Svmuu News: Following the release of Kimi K3, the U.S. semiconductor sector fell amid market concerns over a “DeepSeek Moment 2.0.” However,UBS, Nomura, Bank of America Securities, and Citigroup all believe that Kimi K3 has not weakened demand for AI computing power; rather, it may further drive the expansion of AI infrastructure.
Analysts note that Kimi K3 features 2.8 trillion parameters, a 1 million-token context window, and supports always-on inference, native multimodal capabilities, and the MoE architecture.Its massive parameter count and long context capabilities will increase KV cache utilization and boost demand for HBM, server DDR5, enterprise-grade SSDs, cloud infrastructure, and high-speed interconnects.
Citi refers to this trend as “yet another Jevons Paradox,” meaning that improvements in model efficiency and cost reductions may lead to more use cases and higher token consumption.UBS noted that open-source models typically rely more heavily on memory and storage due to longer context windows; Citigroup believes that the large-scale deployment of Kimi K3 may require supernodes consisting of more than 64 GPUs. Nomura, on the other hand, believes that global competition in large models will continue to drive investment in cutting-edge laboratories and cloud platforms, benefiting the AI infrastructure industry chain.
However, Bank of America Securities cautions that if the rate of improvement in model efficiency continues to outpace workload growth, and actual usage does not expand accordingly, AI infrastructure development may still face the risk of a slowdown.