Orion soft estimates that large Russian enterprises could consume 912.5 trillion AI tokens in 2026. This volume is projected to cost approximately 237.25 billion rubles, based on an average price of 260 rubles per million tokens. The calculation covers 2,000 major companies with annual revenues of at least 10 billion rubles.
The cost model accounts for various usage methods: local solutions (45% of consumption at 180 rubles per million tokens), Yandex and Sberbank APIs (25% at 350 rubles), Russian cloud platforms for models like Qwen and DeepSeek (20% at 220 rubles), and foreign APIs (10% at 450 rubles).
AI agents are identified as a key growth driver. While a standard corporate chat query consumes 2,500 tokens, an agent-based task involving six model interactions requires 34,000 tokens. Consequently, the cost per operation varies significantly, ranging from 0.04 rubles for a client response to 114.77 rubles for a complex coding task.
For 2027, Orion soft forecasts a 25–50% increase in consumption, reaching 1.14–1.37 quadrillion tokens. This growth is expected if 2–4% of standard requests are replaced by agent-based tasks. The projected annual expenditure for 2027 is estimated at 297–357 billion rubles, assuming the current average token cost remains stable.
