美国经济在2025年增长2.2%，但同期月均仅新增约1.5万个岗位、年就业增速仅0.1%，形成了看似“高产出-低招聘”的分化。尽管如此，四季度实际GDP年化增速仅1.4%，且按实际GDP与总工时推算的2025年生产率增速约1.9%，低于约2%的长期均值，也明显不及互联网繁荣期。

产出与就业的缺口未必来自AI本身：自1950年以来，这一缺口在近三分之一年份至少达到2个百分点，而2025年上半年约90%的GDP增长被估计来自数据中心等资本开支。与此一致，剔除投资拉动后的“基础”生产率接近零，同时移民收紧与临时就业下滑通过压缩低生产率岗位，机械性抬高了人均产出指标。

AI扩散在加快但使用强度仍低：美国职场生成式AI使用率从2024年末约30%-31%升至2025年11月的36%-41%，而每日使用者仅约13%，相关工时占比也只从4.1%升至5.7%。单项任务效率提升虽可达12%-40%（更广泛研究多在15%-30%），但综合采用率、强度与任务收益后，对过去一年生产率增速的总贡献仅约0.25-0.5个百分点，且高管每周仅约1.5小时使用、九成看不到可测劳动生产率改善，显示组织重构仍处早期。

America’s economy grew 2.2% in 2025, but employers added only about 15,000 jobs per month on average, equal to just 0.1% annual employment growth, creating an apparent high-output, low-hiring split. Even so, real GDP rose at only a 1.4% annualized pace in Q4, and 2025 productivity growth inferred from real GDP and aggregate hours was about 1.9%, below the roughly 2% long-run norm and well short of the internet-boom era.

The output-employment gap does not necessarily prove an AI surge: since 1950, gaps of at least 2 percentage points have appeared in nearly one-third of years, and an estimated 90% of first-half 2025 GDP growth came from data-center and related capital spending. Consistently, investment-adjusted measures put underlying productivity near zero, while tighter immigration and a fall in temporary work mechanically lift average output per worker by shrinking lower-productivity sectors.

AI adoption is rising, but usage intensity remains limited: workplace generative-AI use increased from roughly 30%-31% in late 2024 to 36%-41% by November 2025, yet only about 13% use it daily and AI-linked work hours rose only from 4.1% to 5.7%. Task-level gains can be large at 12%-40% (with broader evidence often 15%-30%), but combining adoption, intensity, and per-task effects implies only about a 0.25-0.5 percentage-point boost to productivity growth over the past year, and with executives using AI about 1.5 hours weekly and 90% seeing no measurable labor-productivity gain, organizational rewiring still appears early.

Source: The AI productivity boom is not here (yet)

Subtitle: Artificial intelligence is improving fast. Its effect on output, not so much

Dateline: 2月 26, 2026 04:26 上午