Nvidia shares and AI semiconductor stocks sold off following the release of the Kimi 3 model by the China-based start-up Moonshot.
Nvidia shares and most of the AI-related semiconductor sector experienced a sell-off last week after Moonshot released its Kimi 3 model. While the model displayed impressive performance against leading frontier models from Anthropic and OpenAI, the reaction is viewed by many analysts as a buying opportunity rather than a long-term threat. The Kimi 3 model introduced Kimi Delta Attention (KDA), an architecture that significantly reduces memory requirements and increases processing speed. However, the model remains compute-intensive, requiring high-end hardware such as Nvidia’s latest racks to run efficiently. Because the model is massive, it necessitates high-bandwidth memory and advanced networking, which favors Nvidia’s NVLink technology. Furthermore, the Jevons paradox suggests that as technology makes resource use more efficient, overall consumption typically increases, potentially driving higher adoption of AI. While some investors feared the Chinese model might reduce the need for expensive U.S. chips, the high costs of running such a large model and the potential for "pirated" distilled weights mean U.S. enterprises will likely continue to favor high-end hardware.