从历史看，1900 年以来全球 10 次主要股市泡沫平均持续略高于 2.5 年，从谷底到峰值上涨 244%。本轮行情已进入第三年：自 2022 年底以来，标普 500 累计上涨 79%，那斯达克 100 上涨 130%。集中度也是警讯：六档 AI 相关超大型股合计接近标普 500 的 30%，而前 10 大成分股约占 40%——上一次到这种水准要追溯到 1960 年代（1930 年代也有前例）。
基本面与估值让争论分歧。相较网路泡沫时代如 WorldCom 等公司，如今龙头的负债对盈余比更低，且部分企业已回报 AI 带动的获利成长；但以举债支撑的支出正在上升：甲骨文在 9 月 24 日发行 180 亿美元债券后，股价隔日下跌 5.6%，此后累计下跌 37%；Meta、Alphabet 与甲骨文在 2026 年可能合计需要筹资 860 亿美元。以席勒循环调整本益比（CAPE）衡量，标普 500 估值接近历史高位（仅低于 2000 年代初），但英伟达的倍数低于 50 倍，而 2000 年思科一度达 200 倍。市场审视也更强：11 月出现逾 12,000 则提到 AI bubble 的报导；美银调查亦将其列为最大尾端风险。
The AI-led surge pushed US equities to records: the S&P 500 rose 16% in 2025, led by Nvidia, Microsoft, Alphabet, Broadcom and others. Big Tech’s planned buildout is vast: Microsoft, Alphabet, Amazon and Meta capex is projected to jump 34% to about $440 billion over the next year, while OpenAI has pledged more than $1 trillion despite being unprofitable, raising worries about circular funding between it and listed partners.
Historically, 10 major equity bubbles since 1900 lasted a little over 2.5 years on average and gained 244% from trough to peak. The current run is in its third year: since end‑2022 the S&P 500 is up 79% and the Nasdaq 100 130%. Concentration is another signal: six AI-heavy mega‑caps are nearly 30% of the S&P 500, and the top 10 stocks are about 40%—levels last seen in the 1960s (with precedents in the 1930s).
Fundamentals and valuation split the debate. Today’s leaders carry lower debt-to-earnings than dot‑com-era laggards like WorldCom, and some already show AI-driven profit growth, yet debt-funded spending is rising: after Oracle sold $18 billion of bonds on Sept. 24, its shares fell 5.6% the next day and are down 37% since; Meta, Alphabet and Oracle may need $86 billion in 2026. Shiller’s CAPE is near a record (below only the early 2000s), but Nvidia’s multiple is under 50× versus Cisco’s 200× in 2000. Scrutiny is intense: over 12,000 AI bubble stories ran in November, and a Bank of America poll tagged it as the top tail risk.
