Risk and Return under Open-Source LLM Shocks: A CAPM Analysis of U.S. AI-Related Sector ETFs
TL;DR
This paper examines the impact of two landmark open-source releases—DeepSeek-R1 and Kimi K3—on the returns and risk of five U.S. AI-related sector ETFs, finding that both shocks were highly concentrated in the computing-power layer, while the application layer remained relatively resilient.
Abstract
The rapid iteration of open-source large language models is reshaping the competitive landscape of the AI industry, yet systematic empirical evidence on their specific effects on capital markets remains limited. Using the CAPM as the analytical framework and combining it with the event-study methodology, this paper examines the impact of two landmark open-source releases—DeepSeek-R1 and Kimi K3—on the returns and risk of five U.S. AI-related sector ETFs. Three findings emerge: First, both shocks were highly concentrated in the computing-power layer, while the application layer remained relatively resilient, presenting a short-term pattern of repricing within the AI industry chain. Second, the Kimi K3 shock produced shallower losses and a faster recovery, which may indicate that the market had developed a certain degree of immunity to such events. Third, following the DeepSeek shock, the risk structure of the sectors underwent changes that persisted throughout the sample period: the beta of SOXX increased, whereas the betas of software, communication services, and utilities declined markedly, with AI-related systematic risk exhibiting a tendency to concentrate in the core semiconductor segment.