The investment boom in artificial intelligence technologies has raised concerns about a potential bubble. This column argues that the 1920s boom and the Great Depression offer useful lessons. During the 1920s, process innovation grew rapidly, while product innovation stagnated. In turn, process innovation fuelled an investment boom based on an expectation of future demand. When this expectation failed to materialise, a sharp contraction followed. Thus, the optimism and prosperity of the Roaring Twenties help explain the onset of the Great Depression, and offer a cautionary tale for today’s artificial intelligence boom.

The current surge in artificial intelligence investment has brought an old economic question back into focus: what happens when firms pour capital into new capacity in anticipation of profits that have yet to materialise? Will today’s AI boom prove to be a bubble? Will demand grow fast enough to absorb the new capacity being built?

These questions make the 1920s boom, and the catastrophic depression that followed, well worth examining anew. Not because another Great Depression is inevitable, but because the story of how optimistic expectations drove investment during the Roaring Twenties, and whether that very boom left the economy dangerously exposed by 1929, offers a valuable lens through which to understand the risks that accompany rapid technological progress and capital accumulation.

What the existing literature leaves unexplained

Economists have spilled considerable ink on the Great Depression. Leading explanations have emphasised widespread bank failures, misguided monetary policy, the 1929 stock market crash, and other factors (e.g. Bernanke 1983, Friedman and Schwartz 1963, Romer 1990). Yet most of these accounts share a notable gap: they do not explain how the Great Depression started. As Bernanke (1983: 258) himself acknowledges, “It should be stated at the outset that my theory does not offer a complete explanation of the Great Depression (for example, nothing is said about 1929-30).” Instead, they focus on the Depression’s extraordinary severity and persistence, leaving the triggering mechanism unstated. Here, a different, but complementary, question is asked: Did the economic forces behind the prosperity of the Roaring Twenties sow the seeds of the 1929 collapse? Building on ideas in earlier work (Szostak 1995, Yorukoglu 2000), in Cole et al. (2026) we explore this possibility.

A booming economy with saturating markets

The 1920s stand out as one of the most prosperous periods in American history. Between 1921 and 1929, real GDP per capita grew by about 3.4% annually, well above its long-run trend, while real manufacturing investment per person rose by roughly 5.2% per year (see Figure 1). Yet the employment rate remained largely unchanged, and labour’s share of national income fell. In other words, production and investment surged, but employment failed to keep pace — a divergence that standard economic theories struggle to explain. What might account for this pattern? A closer look at the contrasting trends in product innovation and process innovation offers some clues.

Figure 1 Output, investment, employment, and the labour share, 1900–1950

Note: The data sources for all diagrams are detailed in Cole et al. (2026). The left panel shows chain-indexed real investment for manufacturing and real GDP per capita, while employment per capita and the labour share of income are displayed in the right one.

The Second Industrial Revolution transformed American life through the spread of automobiles, electric power, household appliances, and petrochemicals. By the early to mid-1920s, however, many of these groundbreaking innovations had already achieved widespread adoption. The flow of new products entering the market was insufficient to offset the slowdown in demand for these earlier innovations. Indeed, applications for patents and trademarks — imperfect but useful indicators of new product development — both peaked in the mid-1920s (see Figure 2). As a result, some of the economy’s most important new markets were becoming increasingly saturated (Figure 3). Annual growth in automobile registrations peaked around 1925. Housing starts and new construction also reached their highs around the middle of the decade, while house prices began to fall in 1925. Electrification followed a similar trajectory: by 1929, electricity powered about 78% of mechanical horsepower in factories, and roughly 68% of homes had been electrified. The pace of adoption around the end of the 1920s had slowed. Taken together, these trends suggest that the extraordinary demand unleashed by the innovations of the Second Industrial Revolution was becoming increasingly difficult to sustain at its earlier pace, while too few new product markets were emerging to take the place of those approaching saturation.

Figure 2 Indicators of product innovation: Design patent applications and trademarks

Note: Trademarks, trademark renewals, and design patent applications, 1900-1940.

Figure 3 Diffusion and market maturity: Automobiles, housing, and electrification

Note: Upper left, Automobile registrations and factory sales, 1900-1950. Upper right, Number of housing units started per million population, 1900-1950, and the value of new construction in millions of $1957-59, 1920-1947. Bottom panel. The percentage of horsepower for mechanical drive in factories supplied by electricity, 1889-1954; the percentage of dwellings that were electrified, 1907-1950; and the value of new construction by private utilities for electric and gas in millions of dollars, 1907-1950.

Process innovation kept racing ahead

While demand for new products slowed, production methods continued to improve rapidly (see Figure 4). The spread of assembly lines and the widespread electrification of factories made manufactured goods much cheaper to produce. For example, Fabricant (1942) reports that the labour required to produce a car fell by a factor of ten between 1910 and 1930. These remarkable productivity gains extended across manufacturing industries and, importantly, continued even after the Great Depression began. The result was a striking imbalance: firms could produce more of the goods consumers already knew and bought using fewer workers, while opportunities to stimulate demand through entirely new products were not expanding at the same pace.

Figure 4 Process innovation in automobiles and manufacturing, and quality-adjusted automobile prices

Note: Process innovation in autos and all manufacturing (left axis) and real quality-adjusted auto prices (right axis).

Optimism sustained investment

The imbalance between product and process innovation lies at the heart of our explanation. Consumers generally value expanding their consumption baskets with new goods over simply purchasing more of the goods they already buy or own — a preference economists call ‘love of variety’. When a wide range of new products becomes available, rising incomes can support demand for a broader selection of goods. At the same time, greater product variety intensifies competition among producers, helping keep prices in check. Under these conditions, productivity growth raises incomes and can sustain both higher consumption and employment.

The picture changes when process innovation advances faster than product innovation. Consumers have fewer new options through which to spend their rising incomes. Incumbent producers face less competition and enjoy greater market power. As a result, lower production costs may not translate fully into lower prices. Under the hypothesis here, the combination of limited product innovation and rapid process improvements constrains consumption growth even as GDP per capita rises. Sluggish consumption growth, combined with rapid gains in labour productivity, then means that employment can remain flat or even decline, while investment and GDP grow rapidly. This dynamic can help explain the otherwise puzzling features of the 1920s: rising output, substantial investment, stagnant employment, and a declining labour share.

There remains an unanswered question: if consumption growth was weak, why did firms keep investing so heavily? The likely answer is optimism. Perhaps investors believed the slowdown in product innovation was only temporary. They expected that a wave of new products would occur soon. Building factories and acquiring equipment in advance would then appear sensible.

Indeed, financial markets reflected the optimism that prevailed throughout the 1920s (see Figure 5). Stock prices surged, price-to-earnings ratios climbed sharply, and both initial public offerings and merger activity increased. Together, these indicators suggest that investors remained confident in future economic growth, even as important markets, from automobiles to housing, showed signs of saturation. Under our interpretation, rapid process innovation fuelled capital accumulation driven by the expectation that a new wave of product innovation would emerge to sustain demand. When that expectation began to waver, the consequences were severe.

Figure 5 Stock market indices and the price–earnings ratio, 1919–1939

Note: The September Dow and S&P, 1919-1939. Also plotted is the cyclically adjusted June P/E ratio.

When the expected future never arrived

Think about the economy entering 1929 carrying factories and machinery built partly in anticipation of markets that had not yet appeared. If households and firms abruptly revise down their expectations for new product innovation, much of that productive capacity becomes worth far less than anticipated — and investment collapses. At the same time, process innovation continues apace. Firms can still meet whatever demand exists with fewer workers. And with few new products coming to market, consumer spending remains constrained, reinforcing the downturn in both output and employment.

This narrative lines up with the history of the late 1920s and early 1930s. By 1929, automobile and housing markets had shown signs of slowing for several years, although the stock market remained buoyant. After the crash, manufacturing investment contracted much more sharply than output or employment. By 1933, real output had fallen to about 71% of its 1929 level, employment to 82%, and manufacturing investment to just 37%. Yet technological advance did not simply cease. Indeed, Crafts et al. (2016), writing in VoxEU, document strong US productivity growth during the 1930s. Productivity gains can coexist with a collapsing economy when demand and investment fail to keep pace.

The boom and the bust

Our hypothesis offers a plausible explanation for the onset of the Great Depression, complementing more familiar accounts. The stock market crash heightened uncertainty, bank failures disrupted credit relationships, and monetary contraction and deflation deepened the decline in spending (Romer 1990, Bernanke 1983, Friedman and Schwartz 1963). These forces likely magnified a downturn that was already rooted in disappointed expectations about future demand — and they help explain why the Depression proved so difficult to escape.

The broader lesson is not that technological progress is dangerous, or that investment booms inevitably end in disaster. Rather, productive capacity and demand do not necessarily grow in step. In the 1920s, cost-saving innovations brought genuine prosperity. But expectations of continued growth in consumer demand may have encouraged investment beyond what markets could ultimately sustain. The Roaring Twenties may have been both a blessing and a curse: their technological achievements raised living standards for millions of Americans while potentially sowing the seeds of the economic collapse that followed.

Source : VOXeu