the international arm was renamed Nebius and rebuilt as an AI infrastructure company.
What was left is an industry cut off from global expertise, increasingly dependent on China and training on scarce and aging A100 chips. Chervonenkis, once 19th in the world, now ranks 101st.
“In general, no matter how excellent Russia’s AI model might be, the war has set the industry back by decades, because AI doesn’t live in a vacuum,” said Anton Merkurov, former Russian tech entrepreneur now based in Britain. “The country is at war, there are sanctions placed for years to come. Whichever thread you pull, it will unravel a problem: The equipment can’t be acquired, smart people are leaving, drones hit the buildings.”
High-value targets
After this month’s strikes, Yandex was silent for days on the state of its supercomputers, which experts feared could be harmed even if the fire never reached the machines directly, since soot-filled air can corrupt delicate hardware. It finally said they were not damaged but had been relocated to new, undisclosed facilities.
Merkurov and other experts said that data centers will remain vulnerable targets because of their sheer size, and Yandex could face a problem where it rebuilds a facility only to have it destroyed again.
“To get the missing equipment, you’d have to bypass sanctions, but that’s just the start — you need to then transport them, set everything up again, and get it all synchronized,” Merkurov said. “And then, the next batch of drones will come along.”
Sergiy Petrenko, an engineer and former head of Yandex Ukraine, which closed in 2017, said that data centers were simply never designed with such threats in mind, and that repeated strikes on other facilities “would lead to even more extensive disruptions, and there is essentially nowhere to go.”
Ukrainian President Volodymyr Zelensky said that the targeting of Yandex was “responding to Russia’s operation against our data centers,” citing a string of Russian strikes on Ukrainian providers in September. Those took out several facilities belonging to Kyivstar, one of the country’s largest mobile operators, and several hosting providers.
In late September, Russia’s Defense Ministry said it was striking data centers because they “process and transmit intelligence data for the Ukrainian Armed Forces.” In one social media post, the ministry shared an image of a drone with the caption, “Digital detox — Mandatory!”
The attacks on Russian data centers were a logical next step for Ukrainian forces, experts said, as this year Kyiv adopted a strategy intended to disrupt everyday life in Russia and undermine the Kremlin’s goal of keeping the nearly five-year-long war as far removed from citizens’ minds as possible.
Ukraine’s attacks on Russian fuel refineries, which gained steam early in the year, have drawn anger from Trump, who pressured Ukraine to limit targeting energy facilities as he tries to tame global prices ahead of the U.S. midterm elections. In the summer, Ukraine destroyed warehouses of Wildberries and Ozon, Russia’s answers to Amazon.
“Our retaliatory strikes are more painful for Russia because everything there is more centralized,” Zelensky said of the data center strikes.
About 80 percent of Russia’s operating data centers are concentrated in the Moscow region, with St. Petersburg coming second with about 7 percent, according to Russian officials.
Few options for protection
It appears that some Russian officials saw the looming threat before drones flew toward data centers.
“One of the primary tasks right now is how to relocate computing power — to shift all essential IT workloads — out of Moscow, St. Petersburg and the Moscow region to somewhere further away,” Deputy Minister of Digital Development Evgeny Filatov said at a forum just days before the first attack.
But relocating working data centers takes months, said Russian telecom expert Mikhail Klimarev, which means long downtime or expensive parallel construction.
After the strikes, Russian officials put forward proposals to pick a handful of critical data centers and move them to underground bunkers.
Klimarev said the idea is hardly feasible because data centers require huge, well-ventilated buildings, and underground spaces of this size need to be excavated from scratch, which would take years and cost far more than Russia’s cash-strapped economy can afford.
“You need a cooling system, you need an energy supply, and all of this would still be located aboveground,” he said. “If the facility itself is not destroyed, they’ll just take out the power, and you’re back to square one, so it’s expensive and it’s pointless.”
An easier solution would be to store some data in other countries, which is what Ukraine did at the start of the war.
But that would run against Russia’s years-long push for independence from foreign tech companies, the same push that kept data infrastructure concentrated on Russian soil and easier to hit.
Putin extended that logic to AI in July — despite criticism from business leaders, who feared it would complicate AI development further, and from experts, who called the demands unrealistic given Russia’s reliance on Western chips and Chinese open-source models.
The law he signed then calls for “sovereign” models as well as “national” models, the latter of which may include foreign parts. Both must keep their servers and data in the country, where the centers running them will be just as exposed to attacks.
The post These data centers hold Putin’s AI aspirations. They’re big targets. appeared first on Washington Post.
