Yandex reports Alice outages after two data centers are damaged

Yandex has acknowledged that Alice and its related services are running unstably: two of the company's data centers have suffered serious damage. The press office reported this, and RB.RU relayed the statement on 9 October 2026. Some Alice AI features are temporarily unavailable, and the outages have affected smart speakers and some smart home devices. The statement gives neither the cause of the damage nor a recovery timeline.

What is known so far

There are few facts, and it is important not to speculate. Here is what the company's statement confirms:

• Two Yandex data centers are damaged. The damage is described as serious.
• Some Alice AI features are temporarily unavailable. Which ones is not specified.
• Smart speakers are affected by the outages.
• Some smart home devices are also working unstably.
• The cause of the damage has not been disclosed.
• No timeline for full recovery has been given.

The headline's phrasing, "Alice has come down with something," sounds almost homey. But "serious damage" at two sites points to an infrastructure problem, not yet another bad model update.

Why two data centers is a lot

Cloud services at this scale are usually built with spare capacity. Data is replicated, load is spread across several zones, and users most often don't notice the loss of a single site. When two go down at once, that margin runs out. This is general fault-tolerance logic, not information about Yandex's internal setup, but it explains the scale of what is happening.

AI workloads are harder than an ordinary web service. A modern assistant's answers are generated by a language model that needs GPU clusters. You can't bring up that kind of capacity on "ordinary" servers: the model either won't fit or will respond with delays unacceptable for voice. So when compute sites are lost, the heaviest features are usually the first to be switched off. Simple commands may keep working in the meantime. That is exactly the picture the company describes: the service is not down entirely, but some capabilities are unavailable.

A voice request passes through a long chain. The speaker records the phrase, the cloud recognizes the speech, the model forms a reply or a command, and only then does the signal go to the smart home service. If any link fails, the user sees the same thing: Alice stays silent or answers "I can't do that right now." Smart home devices that depend on cloud scenarios suffer a second wave: the lamps and sockets themselves work fine, but the connection to them runs through unavailable infrastructure.

What users and businesses should do

Until Yandex publishes details, the sensible approach is to act cautiously.

• Don't reset your speaker to factory settings. The problem is on the cloud side, and setting the device up again after a reset may cause extra hassle when the service returns.
• Check manual controls: physical switches, buttons on the devices, the manufacturers' own apps. If you have them, they will help during the outage.
• If your smart home controls locks, heating or security scenarios, this is a reason to rethink the setup. Critical functions should not depend on a single cloud and a voice assistant.
• Developers of skills and integrations should check their logs, request retries and error handling. Once service is restored, a queue of deferred commands may arrive.
• Follow Yandex's official channels: service status and the company's new statements are the only reliable source for now.

For the market, this story is a reminder of the physical side of AI. Assistants look like pure software, but they live in specific buildings with specific racks. The main question now is not when Alice will return: the company will most likely manage that within a reasonable time. What will be more interesting is the incident review: why the redundancy was not enough, and how Yandex will change its architecture so that losing two sites does not take the assistant offline.