By Elena Fabrichnaya and Gleb Bryanski
MOSCOW (Reuters) – Artificial intelligence will give the countries that can move forward now as much influence as nuclear technology, a top Russian AI executive said, giving them superiority this century.
Alexander Vedyakhin, first deputy CEO of Sberbank, which is evolving from a major lender to an AI-focused tech conglomerate, told Reuters that it was an achievement for Russia to rank among seven countries with home-grown AI.
“AI is like a nuclear project. A new ‘nuclear club’ is emerging globally, where you either have your national large language model (LLM) or you don’t,” Vedyakhin said in an interview at Russia’s annual AI Journey event.
He said Russia should have at least two or three original AI models, not “retrained foreign models”, for use in sensitive areas such as online public services, healthcare and education.
“It is impossible to upload confidential information in a foreign model. It is simply forbidden. Doing so will lead to very unpleasant consequences,” said Vedyakhin.
President Vladimir Putin said last week that home-grown AI models were vital to preserving Russian sovereignty. Sberbank and technology firm Yandex are leading efforts to catch up with American and Chinese rivals.
Vedyakhin said Russia will struggle to match the leaders in computing, especially given Western sanctions that limit access to the technology, and the gap is likely to widen.
AI CLUB MEMBERSHIP CLOSED
Vedyakhin said the United States and China were ahead of the rest of the club, including Russia, by about six to nine months, and that membership was effectively closed.
“In this race, every day is important, but those who have not yet started are falling behind the leaders by more than a day with each passing day. For those who decide to join now, it will be very expensive, almost impossible,” he said.
“We appreciate what the Chinese and American companies have done. We understand that they have a strong start with a lot of money, experts and computing power,” said Vedyakhin.
Vedyakhin said that Sberbank’s GigaChat 2 MAX LLM was comparable to ChatGPT 4.0, while its new GigaChat Ultra Preview was equal to ChatGPT 5.0.
Sberbank is preparing for competition with next-generation models and plans to make some of its latest models open source, including for commercial use.
RUSSIA IMMUNE TO ‘BUBBUBLEA RISK’
Russia will rely on programmers and mathematicians to reduce costs and accelerate machine learning, Vedyakhin said, adding: “What we cannot achieve with sheer numbers, we achieve with skill”.
Still, AI development requires massive investment, he said, estimating the needs of Russia’s energy sector at 40 trillion rubles ($506 billion) for generation and 5 trillion for grids over the next 16 years.
A leap in LLM memory and the emergence of AI architecture not based on pre-trained generative transformers (GPTs) could mark the next breakthrough, he said, similar to what China’s DeepSeek did in 2024.
Vedyakhin warned that energy consumption levels make returns on investment in AI “either very distant or not at all visible”, and warned against “overhype” on infrastructure spending.
“We believe that excessive investments in AI infrastructure may indeed fail to pay off, given the rapid pace of technological development,” he said, adding that Russia was immune to an “AI bubble” because its investment was not excessive.
(Reporting by Gleb Bryanski; Editing by Guy Faulconbridge and Alexander Smith)