By Yan Deli
Senior Expert, Tencent Research Institute
Originally published in "China Informatization," with revisions.
Three Explanations for the Productivity Paradox
Labor productivity is the ratio of total output to total labor hours, measuring the efficiency with which labor input is converted into actual output. It is a crucial economic indicator. In the long run, productivity growth is the only path to raising living standards (Tim Sargent, 2024). Technological progress is the primary source of productivity growth (Brookings Institution, 2024). Yet, we often witness the simultaneous existence of "leaps in technological innovation" and "disappointing productivity growth." This is the productivity paradox.
In 1987, Nobel laureate Robert Solow wrote in an article on deindustrialization: "You can see the computer age everywhere but in the productivity statistics." This offhand remark became the most classic expression of the "productivity paradox" or "Solow paradox." While Solow is generally credited with proposing the paradox, it may have been someone else; Peng Geng and Lu Benfu (2003) suggest it was Stephen S. Roach.
Over the past four decades, scholars have continuously studied the productivity paradox. Paul David and Erik Brynjolfsson in the United States were pioneers and leaders in this field, forming three main explanations: mistaken expectations, measurement errors, and time lags.
First, mistaken expectations. People's optimistic expectations about the potential of technology are wrong; it is not as transformative as imagined. History has seen many exciting technologies fail to meet initial expectations. For instance, nuclear energy is far from being "too cheap to meter"; controlled nuclear fusion is "always 30 years away"; and Marvin Minsky's (1970) prediction that "in three to eight years we will have a machine with the average intelligence of a human being" has yet to materialize.
Second, measurement errors. In practice, measuring productivity is not easy (Tim Sargent, 2024). The productivity gains brought by new technologies are real but have not been accurately captured; the tools people use to measure empirical reality are not functioning effectively. For example, free internet services and tech products with improving performance and declining prices are difficult to fully reflect in traditional statistics.
Third, time lags. The first two explanations attempt to resolve the contradiction between optimistic expectations and disappointing statistical realities by assuming one side is somewhat wrong. The time-lag explanation, however, posits that both seemingly contradictory aspects can be true simultaneously; it is just not yet time. In other words, new technologies take a long time to have a substantial impact on productivity, especially General Purpose Technologies (GPTs).
The Impact of New Technologies on Productivity Has a Lag
Brynjolfsson (2017) argues that the time-lag explanation is the most convincing and the primary cause of the productivity paradox. Paul David pointed out (2000): "One should not expect the largest productivity returns in the early stages of a paradigm shift, even though the diffusion speed of the new technology may be fastest at this time." General Purpose Technologies require multiple rounds of secondary innovation, complementary innovation, and organizational change before they can substantially impact productivity. Brynjolfsson (2020) summarized the lagged effect of GPTs on productivity as a "J-curve." Helpman and Trajtenberg (1994) held similar views, dividing the impact of GPTs on economic growth into "sowing" and "harvesting" stages. In the sowing stage, productivity growth is slow or even declines; only in the harvesting stage does growth truly begin.
Therefore, in the long run, the productivity paradox is not really a paradox; it is merely a phenomenon of a specific stage, such as the 1980s when Solow made his observation. It was not until the mid-to-late 1990s that capital accumulation in information technology reached a level sufficient to affect productivity (Stephen Oliner & Daniel Sichel, 2000). According to research by the European Central Bank (2020), the contributions of electricity and ICT to U.S. labor productivity show great similarity in their historical trends: both were flat in the early period and accelerated later, with turning points in 1915 and 1995, respectively. Historically, the steam engine, the generator, and the computer began to significantly boost productivity only after 118, 91, and 49 years from their invention, and 54, 40, and 21 years from their commercialization, respectively. As shown in the figure below.
Figure: Years from Invention and Commercialization of General Purpose Technologies to Significant Productivity Improvement
Data Source: Compiled from data by Nicholas Crafts (2018), Paul David (1990), Stephen Oliner & Daniel Sichel (2000), and the European Central Bank (2020).
AI Cannot Yet Significantly Boost Productivity
Artificial Intelligence is a new General Purpose Technology (Nicholas Crafts, 2021; OECD, 2024; NBER, 2026), characterized by universal applicability, continuous improvement, and the ability to spur innovation, making it an engine for future economic growth. Although the term "Artificial Intelligence" was coined 70 years ago and the machine learning revolution began 14 years ago, the current AI wave is in full swing, ranging from Large Language Models and multimodal models to world models, agents, and physical AI, emerging one after another. However, productivity growth has not noticeably accelerated; we are even facing a "productivity crisis" (Rogers, 2024).
Since the release of ChatGPT in November 2022, labor productivity in Canada and the European Union has remained basically unchanged, fluctuating around 0%. In contrast, the EU's average hourly labor productivity growth reached 1.5% during 1999-2008 and 1% during 2010-2019. As shown in the figure below. The U.S. has shown strong growth, with nonfarm business sector labor productivity growth reaching 2.2% in 2025, standing out among Western nations, but this is merely equivalent to the long-term average since 1947 (Source: Bureau of Labor Statistics).
Figure: Recent Labor Productivity Growth in the EU (Source: Eurostat)
Media outlets like to use terms like "iPhone moment" or "ChatGPT moment" to describe significant technological changes, giving an impression of instantaneity and suddenness. However, the impact of technology on the economy and society is a long-term process. Nobel laureate Daron Acemoglu pointed out (2024): "Many people think that AI can rapidly and thoroughly transform all aspects of the economy and lead to a significant increase in productivity, even bringing us close to the singularity. While this possibility cannot be completely ruled out, there is no evidence so far that such a revolutionary impact has occurred."
The "AI+" Index Needs to Reach 50%
Penetration rate is a leading indicator that determines the degree of contribution to productivity. The extent of AI's impact on labor productivity can be intuitively reflected through its penetration rate. Paul David pointed out (2000): "Only after the penetration rate of cost-saving technologies reaches the threshold of 50% will it have the greatest impact on the growth rate of total factor productivity." In other words, the "AI+" index needs to reach 50% before productivity growth can significantly accelerate.
However, official data indicates that corporate application of AI is still in its early stages, facing enterprise-level barriers. The corporate AI penetration rate is around 10% in the U.S. and Canada, around 20% in the EU and the UK, and exceeds 30% for AI technology application in manufacturing enterprises above designated size in China. Here, penetration rate measures "whether or not" a technology is used; applying any single AI technology counts. It has not yet reached the advanced stage of "how good" the application is, without considering the depth, breadth, or effectiveness of the application. Due to different statistical calibers, these numbers cannot be directly compared horizontally, but they all indicate that AI is still in an early stage of technological diffusion, with penetration rates far from the 50% threshold. As shown in the figure below.
Figure: Corporate AI Penetration Rates in Major Countries and Regions
Note: For China, this refers to manufacturing enterprises above designated size; for the EU, Germany, and France, it is limited to enterprises with 10 or more employees; survey times for the UK and US were in September of the respective years, while for Canada it was the second quarter.
China is on par with the United States in the field of artificial intelligence (Zhong Caiwen, 2025), yet it has embarked on a different path. The U.S. focuses on heavy investment, performance, and the pursuit of AGI. China, on the other hand, emphasizes open weights and strong applications, vigorously promoting the deep integration of AI with various industries, using "AI+" as a lever to promote technological transformation and empower high-quality industry development. The essence of "AI+" is to increase adoption rates, thereby boosting labor productivity. As the "AI+" initiative advances in depth, labor productivity is bound to see a rapid rise.
References:
【1】Erik Brynjolfsson, Daniel Rock, and Chad Syverson, "Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics," NBER Working Paper 24001 (2017), https://doi.org/10.3386/w24001.
【2】Paul A. David, 2000. "Understanding Digital Technology's Evolution and the Path of Measured Productivity Growth: Present and Future in the Mirror of the Past," https://doi.org/10.7551/mitpress/6986.003.0005
【3】S. S. Roach, "America's Technology Dilemma: A Profile of the Information Economy," Morgan Stanley, 1987.
【4】Daron Acemoglu, 2024. "The Simple Macroeconomics of AI," NBER Working Papers 32487, National Bureau of Economic Research, Inc.
【5】Filippucci, F, P Gal and M Schief (2024), "Miracle or Myth? Assessing the macroeconomic productivity gains from Artificial Intelligence", OECD Artificial Intelligence Papers, No. 29.
Recommended Reading:
Yan Deli: "Does AI Lead to 100,000 Layoffs in Silicon Valley?"
Yan Deli: "The Nature of Technological Innovation"
👇 Click "Looking" to share insights