At the AI Investment Summit themed "Certainty Opportunities in the Era of AI Infrastructure," held in Beijing, Wang Qing, Chairman of Chongyang Investment, shared his insights during a panel discussion. He highlighted that for those on the front lines of the capital markets, the emergence of ultra-nodes heralds expansive new possibilities for the entire AI landscape.

Wang pointed out that on the AI inference front, representative domestic large-scale models, such as Kimi3, have already reached a trillion-parameter scale, with expectations of climbing even higher within the year. This escalating scale imposes greater demands on computational power, requiring model architectures like the Mixture of Experts (MoE) framework. Such sophisticated parallel computing strategies necessitate system integration solutions that only ultra-node technology can effectively provide.

Should this technological pathway be successfully established, AI model inference could potentially enter a brand-new development cycle. Regarding AI training, Wang acknowledged that domestic progress currently lags behind, primarily due to physical constraints in advanced chip manufacturing processes. While domestic chips are playing an increasingly vital role in inference tasks, breakthroughs in training capabilities may similarly depend on engineering innovations to circumvent individual chip limitations and meet the substantial training demands. If ultra-node technology experiences a significant surge in adoption, it could very well propel domestic AI training into a fresh growth cycle—a development of considerable strategic importance.