How Russia is losing the AI race

Russia has its own self-driving technology, a search engine to rival Google, and voice assistants and smart speakers that are better conversationalists than Alexa or Siri. But it has no generative neural networks that can compete with the latest versions of ChatGPT or surprise everyone like China’s DeepSeek. How did this happen, and does it mean Russia has already lost the AI race?

• Russian companies began seriously exploring language models after ChatGPT’s success in late 2022. At that time, Russian developers realized that the future lay in text generation, rather than training language models for specific tasks, and in “transformers” – a special architecture that allows a language model to see the entire context at once and understand how words relate to each other. Before that, neural networks understood text word-by-word and could “forget” information if the text was long. 
• The emergence of ChatGPT kicked the Russian market into action and developers responded. Sber, Russia’s biggest bank, greatly increased investment into developing its chat bots, two market sources told The Bell. Their investment alone was serious money for the Russian market, one of the sources said, without specifying a figure. But the Russian market was not ready for this boom. The war in Ukraine made it much more difficult to acquire hardware and nobody bought expensive GPUs, advanced video cards for training neural networks that can cost$30,000–40,000 each, ahead of time. They are now even more expensive in Russia thanks to the extra logistics costs via parallel imports. Many Russian developers had to make do with their own hardware since it is not available to rent in Russia and Western players left the market.
• Another problem, which faced Yandex and Sber was staffing. Few people are capable of creating something revolutionary, and most of them are already working for OpenAI or other US-based rivals. “I know at least a few guys from Yandex who had important positions in the LLM development team have gone to work abroad, and some of them ended up at OpenAI,” said one of The Bell’s sources. “If you have the necessary skills, you can live very well in Russia. But it doesn’t compare with what you would get if you went and worked in the West in terms of money, or professional satisfaction. So it’s no surprise that we saw this outflow,” said another source.
• Censorship is another issue. Russian AI chat-bots cannot discuss “problematic” issues such as the war, LGBT rights or criticism of Vladimir Putin. For example, Sber has a special censorship model, but it is so strict that the GigaChat bot is simply unusable. 

Why the world should care

The Russian language model market is forever lagging behind, according to every developer or expert who spoke to The Bell. “We are already back about a generation and a half. And, given high interest rates, this will get worse: we need large capital investment, money is expensive and there’s no liquidity,” said Valery Babushkin, author of the book Machine Learning System Design.