Russian business has made a rapid leap in the consumption of foreign large language models (LLMs). In the first half of 2026, corporate clients on the Yandex AI Studio platform spent 597 billion tokens — 18 times more than in the same period last year. The figure is impressive, but the demand structure is even more interesting: more than half of all tokens (54%) came from Yandex's own developments.
Chinese models are squeezing out competitors
An analysis of traffic distribution on Yandex AI Studio reveals a clear trend: Chinese models confidently hold second and third place. Qwen controls 21.2% of consumption, DeepSeek — 13.5%. American GPT-OSS from OpenAI has to settle for just 7.5%, indicating a shift in the market toward Eastern providers.
This dynamic is also confirmed on other cloud platforms. On the MWS GPT platform from MWS Cloud, token consumption in the first half of 2026 grew 11 times compared to all of 2025, reaching 400 billion tokens. Here, Qwen also holds the lead (261.1 billion tokens), followed by GLM (91.2 billion) and Kimi (81.4 billion), which overtook DeepSeek. A similar picture was recorded on Cloud.ru: over less than eight months, demand surged 18 times, with a significant share of requests going specifically to Chinese models.
Such sharp growth reflects a fundamental shift: business is moving from point experiments to the ongoing operation of generative technologies. Companies are increasingly selecting models for specific production tasks, embedding them into routine processes.
How much is the Russian LLM market worth
According to MWS Cloud estimates, the entire LLM market in Russia will grow by 35% in 2026, reaching approximately 19.6 billion rubles. The cloud segment, which includes subscription-based access to models and API access, will amount to nearly 1.5 billion rubles. However, this figure is disputed. At Turbo Cloud (a Rostelecom structure), the forecast is considered underestimated: according to Denis Bozhenko, director of AI services development, the real commercial capacity of cloud-based LLM access by the end of the year will be at least 10–12 billion rubles.
The dispute over the market size fits into a broader global context. According to AIANA analysts' calculations, the entire AI market in the country grew by 29.7% in 2025, reaching 316.1 billion rubles, and by 2030 the baseline scenario suggests growth to 830 billion. A significant portion of these funds is already being directed by business not toward purchasing equipment, but toward renting capacity and ready-made solutions.
Neural networks are being adopted most actively in HR processes: 49% of Russian employers already use AI in hiring. In the financial sector, the return on technology is also obvious — Sber alone estimated the effect of AI for 2025 at more than 475 billion rubles. The growing consumption of tokens on cloud platforms is a direct reflection of this demand.
My view: China is not just filling the vacuum left by the departure of Western vendors — it is shaping a new industry standard. Qwen and DeepSeek offer comparable quality with aggressive pricing, making them a natural choice for Russian business. In the next 12–18 months, we will likely see a further strengthening of this trend, especially in the segment of specialized industry-specific models.