AI安全担忧引发市场剧烈轮动：AI争论转向前沿模型快速无序发展的安全风险。9月12日Anthropic CEO呼吁有节制地推进能力升级并引入第三方评估，其他美国AI巨头随后跟进，市场资金迅速撤出半导体、资本品等AI资本开支受益板块，转向食品饮料等低AI敞口的防御性板块。美银科技分析师认为，巨大的竞争压力与经济利益意味着美国AI实验室和政府无法承受落后风险；即便监管收紧导致前沿模型训练需求下降，用于「监管」（AI监控、验证与遏制）的算力需求也可能形成对冲。我们则持怀疑态度：竞争已十分激烈、模型定价承压，利润率存在被压缩风险；若美国实验室放缓而中国低成本竞争者不放缓，能力差距将收窄并进一步加剧价格战；AI安全争论还强化了政治反弹风险，中期选举若民主党大胜可能加剧监管收紧。Ramp数据显示AI采用速度放缓、企业支出下降，韩国与台湾对美出口增速也大幅放缓。
AI safety concerns trigger a sharp rotation: the AI debate has shifted towards the risks of rapid and unconstrained frontier model development. Following calls from leading US AI companies for pacing capability advances and independent evaluation, markets rotated sharply away from AI capex beneficiaries such as semiconductors and capital goods and towards defensive, lower-AI-exposed sectors such as food & beverages. Why AI bulls remain confident: BofA tech analysts argue that the enormous competitive pressures and economic stakes in AI model development mean that the US AI labs and government simply cannot risk falling behind. Moreover, even if tighter regulation reduced frontier model training activity, increased demand for "policing" (AI monitoring, verification and containment), could provide an offsetting source of compute demand. Why we remain sceptical: we agree that the strategic incentives supporting AI capex remain strong, but in our view the key question is whether the econom