Why the AI Productivity Dividend Will Take Longer Than Markets Expect
Every general-purpose technology of the past century took a decade or more to register in productivity statistics. Generative AI will not be an exception.
The history of general-purpose technologies offers a humbling lesson for those expecting an imminent productivity dividend from generative AI. Electricity took roughly four decades to translate factory electrification into meaningful manufacturing productivity gains. The personal computer took two decades.
Generative AI shows every sign of following the same pattern. Enterprise pilots have proliferated, but the share of those that have crossed into production deployment with measurable bottom-line impact remains under 25 percent according to a recent Bain survey.
The bottlenecks are organisational rather than technological. Workflow redesign, data infrastructure, change management and talent reallocation are all multi-year programmes — and the firms doing them seriously are doing them quietly.
The implication for capital markets is straightforward. The infrastructure cycle will lead the productivity cycle by many years, and the eventual dispersion in returns within nominally "AI-exposed" sectors will be considerably wider than current valuations suggest.
More from Insights
The Coming Reckoning in Private Credit
The asset class that quietly absorbed the run-off of post-GFC bank deleveraging now faces its first genuine credit cycle. The results will not be uniform.
Inside the Mind of the Modern Allocator: A Conversation With Hala Al-Fayez
The chief investment strategist at one of the Gulf's largest sovereign allocators on the rebuilding of duration, the limits of private markets, and AI as an investment process.
The Fragmentation of Global Capital Markets
Three decades of globalisation produced a deeply integrated global capital market. The next decade will produce something more fractured — and more interesting.