美国 AI 资本支出已占 GDP 的 1.5%，与 1860 年代铁路投资占比相当。McKinsey 预测未来五年在芯片与数据中心上的投入将达 5.2 万亿美元。科技股高度集中：“七巨头”占 S&P 500 超过三分之一，而 1880 年代铁路公司曾占美国股市约 60%。OpenAI 自 2015 年成立后估值已达 5000 亿美元，并预计到 2029 年将消耗 1150 亿美元现金。历史先例呈现同样的建设冲动：1868–1871 年期间美国铁路新建里程达 33,000 英里，并伴随激进财务操作与投机周期。
铁路泡沫曾引发 1837、1857、1873、1893 四次重大恐慌，其中 1873 年因跨大陆铁路融资失败导致 Cooke & Co. 倒闭并引发 1873–1890 的“长期萧条”。19 世纪铁路巨头如 George Hudson 曾控制英国约四分之一的铁路资产（1,000 英里），其扩张依赖极端推销而最终破产。同样，今日科技企业提前在盈利前过度投资，并发行债券、构建相互竞争的 AI 体系。AI 人才争夺激烈，薪酬可达 2 亿美元，且公司治理动荡，如 OpenAI 董事会短暂罢免其首席执行官。
历史模式指示警惕：尽管现代科技公司资本实力与市场监管均优于 19 世纪，部分融资结构仍呈现循环特征，数据中心与芯片的投资规模与真实需求不匹配。铁路泡沫对普通投资者与金融系统有害，却为企业主积累巨额财富，并在长期内带来生产率提升与经济成本下降。AI 可能同时具备泡沫性与变革性，其未来效益虽未定，但当前已显著降低信息搜寻与组织事务的成本。
US AI capital spending has reached 1.5% of GDP, matching railroad investment levels in the 1860s. McKinsey forecasts $5.2 trillion in chip and data-center spending over the next five years. Tech-market concentration is extreme: the Magnificent Seven comprise over one-third of the S&P 500, echoing the 1880s when railroads were roughly 60% of US market value. Founded in 2015, OpenAI is valued at $500 billion and expects to burn $115 billion in cash by 2029. Historical precedent mirrors today’s build-out: from 1868–1871 the US laid 33,000 miles of new track, fueled by aggressive financial engineering and speculative manias.
Railroad excess triggered the major panics of 1837, 1857, 1873 and 1893; in 1873 overleveraged transcontinental projects collapsed, taking down Cooke & Co. and initiating the 1873–1890 Long Depression. Figures such as George Hudson, who controlled 1,000 miles of British rail (about a quarter of the network), expanded through extreme promotion before financial ruin. Today’s tech firms similarly overbuild ahead of demand, issue bonds and race to construct rival AI systems. Talent bidding resembles a frenzy, with pay packages reaching $200 million, while governance instability persists, as seen in OpenAI’s brief ousting of its CEO.
Historical patterns counsel caution: though today’s firms are better capitalized and markets more diversified, some financing resembles past circular structures, and investment volumes in chips and data centers exceed proven consumer willingness to pay. Railroad bubbles harmed investors and destabilized finance but ultimately delivered major productivity gains and lower economic costs. AI may be both bubble and transformative force; its long-run payoff is uncertain, yet it already reduces the cost of information management and everyday tasks.