What Technology Does to the Numbers, and When
Steam, electricity, computers and now AI have all arrived the same way: a surge of investment, a long wait, and then a productivity line that moves. The wait is the part nobody prices in.
October 28, 2026 · updated November 9, 2026 Revised Nov 9
The pattern
Every big technology has reached the economic statistics in the same order. First investment, which shows up in GDP immediately because building things counts. Then, after a gap of a decade or more, productivity: output per hour worked, the number that decides whether the country is getting richer. Then, later still and only partly, wages. The gap between the first and the second is where bubbles form, because investors can see the technology and can't yet see the return, and the gap between the second and the third is where the arguments about who benefits come from.
The almanac now carries the productivity line as a companion on the GDP chart, back to 1947. This article walks the four waves and ends with where the fifth stands by the same clock.
New articles and charts arrive in one short email on Monday mornings. Get the weekly digest
Steam, and a fifty-year wait
The Industrial Revolution is the clearest case because the record is long. British output per worker rose steadily from the 1780s. Real wages for ordinary workers barely moved until the 1840s, a stretch economists call "Engels' pause" after the man who wrote it up from Manchester. For half a century the gains went to capital, factories and cities; then wages caught up and kept rising. Nobody alive in 1810 saw the second half of that sentence.
The dynamo, and a forty-year wait
Edison's first power station opened in 1882. American factory productivity did not visibly respond until the 1920s. The economic historian Paul David worked out why: a factory built around a steam engine had one shaft running the length of the building, with every machine belted to it, and swapping the engine for a motor changed nothing. The gains came only when factories were rebuilt around small motors at each machine, which meant new buildings, new layouts, and a generation of managers who had grown up with electricity. The technology was ready in 1882. The economy was ready in 1920.
The computer, the paradox, and the boom that came anyway
In 1987 Robert Solow observed that computers were visible everywhere except in the productivity statistics. He was right: American productivity growth had run about 1.5% a year since 1973, through the entire arrival of the mainframe and the personal computer. Then, from 1995 to 2004, it ran close to 3%, the best stretch since the 1960s. The gap was the dynamo's gap again: the gains came when businesses reorganized around the machines, not when they bought them.
The dot-com bubble sits in the middle of that stretch, and the sequence matters. Investment in computers, software and telecom lines surged through the late 1990s; the Nasdaq rose fivefold and then fell by three-quarters between March 2000 and October 2002; the telecom companies that had laid tens of thousands of miles of fiber went bankrupt with most of it unlit. The recession of 2001 was short and shallow, unemployment rose from 3.8% to 6.3%, and the 10-year yield fell from 6.7% to 3.3%. And productivity growth did not slow. The investors lost their money; the economy kept the fiber, the software and the reorganized firms, and the productivity boom ran three more years. That is the most important fact about the dot-com bust for reading the present: a bubble in the technology's financing is not the same thing as a bubble in the technology.
Where AI stands on that clock
By the same measures, the AI wave is in its investment phase. Spending on AI-specific data centers, chips and networking reached roughly 0.8% of GDP in early 2026, with all computing infrastructure near 1.5%, about double its share in the 2010s, and it has been the largest single contributor to growth in business investment for two years. That is the dot-com shape: the building is visible in GDP now, as the fiber was.
The productivity line is where the question sits. Output per hour rose 2.1% over the year to the third quarter of 2026, better than the 2010s average but inside the range of ordinary cyclical swings, and the Bureau of Labor Statistics does not and cannot attribute any of it to a technology. The pattern of the earlier waves says that is normal: a decade or more between the investment and the reorganization, with the gains arriving when work is rebuilt around the tool, not when the tool is bought. It also says the gains, when they come, do not announce which technology produced them.
Two things are different this time, and both are argued below. The investment is concentrated in a handful of companies rather than spread across an industry, which makes the financing bubble, if it is one, more concentrated too. And the technology's claimed effect is on the kind of work, office and professional, that the earlier waves mostly left alone.
What people argue about
Whether this is 1999. One side: the capital spending is running ahead of any revenue, as it did then, and when the spending slows, so does the growth it has been carrying, perhaps a third of recent GDP growth by some estimates. The other: the 1999 lesson is that the bust hit investors and left the economy with the infrastructure, and that productivity kept rising through the wreckage. Both can be true. The question the data can't answer yet is how much of the current investment will be lit.
Whether it takes the jobs. Every earlier wave moved work between occupations: telephone operators, typists, bank tellers (whose numbers rose for thirty years after the ATM and fell only after 2010), travel agents. Whether this one also raises unemployment for the economy as a whole is the argument. One side says this wave is different because it reaches the professions. The other says every wave was said to be different, and that none so far has kept the unemployment rate up: the investment boom of the late 1990s coincided with unemployment falling under 4% in 2000, the lowest since 1969. The almanac's job numbers will show the aggregate. They won't show the occupations, which is where this argument lives.
Who gets the gains. Engels' pause is the historical warning: half a century of productivity before wages followed. The 1995–2004 boom was faster, and real wages did rise, but the share of national income going to labor fell after 2000 and has not recovered. One side reads that as the pattern repeating and argues for policy to redistribute the gains; the other says the 2021–23 compression showed a tight labor market does the job without it. Real vs. nominal is the page for reading whether wages are keeping up when the productivity line moves.
See also
- GDP: what investment counts for, and the productivity companion.
- You work for a wage and the unemployment rate: the aggregate job numbers the argument is usually about.
- The 10-year Treasury yield: the rate the investment is financed at, and what it did in 2000–03.
Sources
- BLS, Productivity and Costs, the quarterly release, and its third-quarter 2026 preliminary release (November 5, 2026); FRED: Nonfarm business labor productivity (OPHNFB), the companion on the chart.
- Epoch AI, "Data center buildout share of US GDP", the investment estimates for early 2026.
- Paul David, "The Dynamo and the Computer" (American Economic Review, 1990), the forty-year lag.
- Robert Allen, "Engels' pause" (Explorations in Economic History, 2009), the fifty-year lag.
- Robert Solow, review in the New York Times Book Review, July 12, 1987, the productivity paradox.
- FRED: Nasdaq Composite (NASDAQCOM), unemployment rate (UNRATE), 10-year yield (GS10), the dot-com figures.
Corrections
October 29, 2026: An earlier version stated as fact that no technology has ever raised the unemployment rate for the economy as a whole. That is one side's reading of the record, and the paragraph now presents it as such.