AI Spending Boosts U.S. Economy, but Productivity Gains Remain Unproven
nashnova research
AI-related capital expenditure is set to account for one-third of U.S. economic growth in 2026, yet total factor productivity data show the spending has delivered no measurable efficiency gain — growth is coming from buying more machines, not from a technology breakthrough.
How exactly is AI driving economic growth?
ING chief international economist James Knightley estimates AI will contribute one-third of U.S. growth in 2026, spanning computing equipment, software, and data centers.
A St. Louis Fed study in January corroborates: in Q3 2025, AI-related categories accounted for 39% of total U.S. GDP growth.
This means → AI is boosting the economy in a surprising way — not by making workers more productive, but through sheer volume of corporate and government spending on hardware and facilities.
Is the government piling in too?
The Pentagon proposed nearly $30 billion in its FY2027 budget for a new "AI Arsenal" program to build government-owned AI infrastructure.
This means → AI spending is no longer just a tech-sector story. The federal government is becoming another major buyer, amplifying total investment.
What are the side effects of all this spending?
U.S. electricity prices rose 3.8% year-on-year in August, as massive data-center power demand pushed inflationary pressure beyond the tech sector.
In plain terms = data centers consume enormous amounts of electricity; higher demand drives up prices, and ordinary households end up paying the bill.
Has all this spending actually improved efficiency?
Apollo Global Management chief economist Torsten Slok uses total factor productivity — TFP, which measures how much more an economy can produce without adding extra labor or machines — as the closest proxy for genuine technological progress.
San Francisco Fed data show that since the AI capex cycle began, utilization-adjusted TFP has shown no acceleration at all and currently sits slightly below zero.
This means → by the metric that best captures technology-driven improvement, AI has left virtually no trace so far.
Then how do we explain the rise in output per hour?
U.S. output per hour is running at roughly 2.5%, well above its post-2005 average — a figure often cited as evidence that "AI is already working."
But Slok points out: strong output per hour alongside flat TFP is a hallmark of capital deepening, not a technology shock.
In plain terms = output is growing because firms bought more machines and hired more workers — not because AI made the same people and machines more efficient.
So will AI's productivity payoff ever materialize?
Slok concludes that AI is currently getting more credit than it deserves.
This reflects a central unresolved question in the AI narrative: whether the productivity dividend is a matter of "when" or "whether" will determine if this capex wave turns out to be a sound investment — or an expensive bet.
市场有风险，内容仅供研究参考，不构成投资建议。