Sovcombank warns that closing Chinese AI models would hit Russia's AI industry

Sovcombank analysts have warned that if Chinese developers close access to their AI models, Russia's AI industry will face rising costs and a significant slowdown in development. Forbes reports this, citing the bank's analytical study. Building a "world-class sovereign AI" from scratch would require a strategic turn that is neither easy nor cheap.

Why this matters now

Russian AI companies actively use open Chinese models, above all DeepSeek and Qwen, as the foundation for their own products. This approach saves on training and compute, speeds up time to market and sharply lowers the barrier to entry for startups. Sanctions have cut Russian businesses off from direct access to leading American models through commercial APIs, so open Chinese models have become the de facto infrastructure for a large part of the AI market.

Over the past year DeepSeek has reshuffled the balance of power: the company released models comparable in quality to GPT-4 at much lower compute cost, and opened the weight files for free download. That allowed hundreds of Russian teams to quickly build competitive products on a powerful architecture without having to train base models themselves.

A closure scenario looks realistic: as Chinese developers strengthen their commercial positions, business logic may win out over the principle of openness. OpenAI has already taken a similar path: it began as a nonprofit with open code and today runs purely proprietary models available only through a paid API.

What the Russian industry stands to lose

The analysts single out several key risks for Russian AI companies:

• A sharp rise in the cost of training models from scratch, running to tens and hundreds of millions of dollars
• Slower development without ready-made weights, architectures and public research
• A shortage of GPU infrastructure, which is critical for training large language models
• A growing technological gap with global leaders with every quarter of delay
• The need to hire ML specialists on a large scale, a scarce category in the Russian labor market

According to Sovcombank, a full "strategic turn" would require not only substantial private investment but also systemic decisions at the state level, in education, computing infrastructure and industry regulation. Without them, Russian companies risk ending up in a position where neither Chinese nor American models are available, and their own base is too weak to compete globally.

What options exist

Russia's AI industry is not standing still. It has its own developments: GigaChat from Sber, YandexGPT, Kandinsky for image generation, and a number of specialized models from startups. However, none of them yet reaches the level of the top Chinese or American competitors in the balance of capability, speed and inference cost per token.

The analysts see several possible paths. The first is accelerated technological partnership with friendly countries where open development continues and scales. The second is focusing on narrow vertical niches where Russian models can be competitive without parity with GPT-5 or Qwen-Max: legal texts, medical documentation, the public sector. The third and most ambitious is long-term state investment in computing infrastructure and workforce training, with a planning horizon of five to seven years.

"Developing a world-class sovereign AI from scratch is a strategic turn that is neither easy nor cheap," according to

Sovcombank's materials.

What this means

Dependence on open Chinese models is a convenience that may turn out to be temporary. Russia's AI industry should start budgeting now, and building expertise, for a scenario in which DeepSeek becomes a closed commercial service rather than a free tool. Delaying this step will bring not just extra costs; it risks locking in the technological lag for years to come.