MOSCOW, June 3. /TASS/. The minimum necessary investment to create and maintain a competitive world-class large language model in Russia would exceed $10 billion, Senior Vice President of AFK Sistema, Evgeny Chereshnev, told TASS ahead of the St. Petersburg International Economic Forum (SPIEF).
According to him, the total volume of necessary investment in artificial intelligence and related infrastructure when creating such a model could reach $100 billion.
According to Chereshnev's calculations, training a modern typical model with 1.7-2 trillion parameters costs developers about $100 million per cycle. If the complexity of models increases to 100 trillion parameters in the near future, the price would reach $10 billion per cycle. "Models have 2-4 major updates per year, meaning the same number of cycles, and this is a permanent process," Chereshnev emphasized.
In addition to costs, the electricity consumption for one training cycle of a hypothetical 100 trillion parameter model will increase. According to AFK Sistema's estimates, it would be comparable to the annual consumption of Estonia or Slovenia. Chereshnev called international cooperation a solution to the funding problem.
"Russia must unite with other countries in a similar situation of potential dependence on AI technologies from the US and China, on the terms of a fair consortium -- what is available to one is available to all. This strategy has a chance of success, but requires colossal focus, will, professionalism, and an understanding of the real scale of funding," he told TASS.