General-purpose technologies (GPTs) typically don’t move measured productivity immediately on arrival. The usual sequence is invention → diffusion → complement-building (data, processes, skills, governance) → reorganization of production that finally yields measurable gains. During those middle stages, companies spend on intangibles that accounting treats as expenses rather than capital, so the statistics lag the underlying productive capacity. Economic models describe a J‑curve in measured productivity: a dip while complements are built, followed by an upswing once redesigned processes begin to compound.1
The influential study by Bresnahan and Trajtenberg (1992) puts GPTs at the center of the history of economic development. They argue that whole eras of technical progress and economic growth appear to be driven by GPTs, which they define as technologies that are pervasive (used as inputs by many downstream sectors), have inherent potential for technical improvement, and exhibit “innovational complementarities,” meaning that the productivity of Research & Development in downstream sectors rises as the GPT improves. As GPTs diffuse and advance, they spread through the economy and bring about generalized productivity gains.2 This lens captures both the eventual power of GPTs and the long, complement-intensive road to realizing broad-based productivity gains.
Steam, electricity, and digital computing all fit this pattern.3 The familiar shorthand – steam 60 years, electricity 30 years, computing 15 – summarizes the time between core invention and the start of clear macro-level payoff. The question for generative AI is whether that lag can shrink again, perhaps toward seven or eight years, and what it would take for that to happen.
Steam – Invention First, Productivity Much Later
Dating the steam story to James Watt’s separate condenser (1769), the economy-level payoff arrived much later. For decades steam was chiefly stationary power (pumping; then mill drive via rotative engines) until high-pressure designs broadened its scope. Growth-accounting evidence shows steam’s measured contribution to British productivity is minimal before the 1830s; larger effects appear only in the mid- to late-19th century, as steam spreads across industry and transport – especially with the diffusion of locomotives and railways and the rise of large steam-driven mills and weaving sheds – supported by complements such as tracks and depots, signaling and timetables, maintenance shops, and new skills. In short, the engine preceded the economy by roughly six decades – a textbook GPT productivity lag.4
Electricity – A Motor in the 1890s, a Payoff After Factory Redesign
By the 1890s, core electrical technologies – generators, distribution, and the alternating-current induction motor – were established. Yet the big productivity response does not appear until the 1920s–1930s, once factories abandon the steam-era, line-shaft layout. Simply replacing a central steam engine with a single large electric motor did little; the gains arrived when plants reorganized around unit drives (small motors at each machine), enabling one-story factories, flexible flows, and tighter control. Electrification’s payoff was governed by organizational redesign and took decades.5
Computing – A Mid‑1990s Break After Process Replatforming
The personal computer of 1981 is an emblem, but the decisive break in productivity occurs after 1995. Industry-level and growth-accounting studies attribute much of that acceleration to both the production and the use of information and communication technology, with especially strong gains in industries that replatformed workflows, supply chains, and retail formats around digital systems. Buying hardware was not enough; use mattered – meaning redesigned processes and skills to match.6
