SHANGHAI, October 25, 2026 — The promise of enterprise artificial intelligence is colliding with a familiar economic wall. Despite massive investment in new models and tools, the tangible payoff for most companies remains elusive, echoing a historical pattern that has confused economists for decades.
The shift is not just about better algorithms. It is about the painful gap between buying a technology and actually making it work within the rigid structures of a modern corporation. What is changing now is the scale of that gap. We are watching a repeat of the 1970s and 1980s, a period when computing power exploded but productivity statistics barely moved. For the AI industry, this is a critical moment. The question is no longer whether the technology works, but when the economic returns will finally show up in the data.
The Ghost of the Solow Paradox
To understand why the AI payoff is missing, you have to look back. The term “productivity paradox” was coined by Erik Brynjolfsson in a 1993 paper. He was inspired by a famous quip from Nobel Laureate Robert Solow. Solow observed that you could see the computer age everywhere, but not in the productivity statistics. This disconnect became known as the Solow paradox.
In the United States during the 1970s and 1980s, the situation was stark. Computing capacity increased a hundredfold. Yet, labor productivity growth slowed significantly. In the 1960s, growth was over 3 percent. By the 1980s, it had dropped to roughly 1 percent. This was a massive discrepancy between investment in information technology and actual output at the national level.
Many analysts at the time, including Steven Roach and later Paul Strassmann, popularized this idea in the media. They pointed out that companies were spending heavily on IT, but the numbers did not reflect a corresponding jump in efficiency. Some observers argued that there was no real paradox at all. They suggested that the disconnect was simply a result of unwarranted assumptions about how technology impacts productivity. In this view, the issue was not a mystery, but a failure to deploy the technology correctly.
Why New Tech Takes Time to Pay Off
The reason the returns are delayed is not a bug. It is a feature of how disruptive technologies work. History offers clear parallels. The steam engine and electricity did not immediately boost economic output. Their dividends were reaped slowly, over the course of decades.
There are two main reasons for this lag. First, diffusion takes time. A technology must move from early adopters to common use across an entire industry. Second, organizations must restructure. Companies have to reorganize their workflows to master the efficient use of the new tool. During the initial phase of IT adoption, a large number of cutting-edge investments were actually counterproductive. They were over-optimistic and failed to deliver the expected results.
Huang Yiping, a Boya Distinguished Professor at Peking University, highlights this dynamic in his recent work. He notes that computers eventually did boost total factor productivity. However, these gains were masked by measurement difficulties, time lags, and structural adjustments. The data did not show the improvement immediately because the economy was still adjusting to the new reality.
This is exactly what is happening with AI today. The technology is powerful, but the organizational friction is high. Companies are buying tools faster than they are redesigning their businesses to use them. The initial investments are often messy and expensive. This creates a period where costs rise, but benefits remain hidden.
Expert Warnings on Economic Imbalance
The implications of this lag extend beyond individual company balance sheets. They touch on broader economic stability. Huang Yiping raised concerns at the Inclusion Conference on the Bund in Shanghai on September 9, 2026. He warned that artificial intelligence could exacerbate the imbalance between strong supply and weak demand in China’s economy.
This is a subtle but important point. AI can boost productivity, meaning it can produce more goods and services with fewer resources. However, if this does not translate into higher wages for workers, demand may not keep up with supply. Huang cautioned that technological advances do not necessarily translate into immediate productivity gains or higher wages.
He drew on the “Solow paradox” and “Engels’ pause” to make his case. He suggested that persistently weak demand could require public policies that go beyond conventional macroeconomic frameworks. His call for “employment-first AI innovation” reflects a fear that the benefits of this technology might be concentrated in the hands of a few, while the broader economy struggles to find new sources of growth.
This is not just a theoretical concern. If AI drives down costs without driving up incomes, the gap between what an economy can produce and what its people can afford to buy will widen. This creates a structural problem that is difficult to fix with standard monetary policy.
What the Timeline Suggests for the Future
So, what does the historical timeline tell us about the future of enterprise AI? The evidence suggests that the current period of disappointment is temporary, but it will be long. The productivity paradox of the 1970s and 1980s did not last forever. Productivity growth renewed in the developed countries in the 1990s. The paradox disappeared, but the lessons from the research efforts that tried to explain it remain important.
Today, productivity growth has slowed again globally from the 2000s to the present day. The term “productivity paradox” now refers to this more general disconnect between powerful computer technologies and weak productivity growth. For AI, the pattern is likely to repeat. We are in the early stages of diffusion. The most significant gains will come not from the initial purchase of AI tools, but from the deep restructuring of organizations that follows.
Huang Yiping remains relatively optimistic about AI’s potential to improve productivity. However, he emphasizes that the process is complex and that it is too early to draw firm conclusions. We will have to keep tracking developments and analyzing the evidence. The transformation is ongoing, and its eventual outcome remains uncertain.
For the enterprise sector, this means patience is required. The companies that will succeed are those that understand that the value of AI is not in the model itself, but in the organizational changes that allow the model to be used effectively. The shift is from buying technology to redesigning work. Watch this space, because the data will eventually catch up with the reality. The trend underneath is clear: the payoff is coming, but it will look different from what we expected, and it will take longer to arrive.
Sources
- Productivity paradox
- Huang Yiping: AI Could Worsen China's Supply-Demand Imbalance
- Your AI Marketing Transformation Will Get Harder Before It Gets Easier
Photo: Arlington Research / Unsplash (Unsplash License)
