Russia's Sberbank offers AI model to Global South states keen to bridge digital divide

(Amends dateline)

By Gleb Bryanski and Elena Fabrichnaya

ST. PETERSBURG, Russia, June 3 (Reuters) - Russia is marketing "sovereign AI" to Global South countries concerned about privacy and content in Western AI models, a Sberbank executive told Reuters, as emerging states look to benefit from the technology and leapfrog hurdles such as weaker infrastructure.  

Russia lags the U.S. and China in the global AI race, and Sberbank and Russian IT giant Yandex are working to catch up with their flagship models, GigaChat and YandexGPT. Sberbank sees demand for models trained on local content.

"At first it will be slower and not as smart as Anthropic, Grok, or DeepSeek, but it will align with your values," First Deputy CEO of Russia's largest bank, Sberbank, and AI developer Alexander Vedyakhin told Reuters in an interview ahead of Russia's biggest economic conference in St. Petersburg.

"There is significant demand for this from countries of the Global South and those that want to develop sovereign AI but cannot afford it," he said, referring to states in Latin America, Africa, Asia, and Oceania.

Vedyakhin said that AI was developing towards more compact, specialized models that do not waste resources. He added that a model for credit scoring does not need to know rare dialects or cite poetry, which consumes resources.

"We have reached a saturation point regarding the number of parameters in a model. Users don't need billions of parameters; they need solutions to specific problems at a reasonable cost. Therefore, the next step is model compression," Vedyakhin said.

GLOBAL AI RACE

President Vladimir Putin said last week that Russia was among only three countries globally capable of developing homegrown AI models, which can be used in sensitive areas such as government operations and defence.

During Putin's visit to China in May, Sberbank's CEO, German Gref, discussed the purchase of Chinese-made chips to power GigaChat as Western sanctions continue to block the country's access to advanced hardware abroad.

Vedyakhin said that one of the secrets of U.S. chip maker Nvidia's success in controlling the AI industry was its chip programming software CUDA (Compute Unified Device Architecture), which became an industry standard that all existing LLMs are working with. Any alternative chip would face a struggle to set a new standard, he said.

He estimated AI could increase productivity by 11% to 22% in some sectors of the Russian economy, and redistribute human labour to sectors such as construction.

Weary of watching armies of dancing robots at AI events, he said he asked his staff to develop an AI-powered tiling robot that can do something useful. The robot is now being tested.

"However, it can only work effectively on a perfectly prepared surface, which must first be done by humans," he said.

(Writing by Gleb BryanskiEditing by Alexandra Hudson)