Yandex Lost Three Data Centers in Four Days. Cloudflare Radar Watched It Happen Live.

On the night of October 11, a drone strike hit a data center in Vladimir, a city about 180 kilometers east of Moscow. Yandex, the Russian internet company that functions as the country's answer to Google, said the infrastructure at the facility was damaged and operations were completely suspended. The company posted a service alert on Telegram telling cloud customers to activate their disaster-recovery plans and warned that the remaining resource configuration was "unstable." Residents in nearby buildings were evacuated after a fire broke out.

What made the Vladimir strike remarkable was not the fire, or even the evacuation. It was the count. Vladimir was the third Yandex data center to go down in four days.

Four days, three data centers

The sequence began overnight on October 8 at Sasovo, southeast of Moscow. A drone strike started a fire at the facility, and Yandex fully suspended operations there. Reuters reported the attack, which appears to be the first confirmed major drone attack on a Russian data center. The Moscow Times reported that the site housed two of the three supercomputers Yandex uses to develop its AI models, including YandexGPT.

Kaluga, southwest of the capital, was hit on October 9. Yandex said "several modules" of that data center were "completely disabled." The company called its data centers critical infrastructure and said staff were working around the clock to keep services running. Yandex shares fell more than 4 percent after the strike.

Then came Vladimir overnight into Sunday, October 11. Between them, the three strikes left 3 of Yandex's 5 main data center sites disabled. On October 10, the day before the Vladimir strike, Yandex quietly updated its user agreement to add drone and missile attacks to the force majeure circumstances that can excuse it from contractual obligations, according to the Kyiv Independent.

What broke when the buildings burned

The outages were not abstract. Reports put the number of disrupted services at more than 80. Yandex Cloud customers lost access to Compute Cloud, Virtual Private Cloud, Managed Kubernetes, load balancing, Object Storage, and databases including PostgreSQL, ClickHouse, MySQL, and Yandex's own YDB. AI interfaces faltered: the YandexGPT API, SpeechKit, Vision OCR, DataLens, and Yandex AI Studio.

Consumer services felt it too. Euronews reported problems ordering taxis in Russian cities, along with login and navigation issues in Russia, Serbia, Kazakhstan, Belarus, and Armenia. The Alice voice assistant went quiet for many users. Yandex Music, Telemost video calls, and smart-home devices faltered. Enterprise users named Russian Railways and the mobile carrier MegaFon among the affected. Rival cloud providers were reportedly helping customers migrate data and applications off Yandex infrastructure.

None of this required an attack on the internet's protocols. The domain name system kept working. BGP kept announcing routes. What failed was the physical stuff: the buildings, the power feeds, the cooling. Internet monitor NetBlocks recorded Yandex Cloud network traffic falling to roughly 30 percent of normal levels.

None of this required an attack on the internet's protocols. The domain name system kept working, BGP kept announcing routes. What failed was the buildings.

Watching a cloud die on a dashboard

One of the strangest artifacts of the episode is how visible it was. Cloudflare Radar tracks traffic from AS13238, the autonomous system tied to Yandex LLC. Before the strikes, the network was overwhelmingly automated: human requests accounted for just 0.4 percent of traffic, automated activity for 99.6 percent, most of it Yandex's own crawlers, API calls, and cloud-hosted scripts.

After the strikes, the chart went off a cliff. By October 11, bot HTTP requests from AS13238 had fallen to 0.4 percent of the maximum recorded over the preceding two weeks, according to Tom's Hardware. Note the subtle trap: that 0.4 percent is not the same measure as the 0.4 percent human-traffic share in Cloudflare Radar's weekly summary. One is request volume against a recent peak, the other is traffic composition in a given window. Different denominators, same story of collapse.

Security researchers noticed quickly. One flagged the drop on X. Others pushed back on the more dramatic interpretations, noting that the metric tracks Yandex infrastructure specifically, not every Russian botnet or proxy server. The bots that disappeared were mostly Yandex's own: crawlers indexing the web, cloud scripts, API calls. When the buildings lost power, the automation went quiet.

Why this matters beyond Russia

Ukraine framed the strikes as retaliation, with Kyiv saying the Kaluga hit answered earlier Russian attacks that had reportedly disrupted internet service for 100,000 Ukrainian homes. Whatever the military logic, the episode demonstrates something internet engineers have long suspected and rarely seen tested at scale: cloud concentration is a strategic vulnerability.

Yandex is Russia's Google: search, email, maps, ride-hailing, cloud, AI. Its data centers are the physical backbone of a national digital economy. Knocking out three of five main sites did not just dent a company's balance sheet. It took down taxis, rail bookings, voice assistants, and development databases in multiple countries.

The pattern is not unique to Russia. Globally, a handful of hyperscalers and a few dozen subsea cables carry the overwhelming majority of internet traffic. The past year has shown how fragile the physical layer is: anchor drags in the Red Sea, the Philippines cable cut in early October, and now drone strikes on data centers. The internet routes around damage gracefully, but gracefully means slower, degraded, and occasionally broken for millions of people while it happens.

What comes next

Yandex has not said when services will fully return, and it has called full recovery "very difficult" to time. Its cloud continues to run in emergency mode, and observers will be watching whether traffic rebounds as workloads shift or the outages persist longer than expected. The deeper question is whether the industry absorbs the lesson the hard way. Redundancy across regions is the standard answer, and Yandex had it: workloads shifted to other zones within hours. But redundancy assumes failures are accidents, fires, floods, power faults. When data centers become deliberate targets, the math changes. Geographic distribution that once served latency now has to serve survivability, and the companies running the world's compute will be pricing that in, one fire at a time.