10.12.2025, Moscow.
The explosive rise in the global popularity of Chinese large language models (neural networks) indicates that technological leadership depends not so much on access to cutting-edge chips as on having talented developers capable of finding unconventional solutions, Rossa Primavera News Agency IT Desk notes.
On December 8, a report was published by the AI-model aggregator OpenRouter and the venture firm Andreessen Horowitz, stating that the share of global requests handled by Chinese neural networks increased within a few months from 1.2% to nearly 30%.
This growth resulted from the popularity of models such as Qwen from Alibaba Group Holding, R1 from DeepSeek, and Kimi K2 from Moonshot AI. At the same time, Chinese neural networks consume far fewer computing resources, meaning the US ban on access for Chinese companies to advanced US chips has not halted progress.
Competition with expensive American models — which rely on building numerous gigafactories — facilitates the emergence of developers and companies worldwide, including in Russia. Importantly, Chinese models are open source, meaning their code can be taken, improved, and adapted for specific needs.
In addition, open-source neural networks can be installed on local hardware and used without an internet connection, preventing the leakage of confidential data. This cannot be done with GPT and other American models.
Thus, competition in the AI market is shifting from a race for raw computing power to a race for efficiency and system optimization. At the same time, China’s experience demonstrates that technological sovereignty in this field is achievable even under restrictive conditions. For Russia, this opens a window of opportunity. The remaining question is whether there are enough domestic talents capable of using these tools effectively.
Source: Rossa Primavera News Agency
