There is broad consensus that artificial intelligence (AI) will have a significant economic and social impact, similar to past general-use technologies. However, the characteristics of this impact and the timeline for its realisation are highly uncertain.
The use of AI in production, one of the main avenues for the technology's impact, will depend on corporate decisions in three key areas: initial adoption of the technology, internal diffusion within the organisation, and final application in job roles and tasks.
The National Statistics Institute (INE) compiles data from the EU-ICT-Firm survey, which provides statistics on companies employing one of a list of eight AI technologies. The data indicates the adoption rate, which is the percentage of companies that have completed the first stage of the AI implementation process—deciding to use the technology. While this data is limited in scope, it is relevant as adoption precedes all subsequent stages.
According to INE data, in the first quarter of 2026, just over one-third (33.6%) of non-financial spanish companies with ten or more employees reported using one of the listed AI technologies. A year earlier, this proportion was 21.1%. The increase is notable: a rise of 12.4 percentage points in absolute terms and nearly 60% in relative terms.
The theory of technological diffusion can help address questions raised by these data: why some companies adopt new technology sooner than others and how these decisions translate into a penetration path within the productive fabric. Essentially, the process can be framed as follows: each company decides on the adoption of new technology by weighing its advantages (as technology becomes cheaper and its outcomes less uncertain over time) against its disadvantages (the risk of falling behind competitors). As this balance varies from company to company and moment to moment, adoptions are spread over time, creating a penetration path for the technology within the productive fabric.
This diffusion path or process can be described using a logistic function, as explained in a recently published study by Funcas. The results show a well-known pattern, an S-curve. The initial phase is slow; the pace accelerates when many companies are adopting, and it slows down again towards the end when few remain.
The function requires two data points: the rate at which the process advances and the starting level, which determines when half of the companies are reached. In the cited study, both are estimated using observed rates up to 2024, resulting in a central trajectory bounded by a faster and a slower path, corresponding to EU countries with faster and slower diffusion.
The logistic function represented in the study is y(t) = 1 / (1 + e^(-(-4.54 + 0.65t))). It is expected that the predicted diffusion path for the years leading up to 2024 aligns with the observed diffusion rate values, given that the parameters of the function were calibrated based on observed values.
What is truly relevant is whether the projected values for subsequent years, which were not used in the calibration, follow the expected path. The answer for the 2025 values (the latest data published by INE) is affirmative: the model anticipated a 34.5% adoption rate for that year, while INE published a 33.6% rate, a difference of less than one percentage point. If the path holds, the next INE statistic may approach 50%, with adoption expected to exceed 90% by 2030.
However, it is essential to accurately assess what it means to succeed or fail in the forecast. Success indicates that the underlying factors determining past adoption rates remain constant over time. Conversely, if the projection based on the logistic function does not hold, it suggests that these underlying factors have changed, prompting further inquiry into why.
Finally, it is important to remember that the AI adoption rate only informs about the speed of diffusion, the first stage of the process, until AI eventually impacts productivity and employment.
