What the penetration of new technologies in Spain really measures

CaixaBank Research’s indicator starts with a specific question: what share of the Spanish economy is represented by capital associated with new technologies? Published on 13 February 2020, the study builds a measure to estimate that penetration and examine its possible relationship with future growth. Its subject is the stock of technological capital, not how often a particular application is used.

To construct the measure, the analysis uses EU KLEMS data covering four types of capital: software and databases, research and development, computers, and telecommunications equipment. It relates their combined stock to hours worked and expresses the result in 2010 euros per hour. The unit connects capital with labour, rather than counting companies that use technology.

The authors explain that their choice reflects a measurement challenge: when the study was conducted, there were no data directly quantifying artificial intelligence in economic terms. They therefore use asset categories that can contribute to developing such technologies, even though they also serve other purposes. The measure represents a broad set of technological capabilities, not a separate valuation of AI.

The date matters too. The study was published in 2020, and its results describe the series and period analysed at the time. It helps explain a methodology and a historical trajectory, but it should not be read as a current measurement of the Spanish economy.

An indicator of capital per hour worked and a survey of business technology use do not measure the same thing. The first relates a stock of assets to labour; the second asks companies whether they use specified technologies. They offer complementary perspectives, but their percentages and units are not interchangeable.

The INE’s ICT survey also distinguishes between reference periods: technology-use variables refer to the first quarter of the survey year, while e-commerce sales relate to the previous year. In the edition published on 22 October 2026, therefore, the AI-adoption figure describes the first quarter of 2026, while the e-commerce sales data correspond to 2025.

Among companies with 10 or more employees, the INE recorded that 33.6% used artificial intelligence technologies in the first quarter of 2026 and 39.4% purchased cloud computing services. Among companies with fewer than 10 employees, 21.4% reported using AI. Company size changes the observed rate.

These figures make it possible to compare groups by reported use, but on their own they do not show the investment level of each company or the associated economic results. Other metrics are needed to examine those dimensions, such as spending, intensity of use, or changes in operational indicators.

In the study’s comparison of advanced economies, the United States ranked first in technological capital per hour worked. France and Germany were in the middle range, while Spain and the United Kingdom appeared at lower levels. CaixaBank Research also described rapid growth in Spain’s measure over the two decades examined. A lower level can coexist with rapid progress.

The article suggested that Spain might approach France’s or Germany’s levels in about ten years if the observed trend continued. This was a conditional projection, not a subsequently confirmed outcome or a guarantee. The estimate depended on the pattern persisting that had been recorded in that analysis.

The sector breakdown for Spain showed differences between activities with high levels of technological capital and those with medium or low levels. The study associated higher levels with sectors classified by Eurostat as high-technology industries or knowledge-intensive services. The analysis covered 31 industry and service sectors, excluding agriculture and mining from that breakdown; both activities were, however, included in the national aggregate used for cross-country comparisons.

The authors noted that sectoral differences did not converge during the period analysed. They also considered it possible that future advances might extend to less technology-intensive activities, but presented this as a possibility. The observed relationship does not by itself establish what causes the differences or how they will develop.

The INE survey offers a more recent perspective on companies’ reported adoption. Alongside AI use, it recorded that 34.3% of companies with 10 or more employees carried out data analytics using their own staff, and 39.4% purchased cloud services in the first quarter of 2026. These figures measure reported practices, not technological capital per hour worked.

The difference by company size matters when interpreting adoption: AI use stood at 33.6% among companies with 10 or more employees and 21.4% among those with fewer than 10. These are proportions for different groups, so the figure for larger companies is not automatically an appropriate target for a small business.

The INE cautions that 2025 data are not strictly comparable with those of previous years because of changes to the CNAE activity classification and to the population covered. Comparability between survey editions has limits that should be considered when assessing changes over time.

Taken together, the two measures distinguish the accumulation of technological resources from companies’ reported adoption. Their figures should not be added or combined: they describe different units, questions, and reference periods.

The technological-capital measure brings together broad categories. Software, databases, computers, telecommunications, and R&D can all contribute to AI deployment, but they are also used for other purposes. The indicator does not isolate investment exclusively in AI from the wider set of assets used as a proxy.

The ratio per hour worked relates capital to labour and makes it possible to compare relative stocks of assets. On its own, however, it does not quantify the returns on each investment or the share of productivity attributable to technology. Examining those effects requires relating the measure to economic outcomes and other factors.

The article also notes that some industrial robots may be classified under general machinery categories. This makes their technological content harder to identify in capital statistics. The observation helps clarify the classification’s scope, but it does not turn the indicator into a direct count of robots.

Consequently, CaixaBank Research’s analysis provides an approximation of the economic importance of technological capital during the period studied and of differences between countries and sectors. Measuring the stock and evaluating its effects are separate tasks: the first does not, by itself, establish a causal impact on growth.

For a company, the practical question is not only whether other organisations are adopting AI or cloud services, but also what specific problem it wants to solve and how it will identify an improvement. The INE statistics put reported use of certain technologies in context; CaixaBank Research’s measure offers a historical reference for technological capital. Neither replaces analysis of the business itself.

An internal assessment can specify which asset or service is being introduced, which process will use it, and what result will be observed. Depending on the objective, that result might be the time needed to complete a task, the error rate, or operating costs. Establishing a baseline beforehand makes it possible to compare changes without confusing adoption with benefit.

When reading national figures, it is worth checking the year, the size of the companies included, and the indicator’s definition. Accumulated capital, reported use, and outcomes should also be kept separate. For example, the INE survey assigns different reference periods to ICT use and e-commerce sales.

The conclusion is limited in scope: the 2020 study estimated the technological capital of the Spanish economy and described differences between countries and sectors in the period analysed; the INE survey published in 2026 records companies’ reported use of AI, cloud services, and data analytics by business group. The two measures clarify different aspects and do not, on their own, determine how much each technology contributes to growth.