AI adoption in Spanish firms is advancing rapidly but remains limited and uneven

Between 2021 and 2025, the adoption of AI technologies in Spain’s economy has more than doubled in firms with over 10 employees,1 going from 8% to 21%, suggesting that AI is no longer an experimental technology. Nevertheless, in 2025, nearly 8 out of 10 firms were still not using it, suggesting that widespread adoption has not yet occurred.2

The first key conclusion is that company size is a decisive factor. AI is present in nearly 3 out of every 5 large firms, but only in 18% of those with fewer than 50 employees. This gap reflects barriers that go beyond technology, related to financial resources, the availability of data, qualified personnel and organisational capacity.

These differences have not diminished in the period 2021-2025. Although adoption is growing across all company sizes, progress has been far more rapid in large firms (+25 pps) than in small ones (+11 pps). Medium-sized companies (between 50 and 250 employees) reflect a turning point: their level of adoption (31%) is significantly higher than that of small businesses (18%), suggesting they meet a threshold of sufficient resources to experiment with AI.

If microenterprises were included in the analysis, the aggregate adoption rate would fall considerably: instead of 21%, it would be 14%, as these entities have a very low adoption rate (13%) and account for 95% of Spain's productive fabric.3

In 2025, the sectors with the highest adoption of AI were information and communications and the ICT sector,4 with percentages of around 60% – significantly above the levels observed in 2021 (26-27%). In these sectors, AI has become quite prevalent, in line with their greater intensity in intangible capital, the availability of data and their proximity to technological supply.

In a second tier are knowledge-intensive services, such as professional and scientific activities (38.5%) and real estate activities (35%). Thirdly, a wide range of sectors show moderate adoption levels, at around 20%-26%, spanning both services and industry: basic supplies, hospitality and the automotive and electronics sectors.

At the lower end we find construction, metallurgy, and transport and logistics, where adoption remains low. Overall, although some sectors are approaching widespread AI usage, the economy remains in an intermediate phase, with significant margin for future dissemination concentrated in the sectors and firms that are lagging the furthest behind.

The next issue is to identify the specific functions for which AI is being used in each sector. To this end, we distinguish between two major types of applications. On one hand, we find AI used in processes, aimed at improving organisational and commercial efficiency, such as internal management, administrative tasks or sales support. On the other hand, there is AI linked to the productive core, that is, uses applied directly in the production of goods or in the provision of the main service.5

In this context, adoption is considered superficial when AI is primarily used in processes rather than in the production of goods or services; and it is considered deep when AI is employed to a greater extent in production rather than in management tasks. There is another group of sectors where AI adoption is currently in a transition phase, moving from its use in processes towards greater use linked to production. Most of the adoption in sectors such as hospitality, commerce and administrative activities is found in processes. In these cases, AI is initially introduced in cross-functional management and commercial support roles, where implementation costs and organisational risks are lower, before extending to the productive core.

In an intermediate position we find sectors such as food and textiles, professional and scientific activities, and construction, with a transitional adoption where both types of uses coexist. AI is beginning to permeate operational decisions, but it does not yet play a dominant role in core business activities.

Finally, sectors such as information and communications and the ICT sector show a high level of adoption both in processes and in the productive core. In these fields, AI is not merely an efficiency tool but an integral part of the product, service and digital infrastructure.

Overall, the pattern observed suggests that the recent progress in adoption within the Spanish economy has been primarily driven by horizontal uses, which are quick to implement, low-cost, and more oriented towards administrative tasks than production processes. This is consistent with what Daron Acemoglu has documented for the US, where improving quality and the reliability of processes is the main motivation for adopting AI.6 The big unknown for the impact on aggregate productivity is when the use of AI will become widespread in operational and core business uses, which require greater integration, investment, and the redesign of processes.