In an interview with "Vesti," Sberbank CEO Herman Gref estimated that artificial intelligence's contribution to global GDP will be less than 1% before 2030, with an acceleration of the effect to approximately half a percentage point annually thereafter. According to him, "no other technology has had such a dramatic impact on economic development, on efficiency, and generally on changing the entire way of life of people of all professions." Immediately, a sharper forecast was voiced — "explosive growth and a change in everything" within the next three to four years — and Gref drew a practical conclusion from this: "people need to be prepared for this."
The formulation is noteworthy because it combines two different registers: a restrained quantitative assessment and almost alarmist rhetoric about an imminent turning point. It's worth understanding, step by step, to what extent these two theses are consistent with each other and with independent calculations.
What International Institutions Say
The figures cited by Gref fall within the range indicated by major research centers, although they do not match them exactly.
-
The McKinsey Global Institute, in a 2018 report, estimated the additional effect of AI on the global economy by 2030 at $13 trillion, corresponding to about 1.2 percentage points of additional annual GDP growth on average — no fresh comparable updates to this model have been released since then.
-
Goldman Sachs Research forecast in November 2023 that generative AI would begin to noticeably impact US GDP from 2027 — adding about 0.1 percentage points, accelerating to 0.4 by 2034; for other developed economies, the effect is estimated at an average of 0.3 points, and lower for emerging markets.
-
The IMF, in its World Economic Outlook updates for 2024–2026, indicates a range from 0.1 to 0.8 percentage points of additional global growth per year in the medium term, and in one scenario mentions a figure of about 0.5 points annually for 2025–2030 — which almost verbatim matches Gref's post-2030 forecast.
-
The Bank of Canada, in its May 2026 Staff Assessment of Potential Output, noted that AI would increase the annual growth of total factor productivity in the US by approximately 0.4 points over the next 10 years, while giving a more cautious estimate for the global economy as a whole — about 0.2 points.
In other words, Gref's thesis about less than 1% of GDP before 2030 and about 0.5% after — is not an original hypothesis, but essentially an average of already published scenarios from the IMF, Goldman Sachs, and the Bank of Canada. This suggests that the Sberbank CEO is conveying a consensus assessment rather than deriving his own model.
Where the Gap Between the Numbers and the Rhetoric Comes From
The problem lies elsewhere — in combining this moderate figure with the formulation about "explosive growth and a change in everything" within just three to four years. The very institutions whose assessments are cited above describe the AI effect as non-linear and stretched over time: initially dominated by capital expenditures on data centers, chips, and model training, and only later — by mass adoption in processes. This is precisely why the measured contribution to GDP remains modest so far: according to Goldman Sachs and IMF calculations for 2025–2026, the "official" effect on US GDP remains at around 0.1 percentage points in annual terms, with a "true" contribution of about 0.3 — the main part of the current activity flows through infrastructure investments, not through realized productivity growth in industries.
Given this situation, a sharp, "explosive" change in the economy within three to four years looks more like a statement for the press than a quantitative forecast: the critical mass of adoption that analysts write about typically forms over a longer period, not shorter than the indicated horizon.
Physical AI: Statements Are Backed by Specific Products
At the same time, it would be inaccurate to consider Gref's remarks a spontaneous reaction to a trend. Back on October 1, 2025, during a plenary session at the Moscow Startup Summit, he stated that AI would primarily replace "people in the most unskilled professions — physical AI," and clarified that within 2–3 years, humanoid robots would be able to "sweep streets, serve food, cook food." There, he also spoke about AGI and even superintelligence within a 10-year horizon — formulations almost verbatim repeated in a Vedomosti article from the same date.
A year later, in June 2026, at the "Bolshe, chem menedzhment" (More Than Management) forum, Gref announced that by fall 2026, Sberbank would release a line of humanoid robots called "Green" based on physical AI integrated with GigaChat: height 180 cm, weight 75 kg, 41 degrees of freedom, application — retail, food service, manufacturing, some cognitive tasks without human supervision. The bank's Robotics Center is already testing the Green-VLA model, which translates natural language tasks into physical robot actions, in real retail scenarios.
A similar vector is noted among global players: on September 2, 2026, OpenAI CEO Sam Altman confirmed the development of their own humanoid robot, naming the machine's "brain" as the main challenge and stating that everyone should have a personal robot in the future. Earlier in 2026, similar steps regarding autonomous agents working with interfaces were announced by Anthropic — specifically, in a deal with Vercept.
A Systematic Bet, Not a Sudden Discovery
The chronology — from fall 2025 through mid-2026 to the September interview — shows a consistent development of the same position, not a one-off remark amid general hype. Over a year and a half, Gref's statements have gained specificity: from an abstract thesis about human-level neural networks and "physical AI" to a specific robot model with stated technical characteristics and a release date. For Sberbank, there is also an internal motive: Gref has repeatedly linked accelerated AI adoption to the labor shortage in Russia — with record-low unemployment and no population growth, mass robotization becomes almost the only available way to increase productivity. The bank's own analysts in 2026 estimated the productivity growth potential from GenAI, robotization, and digitalization at 21–33% by 2032 — a figure significantly higher than average macroeconomic estimates of AI's contribution to GDP, which is natural since it refers to the effect at the level of an individual company and its client base, not the entire economy.
Thus, the quantitative part of Gref's statements fits well within the consensus of the IMF, Goldman Sachs, and the Bank of Canada, while the thesis about three to four years until "explosive growth" outpaces the logic of a gradual, capital-intensive-at-the-start effect described by the same sources. Rather, this reflects the specificity of the moment: Sberbank is simultaneously launching its own physical AI product on the market and formulating a macroeconomic forecast, which is why public rhetoric turns out to be noticeably bolder than the figures that the bank and its CEO themselves cite.
AI Opinion
The analysis reveals an interesting gap in the text: Sberbank's large-scale plans to release the "Green" robot line rely on models like GigaChat and Green-VLA, and such systems require serious computing power for training and inference. Experts at Hash Telegraph estimate the total energy capacity of Russian data centers at 1.2–1.7 GW — for comparison, one xAI campus near Memphis already operates on 1.3 GW with a stated increase to 2 GW.
The technological gap in computational infrastructure creates a risk not discussed in the interview: the stated timelines for "explosive growth" of physical AI may face a bottleneck not in the economics of adoption, but in the availability of chips and energy capacity for training new model versions. Will the Russian computing base be sufficient to compete with global players using orders of magnitude more computational resources?
end-content
