Economists say AI gains may be outpacing official data
A Peterson Institute brief estimates AI generated about $250 billion in activity in 2025, but critics question the scale of the claimed impact.
By Daniel Okafor · Business Editor
3 min read
A new Peterson Institute for International Economics policy brief argues that artificial intelligence may already be contributing far more to the economy than government statistics show. The finding matters because policymakers use productivity and GDP data to set tax, labor and spending policy, and the authors say current measures may be missing a fast-growing sector.
The brief was written by Anton Korinek, a nonresident senior fellow at Peterson and head of Transformative AI Economic Studies at the Anthropic Institute, and Patrick McKelvey, a senior data scientist at the Bank of Canada. They argue that AI-related activity is hard to see in official accounts because it is spread across categories such as software, cloud services and data processing.
Korinek and McKelvey also say official data systems struggle to reflect how quickly AI tools are improving. A service that delivers the same result at sharply lower cost can create economic value, they argue, even if conventional output measures do not fully capture the change.
How the brief measures AI
The authors estimate that AI produced about $250 billion in economic activity in 2025. Fortune reported that they compared that scale with the size of the U.S. airline industry.
Korinek and McKelvey estimate that the amount of AI output the industry can produce is rising at about 2,600% a year. They also estimate that the cost of obtaining the same level of AI performance has been dropping by about 94% annually, meaning spending on AI can buy far more capability than it did a year earlier.
Rather than relying on official economic statistics, the authors built their estimates using GPU rental rates, electricity use, AI inference prices and measures of progress in training algorithms. They calculate that if U.S. statistics adjusted for AI’s rapid capability improvements, economic growth in 2025 could look about 4 percentage points higher.
The authors caution that the 4-point figure is an upper-bound estimate, Fortune reported. In other words, they present it as a maximum plausible effect, not as their main forecast.
Korinek and McKelvey propose a separate statistical framework for AI, similar to the way governments track energy or international trade. Their proposal would collect AI activity across industries and adjust for improvements in the technology’s performance.
Why some economists are skeptical
Diane Coyle, Bennett Professor of Public Policy at the University of Cambridge, told Fortune that she agrees there is a measurement problem but questions the size of the brief’s estimates. She said AI is often used as an input in other products and services, while GDP focuses on final goods and services sold to consumers.
Korinek and McKelvey acknowledge that AI is largely an intermediate input, according to Fortune, and cite that as a reason the 4-point growth estimate should be treated as a ceiling. Coyle’s concern is that AI’s economic value depends on whether it improves the finished product or service enough to register in broader economic measures.
Coyle also told Fortune that there is limited systematic evidence so far that AI is raising productivity at the company level. Faster work by individual employees may fail to lift overall output if other teams or processes create bottlenecks.
“I think AI is a significant technology. It will have these great effects, but I think both the speed and scale that this paper is claiming for it are overdone,” Coyle said.
The dispute leaves economists with two linked questions: how much AI is changing production now, and whether the tools used to measure output can keep up with that change. Korinek and McKelvey argue governments should build better measurement systems before the gap affects policy decisions.
This story draws on original reporting from Fortune.