现在的宏观环境不同。3 月初美国关税约为 11%，接近 1943 年水平，美国政策又回到以国内生产为主轴，因此产业回流美国不太可能压低投入成本。1990s 的财政条件则是：初期赤字高于 4% GDP，但十年末转为盈余。如今即使经济持续成长，赤字仍约为 6%；在移除 2025 年学生债务会计调整后，Treasury 仅为支付既有证券利息就必须新增发行约 1 兆美元债务。这提高了政府对资本市场的需求，也让投资者与政府争抢资金，可能推高中性利率，而非支持更宽松货币政策。
Greenspan 也指出，1990s 的低通膨部分来自就业不安全与工资压抑，但这在当前政治语境难以重提。AI 可能仍提升产出，但也可能削减劳动需求，Citrini Research 曾提出大量 AI 驱动失业的极端情境。反之，如果 AI 投资后出现用工繁荣，劳动供给也可能紧缩：美国劳动力在移民限制趋严与数十万人被遣返后最近明显缩小。Powell 本月表示，联准会已研究 AI 三年，却未发现立刻降息的依据；任何解释 AI 对通膨影响的模型都必须纳入当前的财政与劳动限制，而非只看技术乐观。
The article says the current AI boom is unlike the 1990s internet boom. Chris Anstey says Trump’s team, including Scott Bessent, Kevin Hassett, and Kevin Warsh, argues that AI spending can replicate 1990s-style productivity growth with weak inflation: U.S. tech firms spend hundreds of billions a year on AI and may add about $3 trillion in coming years. They criticize Jerome Powell for high rates and expect Warsh, if confirmed in May, to lower borrowing costs. But Greenspan’s record from 1995 to 2006 shows inflation stayed low because two forces worked together: large gains in computing and telecom spending, and globalization as China and former Soviet economies entered global supply chains, lowering unit labor costs. From 1995, six consecutive years of nonresidential fixed investment added more than 1 percentage point to annual GDP growth, a level unmatched since 1930.
The macro setting now is different. U.S. tariffs were about 11% in early March, near 1943 levels, and policy is once again centered on domestic production, so reshoring is unlikely to cut input costs. The 1990s fiscal backdrop was different too: deficits were above 4% of GDP at first but moved to surplus by decade’s end. Now deficits are around 6% despite solid growth; after removing a 2025 student-debt accounting adjustment, the Treasury must issue roughly $1 trillion in extra debt just to pay interest on existing securities. With stronger government demand for borrowing, investors compete with the state for capital, which can push the neutral rate higher rather than support easier policy.
Greenspan also noted that some disinflation in the 1990s came from job insecurity and wage restraint, a claim difficult to use politically today. AI may still raise output, but it may also reduce labor demand, and Citrini Research has floated a scenario with mass AI-driven job losses. Conversely, if AI spending is followed by hiring strength, labor supply could tighten as a constraint: the U.S. labor force has recently shrunk after tighter immigration limits and deportations of hundreds of thousands. Powell said the Fed has studied AI for three years and has not yet seen grounds for immediate rate cuts; any model explaining AI’s inflation impact must include current fiscal and labor limits, not only technology optimism.