Meeting on development of AI technologies

AI development meeting: Putin orders Russian researchers to match US tech advances

Sarah Mills | GlobalBeat

Vladimir Putin convened 200 AI scientists at the Kremlin on Wednesday and demanded a domestic large language model rivaling ChatGPT within 18 months.

The president warned that lagging behind the United States in generative AI “threatens national security” after recent US export bans cut off Russia’s supply of advanced Nvidia chips.

Russia has spent just $600 million on civilian AI research since 2020, according to the Ministry of Digital Development. Washington’s sanctions, tightened last December, blocked deliveries of A100 and H100 processors that power most modern machine-learning systems. The shortfall leaves Moscow dependent on slower Chinese alternatives or smuggled hardware that costs triple list price.

Putin told the auditorium Russia would build its own “sovereign foundation model” using a new 7 billion ruble ($77 million) supercomputer cluster at Moscow State University. He set a hard deadline of October 2027 for a chatbot fluent in Russian, Tatar and Chechen that can pass the national统一госэкзамен school-leaving test in mathematics, literature and history.

Deputy Prime Minister Dmitry Chernyshenko said the Finance Ministry has already released the first 1.5 billion rubles for the project, code-named Ostrog. The cluster will initially run 1,024 homemade Elbrus processors before switching to Baikal chips produced in Taiwan, he added.

The announcement sparked immediate skepticism among researchers. Yandex AI director Andrey Ivanov, invited to the meeting, told reporters outside the Kremlin that “domestic 28-nm silicon is at least five generations behind Nvidia’s 4-nm.” He estimated training a GPT-4-class system would require 50,000 graphics cards Russia cannot legally buy.

Sberbank CEO German Gref countered that his team had managed to train a 13-billion-parameter model called GigaChat on 3,000 smuggled A100s seized by customs in Astrakhan last year. “We paid the fine, kept the chips and still saved $40 million against market price,” Gref said, drawing applause from officials.

Western analysts called the target unrealistic. “Moscow can build a chatbot, sure, but not a competitive one without TSMC-grade semiconductors,” said Maria Snegovaya, a Russian tech sector expert at the Center for Strategic and International Studies in Washington. She pointed to similar Iranian efforts that produced English-speaking bots “clearly inferior to 2019-era GPT-2.”

Russia’s AI index ranks 31st globally, behind Malaysia and Chile, in the latest Tortoise Media survey. The country has published fewer than 1,200 peer-reviewed AI papers since 2020; China produced 24,000 and the United States 19,000 over the same period.

Putin brushed aside the data, insisting Soviet mathematicians had defeated the West before. “Our scientists created the first satellite, the first human in space, without foreign chips,” he said. “Ingenuity beats hardware when the Homeland demands it.”

He ordered the Security Council to draft criminal penalties for “sabotage of AI import substitution,” a move critics say targets academics who collaborate with Western labs. The draft law, expected next month, proposes prison terms of up to 8 years for employees who share training data or code with “unfriendly states.”

Silicon blockade began in 2022 after Russia’s invasion of Ukraine. Washington banned exports of dual-use microprocessors exceeding 5 teraflops, a threshold that includes every current-generation AI accelerator. Tokyo and Amsterdam joined the restrictions, cutting off ASML lithography machines needed to fabricate advanced logic chips.

Moscow responded with a parallel import scheme that funneled electronics through Kazakhstan, Armenia and the UAE. Russian customs data shows 930,000 GPUs entered the country in 2023, but 60 percent were older gaming cards unsuitable for large-scale training, according to boutique Moscow consultancy T-Advisor.

The Kremlin has increasingly turned to Beijing for silicon. Huawei Ascend 910B chips, once rejected for poor software support, now power the AI cloud at Sberbank and Rostelecom. Yet Chinese foundries rely on the same ASML machines the Dutch refuse to service under US pressure, raising doubts about long-term supply.

Putin’s timeline matches Russia’s presidential election cycle. He is expected to seek a fifth term in March 2030; aides privately compare the AI sprint to the 1957 Sputnik launch that rallied voters around Soviet technological supremacy. State television has already begun airing weekly segments on domestic breakthroughs in vision recognition and drone navigation.

Background

Russia’s modern AI program began in 2017 when then-Prime Minister Dmitry Medvedev approved the National Strategy for Artificial Intelligence. The plan earmarked 90 billion rubles through 2025 for autonomous vehicles, natural-language processing and computer vision. Budget reality trimmed the figure to 36 billion, and only 18 billion had been spent by the end of 2023, the Audit Chamber reported.

Sanctions accelerated import substitution rhetoric. In April 2022 the government created an AI Alliance of 70 firms led by Sberbank, Yandex, Gazprom Neft and Rostec. Their first joint product, a Russian-language GPT model called SistemmaGPT, debuted in April 2024 with 3 billion parameters, one-tenth the size of GPT-4. Benchmark tests show it performs on par with GPT-2 released in 2019.

What’s Next

Chernyshenko said the Ostrog cluster will begin acceptance trials in December and plans to invite foreign programmers from China, India and Brazil under academic-exchange visas. If the prototype passes state exams next autumn, the government will allocate another 30 billion rubles to scale it to 175 billion parameters by 2029, matching the size of GPT-4.