The AI Economy Is Growing at 2,600% a Year — and GDP Can't See It
A new paper estimates U.S. AI GDP at roughly $250 billion in 2025, surging at rates conventional statistics completely miss. The authors warn policymakers are flying blind into a potential labor-market shock.
The U.S. AI economy is growing at roughly 2,600 percent per year in quality-adjusted real terms, according to a new paper from economists at the University of Virginia, Anthropic, and the Bank of Canada. Their estimate puts nominal AI GDP at approximately $250 billion in 2025 — a figure that barely registers in conventional economic statistics.
Why the blind spot? The datacenter construction boom is large but not large enough to meaningfully lift aggregate GDP. The real action is in inference — the running of AI systems — where per-unit prices collapse almost as fast as quality-adjusted output climbs. Nominal revenues grow only modestly, masking an explosion in underlying capacity. The raw numbers tell a starker story: U.S. compute spending jumped from $37 billion to $90 billion to $219 billion over the past three years. Actual computing capacity, boosted by more efficient chips, grew over 200 percent annually. Layer on algorithmic progress and quality-adjusted AI output surged roughly 2,290 percent in 2024 and 2,271 percent in 2025.
The paper draws a pointed contrast with past technology waves. Semiconductors and the internet also posed measurement headaches, but those technologies were complements to human labor at the aggregate level. AI, the authors argue, is "the first plausible candidate for large-scale technological mismeasurement in which the rapidly improving sector may become a substitute for human labor." A finance ministry projecting revenues off conventional data will materially underestimate the probability of a labor-tax-base shock — and won't be ready to design responses like tax reform or sovereign wealth funds.
The authors propose three fixes: create "AI satellite accounts" within statistical agencies to track measures like nominal compute spending, forge partnerships between agencies, companies, and academia to generate better primary data, and bake AI productive-capacity measurements into medium-term economic projections. Their bottom line is blunt: a windfall that cannot be seen cannot be shared.
The Wire · Newsletter
One careful email,
every Monday.
The week's most important AI stories, lightly edited and personally vouched for. No autoplay, no spam, easy to leave.
Comments · 0
Sign in to join the discussion.
Be the first to leave a thought.