AI as an Economic Force

Economists have a concept called a general-purpose technology, or GPT. A GPT is not just a useful tool — it is a foundational innovation that transforms production across nearly every sector of the economy, enables cascading secondary inventions, and sustains productivity growth over decades. The steam engine, electricity, and the internet each qualify. Artificial intelligence is now widely regarded by economists as the next entrant on that short list. Understanding AI as an economic force means understanding it at this systemic level — not as a single product or application, but as a platform that changes the cost structure of cognitive work the way electrification changed the cost structure of mechanical work.

What Makes a Technology General-Purpose?

Economists Bresnahan and Trajtenberg (1995) identified three hallmarks of a general-purpose technology: pervasiveness (it spreads to most sectors of the economy), improvement over time (it keeps getting better), and innovation spawning (it enables a wave of complementary inventions and new business models). AI exhibits all three. It has spread from tech companies into healthcare, agriculture, finance, law, manufacturing, transportation, and education. Model capabilities have improved dramatically with scale — a trend that has continued without obvious plateau. And AI has already spawned entire new industries: cloud AI services, autonomous vehicle startups, AI-native drug discovery firms, and large-scale data annotation labor markets. But GPTs have a paradoxical feature: they often produce a long lag between invention and measured productivity gains. Electricity was commercially available by the 1880s, yet the productivity boom it enabled did not show up clearly in economic data until the 1920s. Firms and workers needed time to reorganize processes, buildings, and job roles around the new technology. Many economists believe AI is in a similar early phase — the transformation is underway, but much of the aggregate productivity gain is still ahead.

The economic impact of AI can be organized into three levels. At the task level, AI automates or augments specific cognitive tasks — reading legal contracts, analyzing X-rays, writing first-draft code, routing customer service calls. At the firm level, firms that adopt AI effectively can produce more output with the same inputs, shift their product mix toward higher-value offerings, and undercut competitors on cost. At the economy level, if AI raises firm-level productivity broadly, it can raise GDP growth rates — in principle allowing an economy to produce more healthcare, education, and goods without proportionally more labor. These levels interact. A task-level automation at a law firm changes which types of lawyers add value. That changes wages and hiring. Multiplied across thousands of firms, it changes the composition of employment across the whole economy.

AI's Distinctive Economic Character

AI has economic properties that make it unusual even among general-purpose technologies. First, AI output is software — it can be reproduced at near-zero marginal cost. An AI model trained once can serve a million users with almost no additional cost per query, unlike a factory that must invest more capital for each additional unit produced. This creates extreme returns to scale. Second, AI improves with data. More users generate more data, which can improve the model, which attracts more users. This data flywheel creates strong winner-take-most dynamics, because a firm with ten times the data may produce a substantially better model, not just a slightly better one. Third, AI is a capital-intensive technology to build but a relatively cheap one to deploy. Training GPT-4 reportedly cost tens of millions of dollars. But inference — generating a single response — costs fractions of a cent. This cost structure means that early AI leaders can amortize enormous fixed training costs across billions of interactions, creating a moat that is very hard for a later entrant to cross. These characteristics — zero marginal reproduction cost, data-driven improvement, and extreme fixed-cost amortization — mean that AI's economic gains may concentrate powerfully among a small number of leading firms and countries, rather than spreading evenly.