BlockBeats News, July 21st, following the release of Kimi K3, the US stock semiconductor sector experienced a "DeepSeek Moment 2.0" downturn due to market concerns. However, UBS, Nomura, Bank of America Securities, and Citi all believe that this model did not reduce the demand for computing power. Instead, it may further drive the demand for AI infrastructure.

Kimi K3 has 2.8 trillion parameters, a context window of 1 million tokens, and supports continuous inference, native multimodality, and MoE architecture. Institutions believe that its scale and long context will increase KV cache utilization, driving up demand for HBM, server DDR5, enterprise SSDs, cloud infrastructure, and high-speed interconnects. Citi refers to this trend as "Another Jensen's Paradox," where model efficiency and price improvements may lead to more applications and token consumption.

UBS points out that open-source models usually rely more on memory and storage due to a longer context window. Citi believes that large-scale deployment of K3 may require super nodes composed of 64 or more GPUs. Nomura, on the other hand, believes that global large model competition will prompt leading-edge labs and hyperscale cloud platforms to continue investing, benefiting TSMC, NVIDIA, storage vendors, optical module suppliers, and data center operators in the AI infrastructure space.

However, Bank of America Securities cautioned that if the speed of model efficiency improvement continues to outpace workload growth and actual usage does not expand synchronously, there may still be a certain fallback in AI infrastructure development.