Selectel: 35% of Russian companies increased IT capacity for AI projects over the past year

Half of surveyed Russian companies already use IT infrastructure for artificial intelligence tasks, and 35% increased consumption of such capacity over the last year. This comes from a Selectel study among more than 400 business representatives responsible for infrastructure development and technology projects.

Demand for Capacity

In the Selectel study, AI infrastructure is understood not only as servers with GPUs, but also as platforms for training and inference of models, as well as scalable cloud storage and related services. Against the backdrop of growing numbers of pilots and production scenarios, 35% of companies reported increasing consumption of such resources over the year. Another 14% kept it at the same level, and only about 1% reported a decline. This suggests that AI infrastructure is ceasing to be a niche story: roughly 50% of companies are already deploying it in practical projects.

According to Selectel data, demand is growing not only in surveys but also in commercial metrics. The company's revenue from GPUs in cloud servers by the end of 2025 increased almost threefold year-over-year. This clearly shows that business is transitioning from interest and experimentation to attempts to scale working use cases. At the same time, market maturity remains uneven: some have already built models into their products and operations, while others are still testing scenarios and trying to understand where AI will provide real returns.

"The market is at the stage of pilots, which are very numerous now," says

Ilya Marsharov from Kolmogorov AI.

This picture is confirmed by responses about implementation stages. Among companies already working with AI, 33% use it in real business processes, such as product creation or service delivery. Another 28% are at the piloting stage of individual solutions. Only 3% of respondents said they do not plan to apply AI in their main activities, while the rest either plan implementation or are still forming their approach. This is an important signal: a notable part of the market has already moved out of experimentation mode and is shifting models to production, but many are still testing hypotheses and selecting working scenarios.

Where Results Are Already Visible

For companies that have reached practical application, results are already quite measurable. AI is often evaluated not by grand promises, but by process speed, service quality, and impact on economics. The most noticeable effect so far is connected not with new revenue, but with improving the efficiency of current operations. Therefore, for most companies, the main KPI right now is quite pragmatic: process tasks faster, relieve the team, and reduce losses in daily work.

• 30% of respondents noted acceleration of business processes
• 27% reported increase in employee productivity
• 22% saw improvement in customer service quality
• 15% fixed cost reduction
• 11% linked AI to revenue growth or new sources of income

At the same time, 20% of organizations have not yet felt a significant effect, which is also important: implementing a model does not guarantee a business result. Real cases in the study show that effects appear where AI is built into repeatable processes. At Chity-gorod — Bukvoyed, it is used for content generation, advertising creatives, review moderation, and support request classification; the company claims that its AI assistant already closes more than 42% of requests. At Ostrovok, models help consolidate hotel data, rank tariffs, identify risky bookings, and personalize search results, while a separate loop processes more than 2 billion images.

Barriers and Personnel

The main brake on the next wave of implementation is now linked not to lack of interest, but to organizational constraints. About 27% of surveyed companies do not yet see suitable business tasks for AI. Another 25% complain about lack of expertise, 23% about high implementation costs, 21% about difficulty assessing payback, and 18% about difficulties integrating into current processes.

At the same time, 19% of respondents say they have no serious limitations. The distribution of answers depends on whether the company has real experience. Those who have not yet used AI most often run up against scenario uncertainty and skill shortages.

Those who have already gone through the initial launch stages think more about unit economics, ROI, and information security risks. In other words, as the market matures, the question shifts from "why do we need AI" to "how do we make it useful and profitable."

Against this backdrop, companies are first and foremost investing in internal teams. 40% of respondents are betting on training existing employees. Only 9% hire new specialists specifically for AI directions, and 7% engage external IT teams. This distribution shows that business still prefers to build competencies gradually, without sharp staff restructuring. For the infrastructure and integrator market, this is also a signal: demand is not only for hardware and clouds, but also for support, expertise, and hybrid launch models.

What This Means

The Russian AI market is entering a phase where the main resource becomes not the models themselves, but computational capacity, clear business scenarios, and teams capable of bringing a pilot to operational results. For infrastructure providers, this is a growth window, and for companies, it is a moment when experimenting is no longer enough: you need to count the economics and choose tasks where AI truly relieves business burden.