2026年8月28日,美联储主席凯文·沃什出席在怀俄明州杰克逊霍尔举行的经济政策研讨会。图片来源:Natalie Behring/Getty Images
华尔街分析师正用“疯狂时刻”和“躁动期”来形容眼下的美股。与此同时,一位最看空市场的预测人士警告称,AI泡沫已进入“后期阶段”。
过去几个月,总部位于伦敦的预测机构凯投宏观(Capital Economics)一直在系统论证一个观点:AI交易已经成为“后期泡沫”。高级市场经济学家詹姆斯·赖利考察了八类市场指标,发现其中大多数已经处于或接近历史上重大市场顶部出现前的水平。面对如此严峻的警讯,凯投宏观甚至预测:标普500指数将于明年开始显露崩盘迹象,最终较其高点下跌至少30%,沦为过去一个世纪最惨烈的七次美股暴跌之一。
摩根士丹利则对“泡沫破裂”的警示不以为意。该行全球投资委员会重申,标普500指数今年年底的目标位为8,000点,2027年中期的目标位为8,300点——这仍属于看涨区间,而非崩盘场景。沙莱特的论点基于一个判断:市场已消化空头要求的部分调整。标普500指数的远期市盈率已从今年1月的22.5倍降至目前约19.5倍,但股市并未出现大幅下跌,主要原因是企业盈利同比增长超过30%,为估值压缩提供了缓冲。即便在利率问题上,该委员会也认为,AI基础设施建设“基本不受利率影响”。换言之,无论短期内债券收益率如何变化,美联储即便加息一两次,也不太可能真正阻止超大规模云服务商继续投入资本开支。
凯投宏观屡次引述的历史先例
在预测未来市场走势时,凯投宏观直接将互联网泡沫破裂作为参照模板,而货币政策正是对比的核心。1999年6月至2000年5月间,美联储为遏制经济过热,将政策利率从4.75%一路上调至6.5%。此后,标普500指数累计跌去近一半,且震荡下行的过程持续了长达两年半。
这家分析机构近期一直预测,AI泡沫即便回调,幅度和持续时间也会小于当年的互联网泡沫破裂,原因正是它此前认为,美联储这一轮不会像1999年至2000年那样收紧政策。
即便没有美联储推波助澜,凯投宏观提出的其他论据本身也已经足够令人警惕。市场对标普500指数盈利增长的一致预期,已升至与互联网泡沫时期峰值相当的水平,且增长预期高度集中于科技和半导体公司。四大超大规模云服务商的资本开支持续膨胀,预计到2027年合计自由现金流将转为负值,而过去一年它们的债券发行规模已增长一倍以上。与此同时,新一轮股权发行潮也已初显端倪,Anthropic计划于今年秋季推动的IPO便是明证。在上市计划推进之际,该公司一名研究人员公开宣称:“我们确实极其认真地相信,AI可能会毁灭全人类。”
欧文·拉蒙特在一篇语带讽刺、题为《末日IPO》的博客文章中,将这种说法称为“加强版ESG:灭绝、奇点与护栏”。他指出,上一次有IPO成为市场过热的象征,是2000年2月的Pets.com,后来这次上市被“广泛视为科技股泡沫开始破裂的标志”。他最后写道:“但愿未来的历史学家回望Anthropic的IPO时,不会把它视为人类文明开始崩塌的标志。”(财富中文网)
《财富》杂志使用生成式AI作为本文的研究工具。编辑在发表前已核实信息的准确性。
译者:刘进龙
审校:汪皓
华尔街分析师正用“疯狂时刻”和“躁动期”来形容眼下的美股。与此同时,一位最看空市场的预测人士警告称,AI泡沫已进入“后期阶段”。
过去几个月,总部位于伦敦的预测机构凯投宏观(Capital Economics)一直在系统论证一个观点:AI交易已经成为“后期泡沫”。高级市场经济学家詹姆斯·赖利考察了八类市场指标,发现其中大多数已经处于或接近历史上重大市场顶部出现前的水平。面对如此严峻的警讯,凯投宏观甚至预测:标普500指数将于明年开始显露崩盘迹象,最终较其高点下跌至少30%,沦为过去一个世纪最惨烈的七次美股暴跌之一。
摩根士丹利则对“泡沫破裂”的警示不以为意。该行全球投资委员会重申,标普500指数今年年底的目标位为8,000点,2027年中期的目标位为8,300点——这仍属于看涨区间,而非崩盘场景。沙莱特的论点基于一个判断:市场已消化空头要求的部分调整。标普500指数的远期市盈率已从今年1月的22.5倍降至目前约19.5倍,但股市并未出现大幅下跌,主要原因是企业盈利同比增长超过30%,为估值压缩提供了缓冲。即便在利率问题上,该委员会也认为,AI基础设施建设“基本不受利率影响”。换言之,无论短期内债券收益率如何变化,美联储即便加息一两次,也不太可能真正阻止超大规模云服务商继续投入资本开支。
凯投宏观屡次引述的历史先例
在预测未来市场走势时,凯投宏观直接将互联网泡沫破裂作为参照模板,而货币政策正是对比的核心。1999年6月至2000年5月间,美联储为遏制经济过热,将政策利率从4.75%一路上调至6.5%。此后,标普500指数累计跌去近一半,且震荡下行的过程持续了长达两年半。
这家分析机构近期一直预测,AI泡沫即便回调,幅度和持续时间也会小于当年的互联网泡沫破裂,原因正是它此前认为,美联储这一轮不会像1999年至2000年那样收紧政策。
即便没有美联储推波助澜,凯投宏观提出的其他论据本身也已经足够令人警惕。市场对标普500指数盈利增长的一致预期,已升至与互联网泡沫时期峰值相当的水平,且增长预期高度集中于科技和半导体公司。四大超大规模云服务商的资本开支持续膨胀,预计到2027年合计自由现金流将转为负值,而过去一年它们的债券发行规模已增长一倍以上。与此同时,新一轮股权发行潮也已初显端倪,Anthropic计划于今年秋季推动的IPO便是明证。在上市计划推进之际,该公司一名研究人员公开宣称:“我们确实极其认真地相信,AI可能会毁灭全人类。”
欧文·拉蒙特在一篇语带讽刺、题为《末日IPO》的博客文章中,将这种说法称为“加强版ESG:灭绝、奇点与护栏”。他指出,上一次有IPO成为市场过热的象征,是2000年2月的Pets.com,后来这次上市被“广泛视为科技股泡沫开始破裂的标志”。他最后写道:“但愿未来的历史学家回望Anthropic的IPO时,不会把它视为人类文明开始崩塌的标志。”(财富中文网)
《财富》杂志使用生成式AI作为本文的研究工具。编辑在发表前已核实信息的准确性。
译者:刘进龙
审校:汪皓
Wall Street analysts are describing the current moment as “crazy days” and “silly season” even as one of its most bearish forecasters warns of a “late-stage” AI bubble—and now the Federal Reserve may be about to do the one thing that has historically ended booms like this, over the objections of two of its own governors.
London-based forecaster Capital Economics has spent the past several months building the most detailed public case yet that the AI trade is a “late-stage bubble.” In a September 10 report, senior markets economist James Reilly screened eight categories of market indicators and found most sitting at or near levels that have historically preceded major peaks—a warning serious enough that the firm now forecasts the S&P 500 will start cracking next year, eventually falling by at least 30% from its high, one of the seven worst crashes in the past century.
Morgan Stanley shrugged off the bubble-bursting concerns, with its Global Investment Committee reiterating a year-end S&P 500 target of 8,000 and a mid-2027 target of 8,300—bullish territory, not a bust. Shalett’s argument leans on the idea that markets have already absorbed some of the adjustment that bears are demanding: the S&P 500’s forward price-to-earnings ratio has fallen from 22.5x in January to about 19.5x now, without major equity losses, largely because earnings growth above 30% year-over-year has cushioned the compression. Even on rates, her committee argues the AI build-out “is mostly rate-insensitive,” meaning a Fed hike or two is unlikely to actually stop hyperscaler capital spending, whatever it does to bond yields in the short run.
The precedent that Capital Economics keeps returning to
Capital Economics has pointed directly at the dot-com collapse as its template for what happens next — and monetary policy is central to that comparison. Between June 1999 and May 2000, the Fed raised its policy rate from 4.75% to 6.5% to cool an overheating economy. The S&P 500 fell by nearly half afterward, in a decline that dragged on for roughly two and a half years.
The analysis firm has been recently forecasting a smaller, shorter AI-bubble unwind than the dot-com crash specifically because it did not expect the Fed to tighten this cycle the way it did in 1999 and 2000. But if Warsh hikes Wednesday and signals more increases ahead, it would mark the first meaningful tightening cycle since the AI trade took off in 2023.
The rest of the firm’s case doesn’t need help from the Fed to look concerning on its own. Reilly’s September 10 report found consensus estimates for S&P 500 earnings growth now matching dot-com-era peaks, concentrated overwhelmingly in technology and semiconductor names. The combined free cash flow of the four largest hyperscalers is projected to turn negative in 2027 as capital spending balloons, even as their bond issuance has more than doubled over the past year. And a fresh wave of equity issuance — the clearest signal, in Reilly’s view—is already visible in Anthropic’s planned initial public offering this fall, which arrives amid one of its own researchers publicly declaring that “we really do earnestly believe AI could kill all humans.”
Owen Lamont, in a blog post dryly titled “IPOs of doom,” called that reasoning “ESG on steroids: Extinction, Singularity, Guardrails,” and noted that the last time an IPO became a symbol of excess—Pets.com, in February 2000—it came to be “widely seen as marking the beginning of the collapse of the tech stock bubble.” He concluded, “Let’s hope that future historians do not look back at the Anthropic IPO and see it as marking the beginning of the collapse of human civilization.”
For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.