The Government of the Russian Federation intends to consider, by the end of September 2026, the issue of dividing data centers (DCs) into two separate categories: facilities intended exclusively for the needs of artificial intelligence (AI) and all other sites. This initiative, overseen by the Ministry of Digital Development and a working group led by Sberbank Senior Vice President Andrei Belevtsev, aims to make state support more targeted and prevent the misuse of budget subsidies and benefits.
The idea of the reform has been under discussion since 2025, following instructions from Deputy Prime Minister Alexander Novak. It is expected that the new special status will exclude ordinary sites from being classified as AI infrastructure, which should stimulate fair competition in the market. However, many questions arise along the path to implementing this plan, the main one being the lack of clear technical criteria for the division.
Why is this important?
The scale of investment in AI infrastructure is colossal. According to expert estimates, the cost of building a conventional data center with a capacity of 10 MW is about 20 billion rubles, while an AI center of similar capacity would cost 40 billion rubles. At the same time, the projected market demand for new capacity by 2030 will reach 2.5 GW — twice the current volume of all commercial sites in the country. Representatives of the company "Atomdata" (a Rosatom structure) insist on the urgent introduction of targeted support, emphasizing that the architecture of AI sites is fundamentally different: they require high energy density, specific engineering solutions, and can be located far from end users.
Criticism and risks
The largest operator, "RTK-DC" (Rostelecom group), has opposed the new initiative. Company representative Dmitry Panyshev notes that high-tech infrastructure can be easily adapted for traditional computing, so tying subsidies to the wording "for AI" seems unjustified. Instead, he proposes evaluating facilities solely by their level of reliability.
Market analysts, including leading consultant at iKS-Consulting Stanislav Mirin, point to the complete absence of clear technical boundaries. One cannot simply label racks at 15, 20, or 50 kW as high-load, and liquid cooling cannot serve as the sole criterion. Powerful equipment is often used for standard tasks such as 3D rendering, modeling physical processes, or strength calculations — and none of this relates to AI.
My view: Until objective criteria for a strict division exist, the authorities risk creating a precedent where only select players receive support, rather than the entire industry. Instead of creating a new class, it might be wiser to administer support manually for a limited circle of companies genuinely engaged in AI development. Otherwise, we risk getting not market stimulation, but new administrative barriers.