Wall Street Institutions: Kimi K3 Is Not a “DeepSeek Moment” Sequel, But Rather Strengthens Computing Power Demand

Odaily reported that after the release of Kimi K3, the U.S. semiconductor sector declined due to market concerns over a “DeepSeek Moment 2.0.” However, UBS, Nomura, BofA Securities, and Citigroup all believe that Kimi K3 has not diminished the demand for AI computing power; instead, it may further drive the expansion of AI infrastructure.

Institutions noted that Kimi K3 features 2.8 trillion parameters, a 100-million-token context window, and supports always-on inference, native multimodal capabilities, and a MoE architecture. Its massive parameter scale and long-context capabilities will increase KV cache occupancy, thereby boosting demand for HBM, server DDR5, enterprise SSDs, cloud infrastructure, and high-speed interconnects.

Citigroup described this trend as “another Jevons paradox,” where improvements in model efficiency and cost reductions may lead to more application scenarios and higher token consumption. UBS stated that open-source models typically rely more on memory and storage due to longer context windows. Citigroup believes that large-scale deployment of Kimi K3 may require supernodes composed of more than 64 GPUs. Nomura argued that the global competition in large language models will continue to drive investment from frontier labs and cloud platforms, benefiting the AI infrastructure supply chain.

However, BofA Securities cautioned that if model efficiency improvements consistently outpace workload growth and actual usage fails to expand correspondingly, AI infrastructure construction could still face downside risks.