Goldman Sachs: Kimi K3 Launch Triggers Market Reactions; AI Compute Dominance May Be Coming to an End

According to Chaoxiang Research, a July 18 report by Goldman Sachs noted that Moonshot AI released the Kimi K3 model with 2.8 trillion parameters, surpassing Claude 3 Opus and GPT-5.6 Sol to top the Arena.ai programming leaderboard, with an API pricing of $2.3 per million tokens—the highest ever for a Chinese model. Two days after its release, Zhipu AI’s stock fell 28%, MiniMax dropped 16%, Nasdaq-100 futures declined over 1.8%, and the Philadelphia Semiconductor Index fell more than 18% from its peak. Goldman Sachs views Kimi K3 as a turning point: a Chinese lab unable to match the scale of Western pretraining compute power has rapidly narrowed the gap with top U.S. models through architectural innovation and reinforcement learning, demonstrating that “scaling” is no longer the only path to success. The firm warns that the era of “more compute is better” may be ending, requiring a fundamental reassessment of AI infrastructure investment logic built on the assumption that greater compute capacity equals superior performance. Goldman Sachs maintains a “Buy” rating on MiniMax and a “Neutral” rating on Zhipu AI. Future focus will be on the imminent launches of other 2–5 trillion parameter models, including Zhipu’s GLM, Alibaba’s Qwen, and MiniMax’s M3 Pro.