Putin orders a national AI plan as Russian businesses shut down 90% of pilots

The Russian AI market is accelerating and ruthlessly culling surplus experiments at the same time. The authorities are preparing a national plan for adopting neural networks, regions are scaling applied use cases, and big business increasingly admits that only a small share of pilots reach a full launch.

A state plan and new rules

After AI Journey 2025, Vladimir Putin instructed the government to prepare, by 1 June 2026, a national plan for adopting AI in the economy, public administration and the social sphere. An interdepartmental commission under the president is to coordinate it. In parallel, Rosatom and Rosseti were tasked with approving data center infrastructure development plans for 2030–2036, taking into account future demand from AI services.

This is no longer talk about a fashionable trend. It is an attempt to allocate computing power, energy and responsibility among agencies in advance.

Regulation is another important layer. The Ministry of Digital Development has published a draft law that introduces definitions of "sovereign," "national" and "trusted" AI models. For the first time, the concept of artificial intelligence itself is set down in law, and mandatory labeling is proposed for content created by neural networks. The draft would give citizens the right to challenge decisions made by government bodies with the help of AI without going to court, and would split liability for unlawful outcomes among the developer, the operator, the service owner and the user. The planned date for these provisions to take effect is 1 September 2027.

Where AI already works

At the regional level, the story looks not theoretical but quite practical. According to the digest, 55 regional governments have already put AI to work, with healthcare, public administration and security the most active areas. The Government's AI Development Center reviewed about 400 solutions, and more than 250 of them passed verification and made it onto the "Digital Region" platform.

That is an important signal: most deployments in Russia today come not through high-profile consumer products but through specific services where the effect is quick to see and the economics are easy to calculate.

It is especially visible in medicine, although public trust in such systems remains low. Only 1% of Russians are ready to fully entrust AI with health matters, yet in practice doctors already use neural networks as a second opinion. In 2025, Moscow's AI-based services processed 8.6 million studies, almost three times Moscow's own volume. The IQDOC service provided a separate data point: from 6 February to 16 March 2026, almost 3.9 thousand medical workers asked the system about 25 thousand questions, and 87% of the queries concerned diagnosis and treatment strategy.

• 55 regions already use AI in the work of government bodies
• More than 250 solutions passed verification for the "Digital Region" platform
• 8.6 million medical studies were processed by Moscow's AI services in 2025
• About 25 thousand questions were put to the AI system by doctors in a month and a half
• 87% of those queries concerned diagnosis and treatment

Pilots, business and China

The harshest part of the picture is corporate adoption. A study of the 50 largest companies found that only 7–10% of projects involving LLMs, chatbots and AI agents reached full industrial use. The rest stay in pilot, get reformatted, or are shut down altogether. And 30–40% were stopped specifically because of weak integration into real business processes.

This is an important fork in the road: the problem is often not model quality, but that a company fails to embed the model in its procedures, data, KPIs and the daily work of its teams.

Still, it is too early to talk of a reversal. 97% of large Russian companies continue to pursue AI adoption, half of industrial players are already ready to apply generative models in production, and 42% use such tools in automation. Where scaling has worked, companies report output and productivity gains of up to 20% and resource optimization of up to 15%.

Against this backdrop, there is a separate political and technological bet on China. Putin instructed the government and Sberbank to develop cooperation with China and other BRICS countries, and Sber explicitly points to DeepSeek as a benchmark for cheaper, more pragmatic solutions.

"DeepSeek proved that even with constrained capabilities, it is possible to achieve quality comparable to

American models."

What this means

The Russian AI market is entering a selection phase in which polished presentations quickly give way to infrastructure, legal frameworks and measurable benefit. The state is building rules and data centers, regions are accumulating working cases, and business is learning to close decorative pilots and keep only what fits into its processes.

If the push for local models and cooperation with China strengthens, the market could gain a more practical and less showcase-driven growth path.