AI without borders: Chinese models capture the US and help Russia
Chinese-made artificial intelligence is beginning to dominate Silicon Valley. DeepSeek, Qwen, and Kimi systems are becoming the focus of attention for American startups and corporations, increasingly replacing Western developments. They also appeared in the Russian segment. How the introduction of Chinese AI models is changing the Russian AI market and affecting the development of domestic technologies — in the Izvestia material.
Chinese AI conquers the American market
The market is changing, and major players are quickly adapting to the new reality. American companies gain an advantage by using affordable Chinese AI systems that outperform their Western counterparts in terms of speed and other parameters.
Chinese AI systems like DeepSeek, Qwen, and Kimi are actively gaining ground in Silicon Valley, pushing aside American developments as more and more local companies switch to them to create their own products. Despite the intense competition between Washington and Beijing in the field of AI, California entrepreneurs choose Asian options due to their budget, capacity and flexibility to customize to specific needs.
For startups and corporations, low tariffs for inference (data output), efficiency and the absence of restrictions in operation are priorities. Chinese open-source solutions allow complete autonomy — from downloading and fine-tuning to local deployment, minimizing the threat of data leaks and eliminating paid services.
As a result, the geography of the origin of LLM models (Large Language Model, large language model) fades into the background, and reliability, coding abilities and financial benefits turn out to be crucial.
For example, Airbnb CEO Brian Chesky admits that he uses Alibaba's Qwen line of models for a new assistant in the famous app. "They are very good, fast and cheap." He adds that the integration with ChatGPT from the American OpenAI is not yet fully suitable for the company's business objectives.
Chamat Palihapitya, fund manager at Social Capital, adds that Kimi K2 is much more productive and much cheaper than OpenAI and Anthropic solutions.
The choice of entrepreneurs highlights the seriousness of the shift towards Chinese technology in American business. Research shows that migration to Chinese models is not a one—time trend, but an absolutely systemic change.
"If at the beginning of 2024 Meta Llama (owned by Meta Corporation, recognized as extremist in Russia) had about 10.6 million downloads, while the Chinese Qwen had only 0.5 million, then by the fall of 2025 the situation had changed radically: Qwen reached 385.3 million downloads, and Llama — 346.2 million." — Hugging Face experts note.
In addition, more than 40% of the new language models on the platform are based on Qwen, while the Llama family accounts for about 15%. According to foreign analysts, these changes confirm the high competitiveness of Chinese solutions.
The Russian context: impact and prospects
Russia is also at the center of change, actively implementing Chinese solutions and developing its own AI models.
The office of Deputy Prime Minister Dmitry Grigorenko points out that Russia is one of the few countries in the world that has its own LLM. While European countries mostly use American and Chinese language models, Russia has two of its own at once: YandexGPT and GigaChat.
— Both models are already in use and show high results. For example, GigaChat 2 MAX from Sberbank was recognized as the strongest neural network for the Russian language in 2024, according to the results of the independent benchmark MERA.
To combine efforts on technology development, Russia and China have recently established a Working Group on AI.
— The Russian Federation and China have outlined promising areas of cooperation in the field of AI development. These include the development of common ethical norms and technical standards in the field of artificial intelligence, conducting scientific research and practical cooperation between Russian and Chinese companies," Grigorenko's office said.
Benefits and challenges
The use of Chinese models opens up great opportunities, but also carries risks.
Ivan Nikanov, deputy director of the Innopolis University AI Center, notes that Chinese open-source models are cheaper than Western ones, they can be run locally, which ensures data security and savings.
At the same time, Vladimir Vinogradov, CEO of Pro-Vision Communications, notes the ambivalent situation in the Russian market: Chinese technologies provide access to advanced models, lowering the entry threshold, but there is a risk of technological dependence. The expert adds that in order to remain competitive, they need their own innovations and investments in R&D.
Andrey Lubenets, Advisor to the CEO for the development of IT integrator DM Solutions, emphasizes that the main advantage is low cost and readiness for local change. The risks are related to the quality of the Russian language and issues of transparency of the models, he emphasizes.
Andrey Mishukov, CEO of iTPROTECT, adds that the current situation is not just a trend, but a structural restructuring of the global AI ecosystem, where China is forming an alternative technology stack.
At the same time, Dmitry Ovchinnikov, vCISO/UserGate information security architect, recalls the need for audit and control. According to him, any AI is just a tool, and security depends on proper use and management.
Savely Baturin, Head of the ML Department at Postgres Professional, focuses on savings and control. According to him, Qwen and DeepSeek can be downloaded, raised locally and retrained, which provides savings of up to 80-90% compared to learning from scratch. And the head of the RAEC Dmitry Gulyaev emphasizes the growing popularity of domestic solutions. For example, according to him, Alice AI is the leader among Russian users with a reach of 14.3%, significantly ahead of competitors.
Переведено сервисом «Яндекс Переводчик»
