Russian data centers will require 26.2 billion kWh of electricity by 2030 — 76% more than in 2025. According to the National Rating Agency's forecast, about 70% of the consumption increase will be driven by the growth of AI workloads.
The energy consumption of AI workloads could quadruple in five years — from 2.6 billion to 10.5 billion kWh. As a result, AI workloads will account for about 70% of the total increase in data center consumption. For comparison, colocation will grow by approximately a quarter, companies' own data centers by about 15%, and regular cloud services without AI by 1.6 times.
The growth of AI may also change the industry's geography. While at the end of 2024, about 76% of data center capacities were located in the Moscow region, by the end of 2025, its share decreased to approximately 72%. By 2030, NRA predicts its reduction to 51%. The share of regions outside Moscow and St. Petersburg, on the contrary, will grow from 20% to 41%.
The reason is simple: for training large models, proximity to the end-user is not as important as the availability of free electricity. Therefore, new computing capacities may appear more actively in Siberia, the Urals, and other regions where it is easier to obtain tens of megawatts for a large project. The NRA expects that by the end of the decade, almost two-thirds of AI workloads will operate outside the two capitals.
However, market participants assess such a shift more cautiously. Moscow remains the largest center of demand for computing power, and existing capital sites are much larger than most regional ones. Nevertheless, for a new wave of AI data centers, the main limitation is increasingly not the cost of electricity, but the ability to obtain the necessary capacity at all and connect it within a reasonable timeframe.