The "lag" from innovation to productivity surge: Important lessons from the information and communication technology revolution for the AI era

The productivity boost brought by AI may be real, but the speed at which it enters macro data may not align with the market's most optimistic timeline. Experience from the information and communication technology (ICT) revolution shows that new technologies often have significant time lags from commercialization, investment expansion, to productivity explosion.

According to "Following the Trend Trading Desk," Elsie Peng from Goldman Sachs U.S. Economic Research team wrote in a study released on July 1, "We expect AI to significantly boost productivity growth over the next decade." This statement means that the core variable of AI is not only whether it can improve efficiency, but also when such improvement will be confirmed by macro statistics.

The key question is shifting from "Is AI useful?" to "When will AI be transformed into observable productivity growth?" There is already considerable evidence of efficiency improvements at the enterprise and experimental levels, but macro data typically only start to reflect these changes after companies complete process restructuring, employee training, and organizational adjustments.

ICT provides an important reference point. Personal computers were commercialized around 1981, ICT investment started rising in the early 1980s, but actual acceleration in U.S. productivity didn't occur until the late 1990s, lagging by about 15 years. AI may be faster, but it won't automatically skip the stage of organizational adaptation.

Core Lesson from ICT: Investment Comes First, Productivity Lags

In the early 1980s, the commercialization of personal computers sparked a wave of ICT innovation. The number of breakthrough ICT patents per capita continued to rise in the late 1980s and early 1990s, and ICT investment increased significantly starting in the 1980s. Industries such as professional services, wholesale trade, transportation, and finance increased their investments early on.

However, productivity did not rise in tandem.

Industry panel estimates show that for every 1 percentage point increase in ICT investment as a share of capital stock, its contribution to productivity growth is actually slightly negative in the first four years. The positive effect only becomes clear around year eight, peaking near year twelve at about 0.6 percentage points.

This is the "J-curve" in new technology diffusion: Companies first invest in equipment, adjust processes, bear adaptation costs, and only then may efficiency be released. Especially when the technology changes information flows, decision-making, and employee collaboration, what shows up first in statistical data is often costs, not output.

Three Drag Factors: Costs, Network Effects, and Intangible Capital

Delayed productivity from ICT is firstly related to costs. Semiconductor and communications equipment prices remained high in the 1980s, only starting to decline when increased competition and regulatory changes opened up the market in the 1990s. The U.S. Telecommunications Act of 1996 fostered market competition for communications equipment, and the increase in chip competitors also drove down key component costs.

The second factor is network effects. Technologies like the internet and mobile communications offer limited value to individual users until user numbers reach a critical mass. Only after the adoption rate of ICT applications like the internet and mobile phones crossed a tipping point in the late 1990s did productivity improvements become easier to realize.

More crucial is intangible capital. Companies need to redo processes, retrain employees, adjust organizational structures, build software and data systems. Estimates show that for every $1 invested in ICT hardware, at least an additional $1.7 in supporting intangible investments is needed. About two-thirds goes toward software and databases, the rest toward workforce and organizational restructuring.

This is especially important for AI. GPU, data centers, and model capabilities are easy for the market to observe, but organizational renovation is not. Historical experience shows the latter often determines when productivity truly appears on the books.

AI May Be Faster, But Not Without Friction

Compared to ICT, a notable difference with AI is the cost curve. The decline in AI model usage costs is clearly faster than the drop in PC prices back then. Although some leading U.S. models have recently increased nominal prices to expand profit margins, competitive pressure and underlying computing cost declines may limit the room for further price increases, making overall token price paths more likely to remain stable.

AI also relies less on network effects than communication technologies. Internal deployment of AI tools by companies does not require simultaneous adoption across the entire industry or society to achieve partial efficiency gains. This gives AI a chance to be reflected in productivity data earlier than ICT.

But slow-moving variables still exist. Companies must redo workflows, train employees, adjust job divisions and organizational structures. AI-related hardware investment as a proportion of total capital stock is rising faster than the ICT build-out cycle back then; but investment in workflow restructuring, measured by the share of employee compensation involved in such reorganization, currently looks slower.

Meanwhile, official statistics may underestimate the organizational transformation underway in companies. A survey by the Atlanta Fed hints that AI-related intangible capital expenditures in 2026 could be about $280 billion. Company data-based estimates show AI-related labor costs in the U.S. could reach $150 billion annually, with organizational capital investment tied to executive time allocation around $40 billion per year.

Early Signals May Arise from Four Types of Industries

Searching for early macro signals of AI productivity shouldn't focus solely on the whole economy. A more feasible approach is to first observe industries with a higher digital foundation, greater AI exposure, and tasks more easily automated or augmented.

The report's framework includes four types of indicators: existing IT exposure, AI adoption rate, AI exposure, and the intensity of work reorganization since 2022. Based on this framework, the leading industries include information and data processing, professional services, film and sound-related sectors, insurance, credit intermediation, and computer and electronics manufacturing.

Among these, information and data processing scores 1.97 on average, professional services 1.48, film and sound-related sectors 1.21, insurance 1.09, credit intermediation 0.98, and computer and electronics manufacturing 0.97.

If combined into broader industries, information, professional services, insurance, and finance are most worth tracking first. These sectors not only have a more robust digital foundation but also have more tasks that can be automated or augmented by AI.

But caution is needed when judging. These potential beneficiary industries have had stronger productivity performance even in past decades. If they continue to outperform in the future, it cannot be simply attributed to AI. More crucial is whether they show new acceleration relative to their own historical trends.

The Market Should Focus on Three Main Lines

The first is cost. As long as model usage prices and computing costs continue to decline, AI diffusion could face less resistance than during the era of ICT.

The second is organizational capital. Whether companies truly redo processes, jobs, and data systems matters more than just buying AI tools. The ICT revolution has already proven that simply purchasing equipment is not enough to trigger a productivity explosion.

The third is sector productivity inflection points. If information, professional services, insurance, and finance see productivity improvements beyond historic trends, it is more likely to serve as early evidence of AI entering macro data.

However, sector-level productivity data itself is published with delays. Even if AI has already yielded efficiency gains at the micro level, macro statistical confirmation may take several more years. The biggest reminder ICT offers for the AI era is: technological breakthroughs can be fast, but productivity realization is usually slower.

~~~~~~~~~~~~~~~~~~~~~~~~

The above brilliant content is from Following the Trend Trading Desk.

For more detailed analysis, including real-time interpretation and frontline research, please join【Following the Trend Trading Desk·Annual Membership】

Risk Disclosure and DisclaimerThe market is risky, investment requires caution. This article does not constitute personal investment advice and does not take into account the special investment objectives, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article fit their specific circumstances. If you invest based on this, you bear responsibility for the outcome.