Economists say markets are pricing in a 32.6% AI productivity gain

Economists at the University of California, Berkeley and the London School of Economics and Political Science estimate that between November 2022 and December 2025 news about AI raised the market's expected present value of software engineering productivity by the equivalent of a permanent 32.6 percent gain. The result appears in "The Macroeconomic Effect of AI: Sizing the Software Engineering Channel," issued in September 2026 as NBER Working Paper 35793 and CEPR Discussion Paper 21944, and reviewed by The Register on 29 September.

Reading productivity out of share prices

The authors do not measure developer output directly. They estimate how sensitive each company's stock price is to an AI stock market index, and how that sensitivity depends on the share of the firm's payroll that goes to software engineering. Companies that spend more on engineering talent react more sharply when AI shares rise. An economic model then maps that cross-sectional relationship into an implied productivity gain.

Chen Lian, assistant professor of finance at UC Berkeley, told The Register: "We empirically measure whether firms with larger software engineering payroll shares experience larger stock-price increases when the AI stock index rises."

From code to GDP

Feeding the estimate into the model produces a 3.6 percent increase in the level of GDP. When higher software engineering productivity also lifts research and development productivity, the figure reaches 6.5 percent. The authors note that 32.6 percent is comparable to the 21 to 56 percent task-level speed-ups reported in the experimental literature (Peng et al. 2023; Paradis et al. 2025), and that two opposing forces explain why the market-based estimate lands in that range.

By mid-2026, as coding agents such as Claude Code and Codex spread through engineering teams, the paper reports that the effect on productivity and GDP had more than doubled relative to the end of 2025.

Why the number is not a promise

The estimate reflects what public markets expect rather than what has already happened, and The Register notes that economists warn markets can get carried away. The paper's introduction points to the wide range of existing forecasts, from roughly one percent of GDP growth over a decade to scenarios in which AI lifts growth rates by an order of magnitude. The authors describe their measure as a forward-looking one that is available in real time.