Russian telecom giant Rostelecom is considering building underground data centers. According to a November 2, 2026 report in the Russian daily Vedomosti, the company is in talks with mining giant Norilsk Nickel. The report is based on anonymous executives and sources, however, and it has not been confirmed that construction has been decided or that any contract has been signed.
The concept described by the paper combines two measures, replicating data at separate sites and geographically distributing computing resources and services, with the physical protection that going underground provides. Not losing data and keeping services running are separate challenges. Existing underground data centers also show that putting a facility underground is not enough on its own: power and cooling systems need redundancy as well.
Talks with Norilsk Nickel reported
According to Vedomosti, Rostelecom is considering underground data centers with strengthened protection against external threats. An anonymous executive described the plan, while a source familiar with the discussions revealed the talks with Norilsk Nickel. Another source reportedly also confirmed a layout plan for the underground data centers. However, the planned location, the depth underground, and the start of construction and operation have not been confirmed, and the budget and processing capacity of the facility are also unknown.
Rostelecom already operates a geographically distributed network of data centers. According to materials the company published on November 2, the network of its subsidiary RTK-DC comprises 27 sites, with a total of 27,800 racks and 235 MW of total power. The reported underground concept is positioned as a protective measure being considered in addition to this existing distributed network.
Still, the number of sites and the power capacity are merely indicators of scale. They alone cannot tell us whether customer services can be automatically switched to another site when a particular site becomes unusable, or how well a facility could withstand an external attack.
Data replication, geographic distribution and underground siting each play a different role
The concept reported by Vedomosti falls broadly into three stages: storing copies of data at backup sites, distributing computing resources and services across multiple geographically separated sites, and strengthening physical protection by putting facilities underground. Each measure plays a different role.
| Stage | What is placed or changed | Main role |
|---|---|---|
| Backup sites | Store copies of data at a separate site | Prevent data loss |
| Geographic distribution | Place computing resources and services at multiple distant sites | Distribute processing and services |
| Underground siting | Install data centers underground | Strengthen the physical protection of facilities |
This classification is based on the concept described by the anonymous executive. It is not a measurement of the configuration of any system actually in operation or of the fault tolerance of each measure. If copies remain at a backup site, data can be preserved even if the original site is lost. But continuing services elsewhere requires securing not only computing resources but also the necessary software and communications environment. Underground siting is the part of this scheme meant to protect the facility itself from physical threats.
Underground data centers still need redundant power and cooling
Even when a data center is placed underground, stable power supply and cooling remain essential for equipment to run. A concrete example is Pionen, an underground data center operated by Swedish telecom provider Bahnhof. The facility is a converted former civil defense shelter in Stockholm, and the company publishes the specifications of its power and cooling systems.
Pionen's uninterruptible power supply (UPS) is rated at 1.1 MW, and battery runtime is listed as 12 minutes at maximum load. Two diesel generators are installed as backup power, with a listed generating capacity of 1.8 MW, and an N+1 redundant configuration is used. N+1 is a design approach that secures one spare unit in addition to the units required for operation, so that operation can continue even if one fails. Note that the 12-minute figure is the battery runtime at maximum load and does not indicate how long the entire facility can operate.
For cooling, the facility has two 500 kW heat pumps for primary cooling and two 300 kW liquid cooling units as auxiliary, each in an N+1 configuration. Strengthening a facility's physical protection through underground siting and making power and cooling redundant need to be considered separately. These are solely the equipment specifications Bahnhof has published for Pionen, and do not indicate the specifications of any facility Rostelecom is planning.
In its quarterly report for August 2026, Bahnhof stated that lessons from Ukraine have made clear the importance of protected infrastructure and backup power. It also announced that the first section of the underground facility "Bahnhof Bunker," which it is developing in Gothenburg, is scheduled to go into operation in the first quarter of 2027. That remains a future plan at this point.
What matters is whether services can continue after losing a site
Alongside protecting facilities, what matters is a mechanism that lets services continue elsewhere when a single site becomes unusable. On November 3, 2022, Microsoft explained that it and others had distributed Ukraine's digital infrastructure to the public cloud and supported critical services through data centers across Europe. This differs from measures to protect the facilities themselves: it is a case of trying to improve service continuity by geographically distributing computing resources and data.
US outlet Ars Technica has argued that in Russia, concentration among a small number of large operators creates wartime vulnerability, while Ukraine's decentralized services showed comparatively high resilience. This is the publication's analysis, however, and not the result of comparing facilities and services in the two countries under identical conditions. As for Rostelecom, the published figure of 27 sites alone does not reveal how much service continuity could be ensured when a failure occurs.
It is difficult to evaluate a system's overall fault tolerance from the scale and protective performance of an underground data center alone. If Rostelecom's concept becomes concrete, the points to watch are how computing resources, software and communication routes would be switched when a site is lost, and how much time the design would need to restore service. Only when both the mechanisms that physically protect facilities and those that keep processing running through failures are made clear will it be possible to judge how effective this concept is.