The hidden Power of AI: What patents reveal about innovation and Productivity

Artificial Intelligence (AI) is widely regarded as the most transformative technology of our time, reshaping how firms operate and how individuals work. From self-driving cars to intelligent personal assistants and advanced medical diagnostics, AI continues to expand the frontier of machine capabilities—moving beyond simple manual tasks to complex cognitive activities such as understanding language, recognizing patterns, and making decisions.

Yet, the full economic impact of AI remains largely unknown, as AI is inherently difficult to measure. The traditional distinction between firms that develop and sell new technologies and those that purchase and implement them at the business level proves inadequate in the case of AI. Unlike conventional technologies, AI is an intangible asset that often emerges through co-invention between producers and users, defying the standard boundaries of national accounting principles.

Recent research in the economics of AI has made progress by examining job advertisements to identify firms hiring AI specialists, such as computer scientists, as an indirect measure of adoption. Other valuable, though less explored, sources of information include firms’ purchases of data and specialized software. More traditional indicators, such as survey data and patent statistics, offer the advantage of broad comparability across countries and over time. However, patents have notable limitations: AI-related innovations do not always meet the criteria for patent protection, and firms may prefer trade secrecy to safeguard their algorithms and models.

Early studies have tracked its implementation and development among early movers, namely those firms that first adopted the new technology. Big techs, primarily American and Chinese companies — like Google, Amazon, and Alibaba — dominate the field. Focused studies in the US show that this mainly occurs when these companies acquire the most successful start-ups or hire talented computer scientists from academia.

Despite the aforementioned caveats, patents remain a valuable tool for identifying which firms develop AI technologies and assessing their effects at the firm level. In two co-authored papers, I address these issues using company-level data from several European countries. In analyzing the determinants of AI patenting, we emphasize key inputs such as firms’ internal and external knowledge, research teams, and other organizational characteristics.

AI builds upon decades of progress in computing power, data storage, networking, and algorithms. Firms already active in these foundational domains possess the resources, expertise, and mindset necessary to explore AI.

In the development of AI, the greatest advantages initially accrued to companies that were already active in digital domains such as Network & Communication and High-speed Computing & Data Analysis. More recently, the lead has shifted toward firms with prior expertise in Cognition, Imaging & Sound. This trend mirrors the contemporary AI boom, which has been largely propelled by breakthroughs in “perception” AI—particularly in areas like computer vision and natural language processing.

Notably, a company’s existing stock of AI patents exerts a strong positive influence on its current AI patent productivity, generating a virtuous cycle of innovation. This dynamic gives first movers a sustained competitive advantage over later entrants, reinforcing their market dominance. These explain why most big tech companies, such as software houses, reinforce their leadership and market power worldwide. This is a key characteristic of AI, more than ICT.

This phenomenon is illustrated in Figure 1, which compares the inventive productivity gains associated with prior patenting activity in AI and ICT. The left-hand panel shows that for the most prolific AI innovators, the positive effect of past patenting success on current innovation output is roughly three times greater than for firms active in ICT fields.

Figure 1: Patenting gains in relation to the size of earlier innovation

In terms of effects, the development of AI appears to generate substantial productivity gains, ranging from 2% to 17% depending on the type of firm. Even those operating far from the technological frontier can experience significant benefits from adopting and implementing the new technology. Figure 2 shows that these productivity improvements are not immediate but tend to materialize a few years after AI development, as measured by patent activity.

Figure 2: Timing of the effects of AI on firm productivity (number of years)

This column is based on the following articles:

• Igna, I.A., Venturini, F. (2023) “The determinants of AI innovation across European firms”. Research Policy, 52 2, 104661

• Marioni, D.S. L., Rincon-Aznar, A., Venturini, F. (2024), “Productivity performance, distance to frontier and AI innovation: Firm-level evidence from Europe”, Journal of Economic Behaviour & Organisation, 228, 106762