What happened
US AI compute spending surged from $37 billion in 2023 to $219 billion in 2025, yet traditional GDP metrics do not fully capture this impact. Economists Anton Korinek and Patrick McKelvey estimate that, when adjusted for a 94% drop in AI output (inference) prices over the same period, real GDP could have been, as an upper bound, four percentage points higher in 2025. This massive multiplier effect between nominal spending and quality-adjusted AI output highlights a significant measurement gap in conventional economic reporting.
Why it matters
The rapid decline in AI inference costs, coupled with substantial investment, means current economic models understate AI's contribution to productivity and GDP. Financial analysts must recognise that nominal AI spending metrics do not reflect the exponential increase in quality-adjusted AI output. This measurement gap limits accurate valuation of AI initiatives and could lead to misinformed capital allocation decisions, echoing concerns about soaring AI costs forcing enterprise rethink.
