The state of AI-powered corporate training in Spain 2026: data, gaps and trends

 

In 2025, large Spanish companies trained 91.5% of their eligible workforce, but only 9.4 hours per participant: corporate training has become universal and thin at the same time.

AI is entering Spanish companies far faster than it is entering their training function.

You can see it by putting side by side the two statistics that are almost never read together: Fundae's, which records how much training people working in Spain actually receive, and the INE's, which measures how much artificial intelligence sits inside those same companies. One is moving fast; the other spreads fewer hours each year across more people.

In this article we bring together the data published in 2025 and early 2026, point out the four gaps that explain the situation, put in writing which concrete signals to watch in the next reports, and describe the product category emerging to solve it.

 

How much training actually happens in Spain: the 2025 numbers

 

Fundae's annual report, published in June 2026, is the most complete source available on training in Spanish companies, because it isn't based on perception surveys but on training that was actually delivered and subsidised.¹

In 2025, 348,443 companies ran subsidised training, 20.5% of the 1,698,311 companies eligible to do so. Between them they trained 3,542,960 people, adding up to 6,124,445 course participations at an average of 13.1 hours each.

The aggregate rises slightly against 2024: 3% more participants and 1,001 more training companies. The coverage rate stays pinned at the same 20.5%.

That 13.1-hour average hides the interesting part, because it behaves very differently depending on company size.

 

 

Nine out of ten large companies train; among micro companies, one in seven does. And the relationship between size and hours per person runs in exactly the opposite direction to what people usually assume.

 

The size paradox: the bigger the company, the fewer hours each person gets

 

A company with more than 250 employees trains nine out of ten of its eligible staff, four times the coverage of a micro company. But each of those people receives 9.4 hours of training a year, against the 37.2 hours of someone working in a company with fewer than ten.

The striking part isn't one year's figure, it's the series. In large companies, the average hours per participant has been falling steadily for fourteen years while the number of participants multiplies.

 

 

Large Spanish companies now train 2.5 times more people than in 2011, with half the hours per person. The total volume of hours has barely moved in fourteen years: it has been spread across many more people.

There are two possible readings of this curve, and it's worth holding both at once, because the Fundae data doesn't let you choose between them.

The first is about composition. As headcount grows, training fills up with short mandatory modules (safety, compliance, data protection, product) that repeat every year and pull the average down. It fits with the fact that 62.1% of participants attend in-person courses of barely 9.3 hours on average.

The second is about capacity. A ten-person training team doesn't grow at the pace of a workforce going from 1,000 to 4,000 employees, so the only variable it can adjust is duration. We call this the Training Production Ceiling: the point where an organisation can no longer add training hours because it can't produce more content with the same team, and starts rationing minutes instead of expanding programmes.

Here is what that looks like on the ground. A plant approves a change to a machine procedure in March. The module explaining it gets rebuilt in June, because there was no production slot before that. The people who joined in April were trained on the previous version, and nobody spots it until the next internal audit. Training didn't fail: what failed was the time it takes content to catch up with the procedure.

Neither reading is a proven cause; they are interpretations of a correlation. But both point to the same place for anyone running training in a company of more than 200 people: at scale, training becomes a content production problem, not a pedagogical one. It's the same tension that shows up when you have to cover the mandatory annual training hours per employee without expanding the team.

 

Half the training budget goes unused

 

If the constraint were budget, the training credit would be exhausted. It isn't.

In 2025 companies drew down 674.3 million euros out of an assigned credit of 1,301.6 million: 51.8%. Almost half the money companies have already paid in to train their people went unused.

 

 

The band that makes worst use of its credit is 10 to 49 employees, which consumes 38.8%. Companies with more than 249 reach 56.9%, and even so leave more than 219 million euros on the table.

The average cost of training one participant was 277 euros, of which the company put in 170 and the subsidy covered 107. On those figures, the credit left unused in 2025 would have paid for several million additional participations.

What matters for a training lead is the diagnosis that follows: the bottleneck isn't funding, it's the capacity to design, produce and update content in time. We break down the full mechanism in our guide on how to make the most of the Fundae training credit.

 

AI is already inside Spanish companies, but not yet inside their training

 

The other picture comes from the INE. According to the definitive data from its survey on ICT use in companies, 21.1% of companies with 10 or more employees were using artificial intelligence in the first quarter of 2025, 8.7 points more than a year earlier.²

By sector, services leads at 25.7%, industry sits at 17.5% and construction at 11.4%. Among companies with fewer than ten employees, adoption reaches 13.4%, almost six points more than the year before.

Two things worth separating here, because they get mixed up. One is teaching AI, training staff to use these tools with judgement. The other is producing training with AI, using it to generate, translate and keep training content current. They are different projects, with different owners, and they rarely get solved together by accident.

The first is just starting and lags behind adoption. Accenture's Pulse of Change survey, published in January 2026, picked up a notable perception gap: 90% of the executives surveyed assumed their workforce had the basic training needed to use AI, against 52% of employees.³ Only 26% of employees said they felt able to use these tools confidently, and 15% felt leadership had clearly explained how AI would affect them.

It's a perception study with a limited sample (around 1,070 executives, 120 of them in Spain, and 700 employees across Europe, in 20 sectors), so it's better read as a directional signal than as a measurement of the Spanish market. Even so, the 38-point gap between what leadership believes and what staff report is hard to explain by margin of error alone.

The second, producing training with AI, is where the quieter mismatch sits. A company can have adopted AI in customer support, in development or in marketing and still maintain its training catalogue exactly as it did in 2019: documents rewritten by hand and videos that have to be re-recorded every time a procedure changes.

 

Four gaps that define 2026

 

Crossing both sources produces four concrete mismatches. We've ordered them from most measurable to most structural.

 

Gap 1: high coverage, low depth

 

Companies with more than 250 employees reach 91.5% of their eligible workforce with 9.4 hours per person per year. In practice, each employee gets a little over one working day of training spread across several short modules through the year. The organisation complies, but it will struggle to transform a skill on that time budget.

 

Gap 2: budget available, execution halfway

 

48.2% of the assigned credit went unused in 2025. When a budgeted line goes unspent systematically year after year, the problem is rarely willingness: it tends to be operational capacity to turn a budget into concrete training before the year closes.

 

Gap 3: what leadership believes and what staff report

 

The distance between the 90% of executives and the 52% of employees who consider basic AI training covered has a direct operational consequence. If the leadership team considers AI literacy solved, it doesn't budget for it. And if it doesn't budget for it, the gap widens precisely as adoption accelerates.

 

Gap 4: AI adoption without training governance

 

One in five companies with 10 or more employees already uses AI, but very few have documented who has been trained, on what and when. That is exactly what the AI literacy obligation in the EU AI Act requires, turning that traceability into a requirement rather than good practice. We break it down in our guide on the mandatory training plan under Article 4.

 

What to watch through the rest of 2026 and into 2027

 

Trends are only useful if they can be checked. These are the four we consider most likely, with the concrete signal that would confirm or disprove each one and the source where it will show up.

 

 

There's a fifth trend that doesn't yet appear in any official statistic and is worth watching closely: the change in how training is measured. As long as reports keep counting hours per participant, they will keep describing a classroom model. In a 2,000-person company, the useful question is no longer how many training hours we deliver, but how long a procedure change takes to reach the person who has to apply it, updated, and how many people can demonstrate they know how to apply it.

 

The category taking shape: from content to system

 

An autonomous learning system takes the knowledge that already exists in an organisation, turns it into structured training, checks that it has been understood, and keeps it current when the procedure changes, without a person having to rebuild the whole cycle each time.

For fifteen years the corporate training market has been organised into two categories. Creation tools solve production: they turn a document or a script into consumable content. LMSs solve distribution and record-keeping: they hand out that content and log who completed it.

Neither is designed for what happens next. Content ages, the procedure changes, and someone has to start over. That accounts for much of the operational explanation of why hours per participant have been falling for fourteen years: the underlying problem usually isn't content, it's knowledge architecture.

What's emerging now, and doesn't yet have a place in any official statistic, is a third category: the Autonomous Learning System. The difference isn't whether the tool uses AI, but who holds up the cycle.

 

 

It's worth being precise about where this actually stands. Today most Spanish organisations work in the first column, some have reached the second, and the third describes where the category is heading more than an already settled standard. There are platforms built on that premise (Vidext is one of them), though it's worth separating the generation side, which is fairly mature, from the autonomous side of the cycle, where agents detect outdated content, which is considerably more recent.

The practical implication for 2027 is about design, not procurement. An organisation that wants to stop rationing minutes has to decide first which part of the cycle it will stop holding up by hand.

 

Conclusion: the question is no longer how much training you deliver, but how fast it updates

 

The 2025 data describes a system that works on coverage and fails on depth. Almost every large company trains, almost half the available money goes unused, and each person receives half the hours they did fourteen years ago.

AI has entered Spanish companies at a pace that's unusual for a technology this recent, but it has entered through processes, not through training. As long as the training function keeps measuring its work in hours delivered, it will keep optimising the wrong variable.

If you're rethinking your organisation's training model for 2027, you can see it applied in a demo, though the underlying decision comes before any tool: a company isn't trained by the hours it delivers, but by how fast its knowledge reaches the person who has to apply it, up to date.

 

Frequently asked questions

 

How many training hours does an employee in Spain receive per year?

 

In 2025, the average was 13.1 hours per participant in company-programmed training. The figure varies a lot by size: 37.2 hours in companies with fewer than 10 employees and 9.4 hours in companies with more than 250.

 

What percentage of Spanish companies run subsidised training?

 

20.5% of the companies eligible to do so, that is, 348,443 out of 1,698,311. The share rises to 80.3% in companies with 50 to 249 employees and to 91.5% in those with more than 250.

 

How much training credit goes unused each year?

 

In 2025 companies drew down 674.3 million euros out of an assigned credit of 1,301.6 million, 51.8%. The band that makes least use of it is 10 to 49 employees, at 38.8%.

 

How many Spanish companies use artificial intelligence?

 

21.1% of companies with 10 or more employees, according to definitive INE data for the first quarter of 2025. That's 8.7 points more than the previous year. In companies with fewer than 10 employees, adoption is 13.4%.

 

Which sectors lead AI adoption in Spain?

 

Services leads adoption at 25.7% of companies with 10 or more employees, followed by industry at 17.5% and construction at 11.4%.

 

Why do large companies deliver fewer training hours per person?

 

The data shows the correlation, not the cause. The two most plausible explanations are catalogue composition (more short, repetitive mandatory modules as headcount grows) and the production limit of the training team, which doesn't grow at the pace of the organisation.

 

Is AI training mandatory for the workforce?

 

The EU AI Act introduces an AI literacy obligation for organisations using these systems, with traceability requirements on who has received training and on what. The specific scope depends on the type of system and on whether the company acts as a provider or a deployer.

 

What is an autonomous learning system?

 

It's a system that takes the knowledge already present in an organisation, turns it into structured training, verifies it has been understood, and keeps it current when the procedure changes. It differs from a creation tool, which only solves production, and from an LMS, which solves distribution and completion records.

 

Where is official data on company training in Spain published?

 

Fundae publishes its "Formación en el trabajo" report each June with the previous year's data. The INE publishes its survey on ICT use and e-commerce in companies in the autumn, which includes artificial intelligence adoption.

 

Sources

 

¹ Formación en el trabajo 2025. Informe anual, total nacional - Fundae ² Encuesta sobre el uso de TIC y del comercio electrónico en las empresas. Año 2024 - Primer trimestre 2025. Datos definitivos - INE ³ IA en 2026: la brecha de confianza entre directivos y empleados se agrava (Accenture Pulse of Change survey) - RRHH Digital