Rostelecom: Russia’s megacities lack electricity for new AI data centers

Rostelecom: Russia's megacities lack electricity for new AI datacenters

The Russian datacenter market has hit not just demand constraints, but energy ones as well. According to Rostelecom president Mikhail Oseyevsky, megacities are running out of available electrical capacity, which means new large-scale facilities for AI and cloud services will have to be built where the grid can still handle the load.

Where the bottleneck is

Oseyevsky raised the issue at the Data Fusion conference and formulated the problem quite explicitly: the market is no longer just discussing a new datacenter, but rather at what distance from Moscow it's even possible to connect another large-scale facility. This is an important signal for the industry. Until recently, site selection was primarily driven by land availability, transport accessibility, and communication channels. Now the critical question is whether there is available capacity at the location and how quickly it can be obtained without years of waiting.

"Right now there is a discussion about at what distance from

Moscow you can build another large-scale datacenter."

The problem is especially acute for the largest cities, where the main corporate demand for cloud services, data storage, and computing is concentrated. This is where customers, operations teams, and network hubs are drawn. But this same concentration means that infrastructure gradually hits a physical ceiling. If you cannot quickly add new capacity near a megacity, the market begins to look further from the center of demand and accept more complex project logistics.

Why AI accelerates demand

For regular server businesses, power shortages are inconvenient, but for AI they become a strategic constraint. Mass deployment of models in banking, industry, telecommunications, retail, and the public sector requires not just software, but actual racks, accelerators, cooling systems, and backup power. Even if a company uses someone else's cloud, there still needs to be a physical datacenter somewhere underneath that service, connected to the grid with a clear power margin.

AI workloads differ from classical hosting in that they scale power consumption density more rapidly. The more tasks shift to recognition, generation, search across corporate data, or customer service automation, the higher the requirements for compute clusters. As a result, electricity transforms from a background infrastructure question into one of the main growth factors. If new capacity is unavailable, facility launches are delayed, and expansion costs for operators and customers begin to rise.

How the market will choose locations

If it becomes harder to build another large facility near Moscow, datacenter developers and operators will need to evaluate new territories not just by commercial criteria, but by engineering ones. For some projects, this could mean relocating facilities further from familiar clusters and earlier negotiations with network companies. Site selection for the next wave of AI infrastructure appears to be building around several basic questions.

• Is there available electrical capacity and clear timelines for technological connection
• How close is the facility to Moscow and other demand centers in terms of latency and communication channels
• Is there enough land and engineering infrastructure for subsequent expansion
• What will cooling, redundancy, and operations cost at the new location
• Can the facility be scaled incrementally rather than built all at once at the limit of capacity

For customers, this means compute availability will depend not just on budget and willingness to adopt AI. Increasingly, project timelines will need to account for connection deadlines, power margins, and regional architecture. In other words, the question "where to launch the service" is becoming nearly as important as "which model to choose." For the market, this is a sign of maturity: AI is moving beyond pilot stage and beginning to hit energy, construction, and telecom as ordinary industrial infrastructure constraints.

What this means

The story of new AI datacenters in Russia is becoming not just technological, but energetic. The more actively companies move processes to models and clouds, the more success will depend on where free capacity still remains and who can convert it to working infrastructure fastest.