AI Computing Expansion Fuels Inflation, Putting the Fed in a Policy Dilemma
Nashnova编辑部
U.S. AI infrastructure capex is set to hit $581 billion this year — 1.8% of GDP — yet only 17%–20% of American firms report using AI at all. Costs are arriving now; the productivity payoff is not, and that mismatch is complicating the Fed's inflation calculus.
Silicon Valley promised AI would make things cheaper — so why are prices rising?
Altman wrote of "near-free intelligence," Musk predicted extreme abundance, and Son forecast a 40% price drop — but per CNBC, none of these visions are close to reality.
The opposite is happening: Goldman Sachs estimates U.S. AI capex at $581 billion this year, with global spending reaching as high as $1 trillion.
This means → the data-center and AI-infrastructure buildout is straining supply chains and pushing up power costs — the money is already out the door, but the deflation payoff hasn't arrived.
All that spending — are companies actually using AI?
A U.S. Census Bureau survey from May: only 17%–20% of American firms report using AI, with adoption far higher at large companies than small ones.
OpenAI chief economist Ronnie Chatterji acknowledged that AI must be widely adopted to move the economic needle — "it will take time before it shows up clearly in productivity statistics."
In plain terms = most firms haven't started, and those that have aren't going deep — the day AI actually lowers prices is still far off.
How much can productivity actually rise? What does the internet era tell us?
Analyst Peter Boockvar compared AI to the last major tech-driven productivity boom — the internet era: U.S. productivity rose just 1.5% over 30 years, with a 50-year average of 2.5%.
His blunt take: "To assume generative AI delivers gains far beyond what the internet did — that's a stretch."
This reflects a sober judgment: technology does raise productivity, but a multi-order-of-magnitude leap has no historical precedent.
The tech is ready — so what's the bottleneck?
Former Lululemon CIO Julie Averill, who led the company's AI rollout, points to organizational friction — "getting people to change behavior, trust the model — that's the hard part."
Stanford professor Charles Jones coined the term "weak links" — tasks within a job that are far harder to automate than others. Jobs are bundles of tasks, and the hardest-to-automate ones are the structural reason AI's productivity dividend isn't arriving in full.
Chatterji cited OpenAI data: power users consume 8× more tokens per user than average firms, a gap that widened from 2× to 8× in just three months — the gulf between leaders and laggards is accelerating.
What does this mean for the Fed?
On one side, AI infrastructure buildout is already pushing up power and supply-chain costs — the inflationary effect is observable and happening now.
On the other, the timeline for productivity gains remains uncertain — the deflationary payoff is still on the horizon.
This means → the mismatch — costs first, dividends later — prevents the Fed from pricing "AI deflation" into near-term policy — inflation pressure is immediate; the deflation promise is distant.
Content is for reference only, not financial advice.