Russian companies and government agencies may soon be required to select sovereign AI models in a wide range of sectors, even when those models are up to 50 percent more expensive than competing solutions.
According to a draft government resolution reported by Kommersant, a model would be considered sovereign only if it is developed entirely by a Russian legal entity and operates exclusively on infrastructure located inside the Russian Federation. At present, the only system meeting these criteria is GigaChat from Sber.
National models, including Yandex’s Alice AI, would be permitted to use open-source components. The proposed scope covers preparation of legislation, state programs and national projects as well as taxation, customs, budgeting, critical information infrastructure, education, healthcare and the Gosuslugi portal.
The Ministry of Digital Development has clarified that the list is still a working compilation of proposals and has not been finalized. Although the initiative has been linked to Sber, the bank stated that the document merely aggregates ideas from various participants in the discussion.
Competitors have expressed reservations. Yandex advocates uniform rules and the freedom for organizations to choose models according to their specific needs. MWS AI warned that restricting the list of acceptable solutions could leave customers without options that meet required quality or price parameters.
Experts also question the cost calculations. Technical Director of the GenAI direction at Just AI, Anton Sipachev, noted that token pricing represents only part of the total expenditure; integration, output validation and error correction can significantly increase overall costs. He further highlighted the risk of market monopolization favoring a single vendor.
