The Russian artificial intelligence market is facing a critical problem — an acute shortage of graphics processing units (GPUs) and their rapidly rising cost. In response, a working group under the government subcommittee is actively developing a set of state support measures, including preferential taxation for the purchase of AI equipment. This initiative is a signal of a systematic approach to technological sovereignty, which is currently a priority for the authorities.
Key deadlines and objectives
According to the action plan, by the end of September, relevant departments must determine the norms that prevent modern AI equipment from being classified as domestic products. In parallel, technical criteria are being developed that will clearly separate AI accelerators (GPUs, NPUs, high-speed networks) from mining equipment. This is necessary to eliminate abuse and ensure that benefits do not go to cryptocurrency mining or gray imports.
By the end of the year, the group will present final proposals on tax preferences and other mechanisms to reduce the cost of owning such equipment. The Ministry of Industry and Trade has been appointed as the responsible body. It is telling that specific rates and support parameters have not yet been disclosed — this indicates the complexity of coordinating the interests of departments and businesses.
Expert opinion: how to separate AI from mining
The expert community is divided on approaches to classification. Nikita Kazaryan, head of the AI Platform at MWS Cloud (part of MTS), notes that over the past year, commercial prices for servers with high-performance GPUs have more than doubled, and in some cases tripled. According to a MWS Cloud study, 47% of respondents expect further increases in GPU resource costs over the next 12 months. Kazaryan proposes dividing equipment by power and class: expensive accelerators like the Nvidia H200 are simply unprofitable for mining, which simplifies filtering.
However, Yuri Tyurin from MD Audit (part of the Softline Group) rightly points out that the same accelerator can be used for both machine learning and cryptocurrency mining. He insists that the criterion should be the purpose of the system, not the chip itself. Dmitry Panyshev from the Data Center (part of Rostelecom) highlights the localization problem: GPUs are not produced in Russia, and production localization is required to receive support from the Ministry of Industry and Trade. He proposes a phased approach — from soft assembly requirements to raising the bar over 10 years — so as not to slow down the market with harsh conditions.
At the same time, Andrei Pankov from Hi-Tech insists on strictly tying benefits only to Russian equipment from the registry under Government Resolution No. 719. He sees the root of the problem in the absence of a separate category in the OKPD 2 classifier for AI chips and accelerators, which makes it impossible to recognize products as domestic.
My analysis: The initiative is correct, but the main risk is bureaucratic stalling. Without clear criteria and rapid implementation of benefits, the Russian AI sector will continue to lose competitiveness amid the global race. The key issue is not only tax relief but also creating real incentives for localization; otherwise, support will turn into subsidizing imports.