Rented Intelligence: Why Russia Is Losing the Battle for AI Sovereignty

An analyst published a detailed analysis on Habr in July 2026 of strategic risks that most Russian companies ignore: AI has ceased to be a SaaS tool and has become a strategic resource — and renting Western models in this logic works just as poorly as renting an oil rig from a state that sanctions you.

Why Renting Western AI Ends

Western frontier models — GPT, Claude, Gemini — are technically inaccessible in Russia due to sanctions imposed after 2022. Companies that build processes around foreign APIs by circumventing restrictions accept a double risk: legal (operating under sanctions prohibition) and operational (infrastructure will collapse the moment access is permanently closed).

Open weights — Llama, Mistral, DeepSeek — are technically accessible and do not require API subscriptions. However, according to the author's analysis, they create a different threat: personal risks for company directors. If the state tightens requirements for the provenance and certification of AI models, an executive without a verified 'clean' chain of model origin will become vulnerable — personally, not just corporately.

Russia's Own Frontier Model: Real Timeframes

• Creating a GPT-4 level model requires tens of thousands of modern GPUs such as NVIDIA H100 or equivalents
• Sanctions restrictions on exporting such chips to Russia have been in effect since October 2022
• Domestic chips ('Baikal', 'Elbrus') lag behind global standards by several generations in performance and software ecosystem
• No Russian company has announced a frontier model competitive with GPT-4 according to public benchmarks
• Building a team of hundreds of world-class AI engineers requires years and access to the international talent market

The conclusion the author draws: Russia will not have its own competitive frontier model in the coming years. This is not pessimism — this is arithmetic.

How AI Pumps Money Across Borders

When an AI tool replaces an employee, a company saves on wages. But if the tool is a foreign API, this savings does not remain in the Russian economy: it transfers to the model owner — an American or Chinese corporation. Payroll decreases, income tax and social contributions are lost, and money flows abroad with monthly subscriptions.

"This is another form of resource dependency — not oil into the

ground, but labor into the cloud".

This effect scales directly proportional to AI adoption in business. That's why the state will sooner or later introduce a fiscal response — and will do so without waiting for AGI, simply reacting to the decline in the tax base.

What the Legislator Will Answer

The author's forecast on regulation is built on three directions: mandatory localization of AI infrastructure for critical sectors — analogous to personal data storage requirements; personal liability of executives for using unverified foreign models; a tax or fee on AI labor replacement as a tool to compensate for lost income tax and social contributions.

Moreover, none of these directions require AGI as a trigger — reaching a threshold scale of automation is sufficient, when the decline in tax base becomes statistically noticeable.

What This Means

Companies building an AI strategy around renting Western models need an audit right now: where does the model come from, where is data processed, what happens if regulation tightens. A wait-and-see position is not neutral: it is accumulating regulatory and operational risk in real time.