人工智慧（AI）的大幅重新定价正在重创亚洲市场。韩国综合股价指数（Kospi）暴跌8%，自6月高点以来的累计跌幅扩大至近40%，其中SK Hynix与Samsung股价跌幅皆超过9%，外资更于本月抛售逾100亿美元的韩国股票。同时，日本日经225指数下跌近5%。引发抛售的导火线除了晶片板块，还包括美国通膨加剧重燃对联准会收紧政策的担忧，以及CoreWeave的债务困境，促使投资人直面AI基础建设背后的庞大融资帐单。

此种紧张局势亦体现在国际货币基金组织（IMF）的最新展望中。尽管IMF受惠于AI对韩国及台湾等经济体的强劲推动力，将该地区2026年成长预测上调至4.8%，但其预测明年成长将放缓至4.4%，并警告当前缓冲亚洲的引擎未来可能成为脆弱点。AI投资或估值的修正将重创位于全球科技供应链核心的亚洲，而孟加拉、柬埔寨与寮国等石油净进口国，更面临承受高能源价格成本却未享受到AI红利的双重困境。

此外，Anthropic的财务数据精准反映了市场对AI争论的转变。该公司公布了亮眼的季度营收168亿美元、正自由现金流以及85%的调整后毛利率，这些指标有望提振预计于年底进行的首次公开募股（IPO）。然而，在资本昂贵的环境下，投资人已不再仅关注AI能否带来爆发性成长，而是转向检验维持其运作所需的庞大运算与基础设施支出能否赚取足够回报，市场关切焦点已从AI能发展得多大转变为谁将为其买单。

The significant repricing of AI is hitting Asian markets hardest. South Korea's Kospi plunged 8%, extending its decline from the June peak to almost 40%, as SK Hynix and Samsung each dropped over 9%, and foreign investors offloaded more than $10 billion in Korean shares this month. Concurrently, Japan's Nikkei 225 fell nearly 5%. Catalysts extend beyond semiconductor stocks: rising US inflation revived fears of Fed monetary tightening, while distress surrounding CoreWeave's debt forced investors to confront the massive financing burden backing the aggressive AI infrastructure expansion.

This tension is mirrored in the IMF's latest economic outlook for Asia. While the fund raised its 2026 regional growth forecast to 4.8% due to AI tailwinds bolstering economies like South Korea and Taiwan, it projects expansion will decelerate to 4.4% next year. The IMF warns that the primary growth driver could become a major vulnerability if AI valuations correct, severely impacting key tech supply chain participants. Meanwhile, net oil importers including Bangladesh, Cambodia, and Laos face elevated energy costs without capturing significant AI-driven windfalls.

Anthropic's latest metrics capture the evolving market narrative. The company posted $16.8 billion in quarterly revenue, positive free cash flow, and an 85% adjusted gross margin, strengthening prospects for an expected year-end IPO. However, in an era of costly capital, investors are no longer satisfied purely with exceptional top-line expansion; they increasingly question whether massive compute and infrastructure expenditures can generate adequate returns, shifting the focal inquiry from the scale of AI growth to determining who will ultimately fund the buildout.