Resale of Foreign AI Models via Proxies in Russia Has Grown to 150 Million Rubles a Month

When official access to GPT, Claude and Gemini is closed to Russian companies or inconvenient to pay for, demand finds workarounds. Vedomosti described how quickly such a market has grown: at the end of 2025 there were practically no sole proprietors selling businesses access to foreign AI models through Chinese proxies. By August 2026 there were about 50 of them, and the monthly turnover of services reached 100–150 million rubles.

One of the market participants, the service 1app.ru, commented on the situation with unexpected candor: there is no know-how here and there can't be. This is the resale of access, and the added value comes from two things: constantly searching for suppliers and stable orchestration between them.

How it works

Imagine a gateway with a single API compatible with familiar SDKs. The client sends a request, and the gateway decides which supplier to route it through: who currently has quota, who is not blocked, who has lower latency and price. If one channel drops, the request goes to another, and the client doesn't notice.

Essentially this is ordinary routing and load balancing, familiar to anyone who has built fault-tolerant systems. The author of the post notes that the task is solvable from an engineering standpoint, though not quick. The difficulty is not in the algorithms but in operational routine: suppliers constantly disappear, keys get banned, prices and limits change.

Why the market is growing and where the catch is

The demand is clear: companies need frontier models, while direct payments and registration are difficult. The low barrier to entry also explains the explosive growth in the number of sellers: to get started, a few suppliers and a simple proxy layer are enough. But that same low barrier means there is no moat. Competition is based on price and uptime.

This model has limitations worth considering:

• Data. All prompts pass through a chain of intermediaries. For personal data, trade secrets and code, this is a serious risk.
• Supplier dependence. If a model provider tightens its checks, some channels will close without warning.
• Legal uncertainty. Providers' terms of use usually restrict resale and access from a number of jurisdictions.
• Stability. Quality depends on how well the reseller manages the pool of channels.

What to do in practice

If you are a developer or product manager considering such a service, do not send sensitive data through it. Use it for low-risk tasks: prototypes, processing public texts, internal experiments. Build an abstraction layer over the provider so you can switch between resellers, Russian cloud models and open-source solutions without rewriting code. Negotiate an SLA and ask how the reseller handles logs.

If you are a founder thinking of launching a similar business, assess the outlook soberly: 50 sellers in a year means a fast price race. Those who build reliable orchestration, monitoring and transparent terms for clients will survive. And the main risk remains external: you cannot control the decisions of foreign providers.

The story shows how quickly a gray market for infrastructure forms when official channels are closed. For the industry, it is a signal of real demand for frontier models, and for every team, a reason to think through a plan B in advance.