Moonshot AI Executive Says AI Bubble Exists but Enterprise ROI Is Clear

Moonshot AI’s Kimi enterprise lead Huang Zhenxin said the AI industry has signs of a bubble, but argued that underlying fundamentals remain strong and that businesses can already calculate return on investment from AI deployments. According to ChainCatcher, Huang said AI-driven productivity gains have already undergone a substantive shift, with enterprise users increasingly able to quantify ROI. Addressing recent price increases by model providers, Huang attributed the trend to rising global computing costs and chip capacity that cannot keep pace with a surge in token demand. He said model cost-effectiveness should not be judged only by input and output unit pricing, but also by cache hit rates, which can materially reduce real compute costs. Huang said Kimi’s native cache hit rate has exceeded 90%. Huang also said Kimi plans to continue pursuing underlying architecture innovation to extend scaling-law progress. He added that Kimi’s Muon optimizer, which he said was first validated at scale by the company, has since been adopted by multiple mainstream large-model developers. On the “last mile” of enterprise AI implementation, Huang said that as base model capabilities improve, application-layer technical approaches are expected to become simpler over time.