Measuring technological penetration in Spain: from installed capital to actual use

Saying that a technology has "penetrated" the economy can refer to different things: that companies have invested in it, that they have it installed, that they use it in specific processes, or that its use already produces measurable effects on productivity. These are related but not equivalent stages. Confusing them leads to converting investment or expectations into a claim about effective adoption.

The CaixaBank Research analysis published on February 13, 2020, addressed a specific difficulty: at that time, there were no economic data directly quantifying artificial intelligence (AI) as an emerging technology. Therefore, its authors approximated the relevance of new technologies through components of the capital stock that could facilitate their development. The study is useful for understanding the technological productive base, but it does not constitute a count of AI user companies or a measurement of their habitual use.

A business adoption statistic answers a different question: what proportion of companies declares using at least one tool, within a defined population and period. Even that data does not reveal by itself the intensity, frequency, or economic weight of that use. The indicator is only well interpreted when it is identified what it observes and whom it includes.

The 2020 study used the EU KLEMS base to observe the capital stock in four groups: software and databases, research and development (R&D), computers, and telecommunications equipment. The authors combined them as a broad approximation to the economic presence of technologies directly or indirectly related to AI. They did not isolate a specific investment in AI: the components cover assets with applications that go far beyond that technology.

To compare the weight of these assets across economies and sectors, the article divided the aggregated stock of the four types of capital by hours worked. The result is expressed as constant 2010 euros per hour worked. Thus, the indicator approximates the available technological capital in relation to work; it does not count users, licenses, AI projects, or automated tasks.

The measure allows contrasting the technological capital stock per hour worked among the countries and sectors included in the analysis. Its counterpart is a broad proxy, not a direct thermometer of adoption. Software, computer equipment, and telecommunications are part of the capital categories included in the indicator; by themselves, these data do not identify which companies use AI or for what purposes.

In the period studied, CaixaBank Research placed Spain, along with the United Kingdom, at lower levels than France and Germany in new technology stock per hour worked, although it noted that Spain had registered high growth during the previous two decades. The article suggested that, if that trend continued, Spain could approach France and Germany in about ten years. It was a conditional extrapolation, published in 2020, not a confirmation of subsequent convergence or an updated forecast.

The sectoral comparison of the work covered 31 branches of industry and services in Spain. The sectors with the highest technological stock tended to coincide with Eurostat's high-technology and knowledge-intensive categories. However, the breakdown excluded agriculture and mining; therefore, its sectoral results do not describe all productive sectors of the country. Moreover, a sectoral classification does not allow concluding that all companies within a branch have the same level of investment or use.

The practical reading is to distinguish installed technological capacity from business adoption. The aggregated capital stock of a sector does not establish how many companies in that sector use AI, with what intensity, or with what results. To answer those questions, different indicators are needed, defined for the population and period to be analyzed.

To know the declared use of AI in Spanish companies, it is advisable to resort to a survey that explicitly asks about that technology and review its definition and population. The Bank of Spain's Survey on Business Activity (EBAE) included, in the fourth quarter of 2024, questions on generative and predictive AI, among other technologies. In its results, 19.9% of the surveyed companies declared using at least one AI system; among users, experimental use or in the form of a pilot predominated. That percentage corresponds to the EBAE and its definition, not to the EU KLEMS capital stock.

The INE series offers another perspective, with its own population and questionnaire. In the definitive results corresponding to the year 2025 and the first quarter of 2026, published on October 22, 2026, the INE reported that 33.6% of companies with ten or more employees used AI technologies in the first quarter of 2026. The percentage should not be compared without more with the EBAE estimate: the coverage, survey design, and rate construction method change.

The table does not order the figures from lowest to highest nor presents a homogeneous evolution. It serves to separate measurement questions: how much technological capital there is and which companies declare using AI. To analyze a temporal trend, the source, population, and definition must be kept constant, or it must be carefully explained why different series are compared.

Market forecasts or growth scenarios can orient about expectations, potential investment, or future size of a sector. They are not, by themselves, measures of observed adoption. To value a forecast, one must identify what variable it projects, the time horizon, the assumptions, and whether it speaks of sales, spending, productivity, or number of user companies. A market projection does not prove that the technology is already deployed.

The CaixaBank Research analysis cited here was published in February 2020 and describes the measurement and trends examined in that work. To value the situation in another period, the comparison must resort to data corresponding to that period and the same variable. The INE reports business use of AI in the first quarter of 2026; the EBAE asks about AI utilization and collects categories of use intensity in its fourth quarter 2024 module. They are references with different dates and designs.

None of those statistics alone demonstrates the causal impact of the technology on productivity. To estimate it, economic results must be studied and factors such as training, organizational changes, complementary investment, and differences between companies considered. Presence, use, and impact are distinct levels of analysis; moving from one to the other requires specific data, not an automatic extrapolation.

For a company evaluating an investment, an aggregated capital indicator serves as context on the technological capacity of a country or sector, but it does not replace an evaluation of own processes. The declared use rate can show that a technology already appears in part of the business fabric, but it does not indicate if its application fits a specific organization, if it solves a need, or if it compensates its implementation costs.

A prudent reading separates three questions. What infrastructure exists? It is answered with stock and investment data. Who uses the technology? It is answered with adoption surveys that specify population and period. What result does it produce? Performance indicators and designs capable of distinguishing the effect of the technology from other simultaneous changes are needed.

The most useful conclusion is not that a technology is "adopted" or "yet to come" in absolute terms. It is to check what each figure measures before using it to decide. The 2020 study helps understand the accumulation of enabling capital; the Bank of Spain and INE surveys describe the declared use of AI in specific universes. Read together, they provide complementary context, as long as their differences in date, unit, and definition are kept visible.