Drone strikes knock out three Yandex data centers in four days
Russia moved its digital economy onto a handful of centralized campuses. Ukraine just demonstrated what that concentration costs — and two of the buildings hit hold the machines Yandex trains its AI on.
Ukrainian drone strikes have knocked out three of Yandex's data centers in four days, including the sites that house the supercomputers the company uses to train YandexGPT. The first attack damaged the Sasovo facility southeast of Moscow on October 8. That campus holds two of Yandex's three AI training systems: the Chervonenkis supercomputer, rated at 29.42 petaflops and ranked 101st on the Top 500 list of the world's fastest machines, and the Lyapunov system at 20.03 petaflops, ranked 161st. A second strike knocked out the Kaluga region data center on October 9 — one of the largest in Russia, with a maximum capacity of 63 megawatts. The third, on October 11, forced Yandex to suspend operations at its Vladimir facility east of Moscow, which holds the third training system, Galushkin, at 20.64 petaflops and 134th on the same list.
How much damage the AI hardware itself took is not public. Ars Technica reports that it is not clear how much damage the supercomputers sustained, and Yandex's own statements describe the buildings, not the machines. That distinction decides how expensive this gets: a wrecked campus is a real-estate problem, a wrecked training cluster is a compute problem with a lead time measured in quarters. Note too that the accounting of Yandex's footprint is not consistent across outlets — the BBC counts five data centers in total, the Financial Times six — so the share of the company's capacity now offline is roughly half to three-fifths, not a precise figure.
The service damage is already large. Yandex said Sasovo was "seriously damaged" and completely shut down, that part of Kaluga's infrastructure was out of service, and that Vladimir's operations had been "completely suspended." The BBC reports the strikes disrupted Yandex's AI chatbot and smart-home devices; the Associated Press found outages in dozens of Russian cities and across Kazakhstan, Belarus and Armenia, with users unable to order taxis or reach banking services. Yandex's search engine holds about 70 percent of the Russian market, and its cloud unit counts more than 64,000 corporate customers — the company waived cloud, 360 and payment fees from October 8 to 31, and its shares fell about 5.4 percent, roughly 77 billion rubles.
The strategic point is the one pro-Kremlin commentators are making themselves. A Russian milblogger writing as Ratio Militaris argued that Yandex's hyper-centralized ecosystem — one dominant player, a few very large nodes — turned from national advantage into wartime weak link. Russia's 2019 Sovereign Internet Law makes the obvious fix hard: Russian companies are effectively barred from hosting core services abroad, and the capacity to mirror them in Belarus or Kazakhstan does not exist anyway. Ukraine's smaller, more distributed estate, plus free hosting from Microsoft and Amazon for government services, has proven harder to knock over. We covered how badly that asymmetry can end for a hyperscaler — AWS can't restore Bahrain and a UAE zone, six months after the drones. The same dynamic shaped Iran war forces UAE to redraw its 5GW AI data center map.
Ukraine frames the campaign as a "symmetrical response" — Russia struck Ukrainian data centers and mobile networks in late September, briefly cutting internet and cell service for up to 100,000 Kyiv households. What is new is the target class. AI training clusters are now, demonstrably, fixed military-adjacent infrastructure: they cannot move, they cannot be hidden, and they take months to replace.
What to watch: an unnamed IT expert told the Russian newspaper Kommersant that replacing the damaged equipment could take up to six months, and Yandex has said it is "very difficult to predict" when the sites come back. If YandexGPT's training pipeline is genuinely idled rather than merely throttled, that is the first time a national frontier lab has lost its compute to a kinetic attack.
If a country's AI capacity can be erased by four days of drone strikes, what does that mean for every data center cluster built next to a coast or a power plant? Tell us in the comments.
