Перейти к содержанию

AI devoured the entire cloud: GPU power consumption in Russia increased by 507%

Businesses are moving neural networks from experiments to work, but there is a shortage of powerful accelerators for large models

Russian businesses have massively turned to AI computing. According to "Reg.Oblaka", in the first half of 2026, the consumption of cloud GPU infrastructure increased by 507%, the number of new connections by 160%, and user activity by 240%.

Graphics cards are no longer being acquired for experiments. They are used in customer support, document processing, anti-fraud, speech analytics, and computer vision. The most popular has become NVIDIA A4000 — it accounts for 63% of companies.

The more serious the tasks, the higher the bill. The average check for GPU servers reached 2.3 million rubles by summer, and the number of clients with expenses exceeding 500 thousand rubles per month more than doubled.

But the market has hit a hardware wall: simple cards are sufficient, but powerful accelerators for training large neural networks are not. Therefore, instead of one giant model, businesses are increasingly assembling a set of small specialized systems for specific tasks.

Read more on the topic: