Woofun AI reports that UBS, Nomura, Bank of America Securities, and Citi argue the release of Kimi K3 does not reduce computing power demand but may instead stimulate AI infrastructure investment. The model features 2.8 trillion parameters, a 1 million token context window, and supports continuous inference and native multimodality. Institutions note that these specifications increase KV cache utilization, thereby boosting requirements for HBM, server DDR5, enterprise SSDs, cloud infrastructure, and high-speed interconnects.

Citi describes this dynamic as "Another Jensen's Paradox," where improved model efficiency and pricing could expand applications and token consumption. UBS highlights that open-source models with longer context windows rely more heavily on memory and storage. Citi suggests that large-scale K3 deployment might necessitate super nodes comprising 64 or more GPUs. Nomura indicates that global competition in large models will encourage leading labs and hyperscale cloud platforms to maintain capital expenditure, benefiting TSMC, NVIDIA, storage vendors, optical module suppliers, and data center operators.

However, Bank of America Securities warns that if model efficiency improvements outpace workload growth without corresponding usage expansion, AI infrastructure development could face a slowdown.