Three Data Centers Hit in One Week: What the Yandex Attacks Reveal About How the Internet Holds Together

On the night of October 7, a drone strike set fire to Yandex's Sasovo data center in the Ryazan region, and the facility stopped operating entirely. Two days later, the company's Kaluga data center was hit, knocking several modules offline. Then on October 11, Yandex confirmed that a drone attack had damaged its Vladimir data center, and operations there were, in the company's words, "completely suspended."

Three of Yandex's five Russian data centers, damaged within five days. Internet monitoring group NetBlocks reported on October 11 that Yandex Cloud's network connectivity had fallen to 30 percent of normal levels. Yandex Pay went dark, taxi ordering stalled, and the company told cloud customers to activate their disaster-recovery plans, effectively telling them to leave.

It is the most dramatic demonstration in years of a truth the internet industry prefers to abstract away: the cloud has an address. Your data lives in a building, on a street, with a power cable and a cooling system, and buildings can be hit.

Three strikes in one week

The sequence matters because each strike removed capacity from a different site. The October 7-8 attack on Sasovo started a fire and shut the facility down completely; Reuters reported that two of Yandex's three supercomputers used to train its AI models are housed there. The October 9 strike on Kaluga took several modules out of service, and Yandex called its data centers critical infrastructure, saying staff were working around the clock. The October 11 strike on Vladimir forced a full suspension of operations.

Vladimir mattered most for the public internet. Yandex Cloud's availability zone ru-central1-a is hosted at the Vladimir facility, and the strike took power out in that zone. It is also the site behind Yandex Pay: Dima Pirogov, head of Yandex's fintech division, confirmed via Telegram that the site supporting the payments service was hit, adding that users' money was safe even as they could not reach it.

Nobody has officially claimed responsibility. Ukrainian president Volodymyr Zelenskyy said Ukraine was responding to Russian strikes on Ukrainian data centers, telling journalists that Ukraine always responds in kind, though he did not name Yandex. The BBC had reported in September that Russian drone strikes damaged a business center in Kyiv containing a data center, leaving around 100,000 households without internet access. The pattern runs in both directions.

What an availability zone actually is

Cloud providers sell resilience in layers. A data center is one building. Several buildings, each with its own power and network connections, form an availability zone, and several zones in the same general area form a region. The promise is fault isolation: a fire in one zone should not take down the others, and your workloads should fail over automatically.

The Yandex attacks stress-tested that model in an unusual way. Each strike hit a different zone, so it was not one failure to route around but three in five days, including ru-central1-a at Vladimir. Researcher Lukasz Olejnik noted that the strikes effectively removed two Yandex Cloud availability zones, leaving some database clusters read-only or down. Redundancy works when failures are independent and uncorrelated; a campaign that deliberately removes sites one by one turns independent zones into a queue.

The cloud has an address. Your data lives in a building, on a street, with a power cable, and buildings can be hit.

Why the cloud bends instead of breaking

And yet, remarkably, much of Yandex stayed up. By October 12, the company said access to the Alice AI assistant, Mail, Maps, Search, and Taxi was being restored, with engineers prioritizing the most-used services. This is what redundancy is for: traffic reroutes, replicas in surviving zones take over, and the service degrades instead of dying.

NetBlocks' 30 percent figure measures something specific: how much of Yandex Cloud's normal network connectivity was observable from the global internet on October 11. The internet itself kept routing around the damage; packets found other paths. That is the core mechanism that has held the network together since its design in the 1970s: there is no single center, so there is no single point of failure.

But 30 percent connectivity also shows the limits. Some customers' data and compute were pinned to the lost zones, and for them, failover was not automatic. That is why Yandex advised customers to host elsewhere and activate disaster-recovery plans. Platform-level redundancy protects you only if your workloads are actually spread across the platform.

The fine print: force majeure

On October 10, between the second and third strikes, Yandex updated its user agreement to add drone and missile attacks to the circumstances of force majeure that can excuse it from contractual obligations, according to the Kyiv Independent. In plain terms, the company was telling customers: if war breaks our service-level promises, you cannot bill us for it.

This is worth noticing even if you will never run a workload on Yandex Cloud. Every cloud contract has a force majeure clause, and every one defines which catastrophes the provider does not have to pay for. Outages from war, sabotage, and infrastructure attacks sit in a gray area that most customers never read until the moment it applies. The practical lesson is the same for any provider: know what your contract promises when physics wins.

The takeaway: infrastructure is policy

Data centers are becoming contested ground everywhere, not just in war zones. The same fixed, power-hungry, expensive-to-replace footprint that makes them targets in a conflict also draws protests over water and electricity, zoning fights, and regulatory scrutiny in peacetime. They concentrate risk: compute, payments, transport, and messaging, all under one roof, in a few locations.

The internet's answer to concentration has always been distribution: more zones, more regions, more paths. The Yandex outages are a reminder that distribution is a design choice you have to make deliberately, not a property that comes free with the word "cloud." Anything pinned to one zone, one provider, one country is one bad day away from being gone. Spread the load, keep backups out of region, test failover for real, and read the contract before you need it. The internet holds together because it was built to route around damage. This week, it had to.