Reg.cloud: Generative AI Has Entered Its 1997 Internet Phase, and the Window to Get In Is Narrowing

Generative AI has entered a stage where the market is moving from curiosity to systematic adoption, and delay is starting to get expensive. In a column on Habr, Reg.cloud CIO Evgeny Martynov draws a direct analogy with the 1997 internet and argues that 2026–2027 will determine who manages to build AI into a product and who stays at the pilot level.

Why it's 1997

The main thesis is simple: the mass first encounter with generative AI has already happened, but the real competitive advantage will not go to those who "played around with prompts." It will go to those who turn experiments into repeatable processes.

Martynov cites Stanford AI Index 2026: in three years, generative AI reached 53% of the world's population. By comparison, the internet needed about seven years to hit the same mark, and personal computers more than ten. In other words, adoption is spreading faster than in previous technology waves.

But the author deliberately separates speed of reach from depth of adoption. In his view, the market is now in the middle of the second wave: users and companies already know the tool but have not yet built it into core processes. That is the window where late entrants have a chance: there is still time to get in without starting from absolute zero, but it is shrinking fast.

"The technology matures together with those who adopt it."

The analogy matters because in the late-1990s internet, the winners were not those who talked about the technology the most, but those who built processes, interfaces and distribution channels before everyone else. The author applies the same logic to AI: once the technology becomes truly everyday, those catching up will pay not only in money but also in time spent accumulating data, tuning teams and integrating it into the product.

What the numbers say

For the Russian market, the column offers another argument: in 2025, the generative AI segment reached 58 billion rubles according to Just AI and Onside, which means 4.5x growth in a year. The forecast for 2030 is 778 billion rubles at an average annual rate of 68%. But the key caveat concerns the near term: this growth scenario depends on whether the 2026–2027 pilots turn into real services rather than remaining presentations and internal demos.

At the same time, the market remains strikingly raw. According to MTS Web Services, only 26% of companies that are already budgeting for AI have a clear adoption strategy. The rest are either testing individual scenarios or have not left the planning stage at all.

Hence the article's second thesis: the "post-hype" era has already begun. There are plenty of experiments, but the winners will be determined not by the number of pilots but by the ability to connect data, build security and achieve a measurable business result.

The author also looks at models separately. In his assessment, Russian companies are not building frontier systems at the level of global leaders from scratch, and so they rationally bet on an open-source base, primarily the Qwen and Llama families and their derivatives. Competition is shifting from a capital-expenditure race to fine-tuning, work with domain data, applied integrations and product quality. For the enterprise segment, this matters more than the idea of having "its own big model."

What adoption is built on

The infrastructure part of the piece follows from this conclusion. If a company cannot send sensitive data to public services, and the stability of APIs and prices matters more than "out-of-the-box magic," it needs a controlled perimeter under Russian jurisdiction: owned or rented GPUs, internal knowledge bases and a clear zone of responsibility on the provider's side.

Against this backdrop, Reg.cloud has set up a dedicated AI division and listed scenarios that can already be launched without a large research team.

• Search across internal documents and RAG over a closed knowledge base
• A corporate assistant with access to sensitive information
• A first-line support helper trained on ticket history and documentation
• Generation and review of contracts and other documents under local compliance rules

As its base stack, the company names GPU bare metal and virtual machines, inference via vLLM, an internal chatbot on Open WebUI and Ollama, automation via n8n, a shared JupyterHub environment, S3 storage and the autonomous agent OpenClaw. In essence, this is an attempt to assemble not a single "magic model" but a practical platform where a business can quickly go from idea to a working scenario, from smart search over a knowledge base to support and document-process automation.

What it means

The Reg.cloud piece is both a market signal and a presentation of the company's own AI stack. Its main point is convincing: the window for calmly studying generative AI is closing, and in 2026–2027 companies will split into those that built working processes around models and those still discussing pilots. For the market, this means interest shifting from "which model to choose" to questions of data, integration, security and speed of adoption.