美国股市陷入所谓“AI恐慌交易”——投资者在美国出售软件股与财富管理股，担忧AI快速进步会蚕食既有商业模式；而中国市场情绪显著乐观，投资者更多关注AI带来的增长前景与终端用户成本节约。中国的本地AI模型发布者如 MiniMax Group Inc. 与 Knowledge Atlas Technology JSC（Zhipu）在2月股价均“翻倍以上”，并因 Morgan Stanley 等华尔街机构的看涨评级而吸引资金，从传统互联网巨头向纯AI标的轮动。

这类分化的核心原因之一是市场结构。Gary Tan 指出，由于监管和地缘政治限制，外国大语言模型对国内市场的进入受限，给予本土模型开发者更大空间；同样，A股与港股公开市场上可投资的LLM建模公司稀缺。Zhipu 和 MiniMax 均于1月在香港上市后，股价分别上涨524%与488%；此外，上海寒武纪？不是该文应是 Shanghai Biren 上涨80%以上（自1月2日）， Montage Technology 自2月9日上市后上涨98%以上。OpenAI 接近以1×10^11美元（$100 billion）的规模募资，估值或超8.5×10^11美元（$850 billion）；Anthropic 当月以3×10^10美元（$30 billion）资金完成估值3.8×10^11美元（$380 billion）。

当前情绪呈“再点燃”态势：任何新模型与商业化进展都被视为催化剂。Zhipu 的 GLM-5 超越 Moonshot，在国际开源模型基准上登顶；DeepSeek 的下一代模型发布预期也在推升行业共振，且更低成本的AI能力有望提升用户采用率。Morgan Stanley、Jefferies、UBS 均给出 MiniMax 等的买入同等评级，Morgan Stanley 预估其2027年收入可达7×10^8美元（$700 million），对应约10倍增长预期；资金正在向纯AI公司集中，阿里巴巴、腾讯等多元平台出现部分获利回吐。但市场也有人警惕：若盈利增长跟不上预期，重估可能难以持续，且模型替代风险或最终压低更广泛的企业利润。

The U.S. market is caught in an “AI scare trade,” with investors selling software and wealth-management stocks over fears that rapid AI advances will erode incumbents’ profit pools, while China is broadly upbeat and sees AI mainly as a source of growth and user cost savings. Chinese investors favor firms that have launched or upgraded models, especially MiniMax Group Inc. and Knowledge Atlas Technology JSC (Zhipu), whose shares rose more than two times in February, supported by bullish coverage from firms including Morgan Stanley and a rotation away from diversified internet giants. (Key numbers: 2)

A key driver of the divergence is structural: China’s AI ecosystem is relatively insulated by regulation and geopolitics, giving local model makers more room as large foreign language-model players face limited access. Both Zhipu and MiniMax listed in Hong Kong in January and were up 524% and 488%, respectively; Chinese chip names also surged, with Shanghai Biren up more than 80% since its Jan. 2 debut and Montage Technology up more than 98% since trading began Feb. 9. Global leaders OpenAI and Anthropic are still private; nevertheless their fundraising reinforces the AI narrative, with OpenAI nearly raising 1×10^11 USD ($100 billion) toward a valuation that could exceed 8.5×10^11 USD ($850 billion), while Anthropic recently raised 3×10^10 USD ($30 billion) at 3.8×10^11 USD ($380 billion) valuation.

Each new AI release in China is treated as a catalyst for both builders and users, especially after ByteDance’s media-app launch and Zhipu’s GLM-5, which outperformed peers and topped a global open-source benchmark ranking. Anticipation around DeepSeek’s next model and its stronger cost competitiveness is also lifting sentiment across the sector. Analysts at Morgan Stanley, Jefferies, and UBS have buy-equivalent views on MiniMax, with Morgan Stanley projecting revenue could reach 7×10^8 USD ($700 million) by 2027—up to a tenfold increase in two years. That has driven funds toward pure AI names while diversified names like Alibaba and Tencent face some profit-taking, though some market participants warn valuation support could fade if earnings fail to keep pace and disruption later broadens to hit corporate profits.