The State of AI in Spain: 2025 Quantitative Report

Four numbers that frame the situation

13.4% of Spanish companies with more than 10 employees use at least one AI technology. The EU-27 average is 13.5%. Spain sits exactly at the mean. Not lagging. Not leading.

That figure comes from the European Commission’s Digital Economy and Society Index (DESI) 2024. Like all averages, it conceals more than it reveals. Break the 13.4% down by company size and the picture shifts dramatically. Large Spanish enterprises (250+ employees) adopt AI at 49.2%, above Germany (48.1%) and just below France (50.3%). SMEs with 10-49 employees sit at 8.7%. That is not a gap. That is a chasm.

This whitepaper compiles and analyzes the quantitative data available on the state of AI in Spain. It is not a speculative exercise about what AI could do. It is an inventory of what it is doing, what it costs, where it works, and where it does not. Sources are public: INE (the Spanish national statistics institute), Bank of Spain (quarterly digital economy reports), European Commission (DESI, AI Watch), and sector reports from AMETIC and Cotec.

Investment: how much and where

Spain allocated an estimated 1.847 billion euros to artificial intelligence in 2024, according to the National Observatory of Technology and Society (ONTSI). That represents 0.14% of GDP. For comparison: France allocates 0.21%, Germany 0.19%, and Finland, the European leader, 0.37%.

The aggregate number is less interesting than the composition:

Public sector: 612 million (33%). The ENIA national AI strategy and Next Generation EU funds are the primary drivers. The Public Administration Digitalization Plan 2021-2025 has a specific AI budget of 500 million euros, of which roughly 340 million had been executed by June 2025. The most visible projects: the Spanish Tax Agency’s virtual assistant (processing 2.3 million monthly queries), Social Security’s fraud detection system, and predictive justice pilots in commercial courts in Madrid and Barcelona.

Large enterprises: 789 million (43%). IBEX-35 companies absorb a disproportionate share. Telefonica, BBVA, Santander, CaixaBank, and Inditex together account for over 40% of this figure. The pattern is clear: large Spanish companies invest in AI at the same level as their European peers. But there are 4,600 companies with 250+ employees in Spain, and half have not started any AI project.

SMEs and startups: 446 million (24%). A relative surprise here. Spain’s AI startup ecosystem grew 34% by number of companies between 2022 and 2024, according to Barcelona Supercomputing Center mapping. Barcelona and Madrid concentrate 78% of these companies, with an emerging cluster in Valencia linked to logistics and agrifood.

The question investment data cannot answer is efficiency. How much return does each euro generate? Granular national data does not exist, but the Bank of Spain estimates that AI-adopting companies report an average labor productivity increase of 4.7% within the first two years. That number does not justify euphoria, but it does justify the investment for those who approach it with discipline.

Adoption by sector: leaders and laggards

AI adoption in Spain is not uniform. The cross-sector variance is enormous, and the leading sectors are not always the expected ones.

Financial services: 38.2% adoption

The most advanced sector, by a wide margin. Spanish financial institutions have invested in predictive models, fraud detection, and process automation for over a decade. BBVA has more than 400 AI models in production. CaixaBank processes 100% of consumer credit applications with ML-based scoring. Santander reports its fraud detection models catch 94% of fraudulent transactions with a 0.3% false positive rate.

What sets financial services apart is not technology (the models are relatively conventional: XGBoost, shallow neural networks, language models for NLP) but operational maturity. They have MLOps teams, robust data pipelines, model governance, and years of regulatory compliance experience. AI is not an innovation project for them. It is infrastructure.

Retail and logistics: 22.1% adoption

The second sector, driven by two forces: supply chain optimization and customer experience personalization. Inditex is the paradigmatic case (its demand forecasting system processes 400 million items per year, adjusting production and distribution in 2-week cycles), but meaningful adoption exists across mid-sized companies.

Logistics specifically has found AI applications with immediate returns: route optimization, demand forecasting, warehouse management, and shipment anomaly detection. A food distribution company we worked with reduced fleet kilometers by 12% with a route optimization system combining constraint programming with traffic prediction. ROI was positive in 4 months.

Manufacturing: 17.3% adoption

Spanish industry is adopting AI, but at a lower rate than Germany (24.8%) or Italy (19.1%). Dominant applications are predictive maintenance, visual quality control (computer vision on production lines), and process optimization. The automotive sector leads, with SEAT/CUPRA and their suppliers as primary adopters.

A barrier specific to Spanish industry: company size. 96% of industrial companies are SMEs, and most operate on margins that cannot absorb 200,000 or 300,000 euro AI projects. Government digitalization programs have helped, but the AI solutions they cover are basic (chatbots, marketing tools) and do not address the high-impact industrial use cases.

Healthcare: 14.6% adoption

The Spanish healthcare sector has enormous AI potential (data from 47 million patients in a universal public health system) but relatively low adoption for two reasons: strict regulation (justified) and system fragmentation across 17 autonomous communities with incompatible information systems.

The most advanced projects focus on diagnostic imaging (Vall d’Hebron Hospital in Barcelona, La Paz Hospital in Madrid) and readmission prediction. But scaling a model trained on Catalan health system data to the Andalusian system requires an integration effort nobody has solved at scale.

Construction: 6.2% adoption

The most lagging sector among Spain’s major industries. No surprise there. Construction is a sector with low general digitalization: only 34% of Spanish construction companies have an ERP, according to the INE. If you do not have digitized data, you cannot apply AI. It is that simple.

Exceptions exist. Ferrovial and ACS have innovation divisions with AI projects in project management, construction planning, and workplace safety. But these are exceptions from companies with revenues exceeding 5 billion euros.

Barriers: why adoption is not faster

The INE’s ICT survey asks directly about AI adoption barriers. The 2024 results are revealing:

1. Talent shortage (67%). The most cited barrier, and the most real. Spain produces roughly 3,200 graduates from programs directly related to AI per year (masters and PhDs in ML, data science, robotics). LinkedIn Economic Graph estimates annual demand at 12,000 positions. The deficit is structural and will not resolve in the short term.

The median salary for a senior ML engineer in Spain is 55,000-70,000 euros gross, per Glassdoor and Levels.fyi data. In Germany, 75,000-95,000. In the UK, 80,000-110,000. In the US, 150,000-250,000. The brain drain is not a narrative. It is a data point.

2. Cost (54%). The second most cited barrier, but more nuanced than it appears. Model costs have dropped dramatically: a GPT-4 call costs 90% less than 18 months ago. The real cost is not in inference but in integration. Connecting an AI model to a company’s existing systems (ERP, CRM, legacy systems) is an engineering project requiring months of work and qualified personnel. Which brings us back to barrier 1.

3. Regulatory uncertainty (41%). The EU AI Act entered partial application in February 2025. Spanish companies, especially in regulated sectors (finance, health, insurance), cite uncertainty about risk classification of their systems as an investment brake. They do not want to invest in a system that might require substantial modifications for compliance.

This is partly rational and partly excuse. The AI Act’s obligations for high-risk systems are demanding (technical documentation, conformity assessment, post-deployment monitoring), but most enterprise AI applications do not fall into the high-risk category. A product recommendation system is not high risk. A customer service chatbot is not high risk. The perceived regulatory complexity exceeds the actual complexity.

4. Data quality (38%). Garbage in, garbage out. Companies that have not invested in data engineering cannot do AI. A demand forecasting model needs clean, consistent, complete historical data. If your ERP has empty fields, duplicates, and inconsistent categories, the model will learn garbage. We have seen AI projects fail not because of the AI but because the first three months went to cleaning data the company believed was already clean.

5. Lack of clear use cases (29%). “We know AI is important, but we don’t know where to apply it.” It is the fifth barrier, but perhaps it should be the first priority to resolve. Companies that start with a concrete, measurable, bounded use case (classify invoices, predict cancellations, optimize scheduling) have significantly higher success rates than those that start with “we want to do something with AI.”

Measurable outcomes: what the evidence says

Talking about potential is easy. Showing results is another matter. What do the data say about the real impact of AI on Spanish businesses?

The Bank of Spain published in March 2025 an analysis based on the Central Balance Sheet Office that cross-references AI adoption data with financial indicators. The findings:

• AI-adopting companies show labor productivity growth 4.7% higher than non-adopters, controlling for size, sector, and company age.
• The effect is non-linear. Companies with AI in production for over two years show 7.2% gains. Those with less than one year, 1.8%. AI needs time to generate returns.
• There is no evidence of net employment destruction. AI-adopting companies hire, on average, 2.3% more than non-adopters. The profile shifts (more technical roles, fewer administrative roles), but the net balance is positive.
• The sector with highest measured impact is financial services (+8.1% productivity), followed by logistics (+6.3%) and professional services (+4.9%).

These numbers are consistent with international literature. Goldman Sachs’ January 2025 report estimates a global generative AI productivity impact of 1.5% cumulative annually. McKinsey estimates that 60-70% of generative AI’s economic potential comes from automating tasks within existing functions, not from new functions.

What these reports do not say (because they lack the data) is how many AI projects fail. Our experience and that of other sector consultancies suggests 60-70% of enterprise AI projects do not reach production. The reasons are the barriers already described: insufficient data, lack of integration, misaligned expectations. The 4.7% productivity gain is the result of surviving projects, not all initiated projects.

European comparison: where Spain actually stands

Spain’s position in the European context requires more nuance than a simple ranking. It depends on what you measure:

General enterprise adoption: Position 14 of 27 in the EU. At the mean. Denmark leads at 24.1%, followed by Finland (22.8%) and the Netherlands (19.7%).

Research: Spain ranks 4th in the EU for scientific publications on AI (behind Germany, France, and Italy) and 5th in patents. CSIC, the polytechnic universities of Madrid, Barcelona, and Valencia, and centers like BSC produce world-class research. There is no research capability deficit.

Startup ecosystem: Spain ranks 8th in Europe by number of AI startups, per the Stanford AI Index. Barcelona is Europe’s 4th AI hub after London, Paris, and Berlin. Madrid does not appear in hub rankings but has more AI companies focused on finance and telecommunications than Barcelona.

VC investment in AI: 487 million euros in 2024, per Dealroom. Far from France (2.1 billion) and Germany (1.3 billion), but growing at 28% year-over-year. The capital gap is real but closing.

Compute infrastructure: BSC hosts MareNostrum 5, one of Europe’s 10 most powerful supercomputers. But GPU cloud access for companies remains expensive and data center supply in Spain is limited compared to Ireland, the Netherlands, or the Nordics. Planned openings by Amazon (Aragon), Google (Malaga), and Microsoft (Madrid) of cloud regions in Spain will change this in 2025-2026.

The SME gap: the central problem

If one data point summarizes the AI problem in Spain, it is this: AI adoption among companies with 250+ employees is 49.2%. Among companies with 10-49 employees, it is 8.7%. And companies with 10-49 employees represent 85% of Spain’s business fabric.

This is not a willingness problem. It is a resource problem. An SME with 20 employees and 3 million in revenue does not have a data department, does not have an ML engineer, and its “data infrastructure” is probably a shared Excel and a half-implemented CRM. Telling that company to “adopt AI” is like telling someone without a car to enter Formula 1.

Realistic solutions for SMEs follow three paths:

1. AI embedded in existing tools. You do not need your own AI project. You need business tools with built-in AI. Your CRM already has predictive scoring. Your email marketing platform already has subject line optimization. Your accounting tool is already adding automatic classification. This is the highest-impact short-term path.

2. Pre-configured vertical solutions. SaaS products built for a specific sector and use case. A warehouse management system with integrated demand forecasting. A customer service platform with automated responses pre-trained for your sector. The cost drops from 200,000 euros for a custom project to 500-2,000 euros per month for a vertical SaaS.

3. AI managed services. Outsource AI implementation and operation to a specialized provider, the same way companies outsource accounting or digital marketing. This model is emerging in Spain and is, in our view, the best fit for the country’s business structure. The SME does not need to understand machine learning. It needs someone to do it and charge for results.

Public policy: what works and what does not

The ENIA strategy has three pillars: talent, ecosystem, and adoption. A quick results review:

Talent: Five AI chairs have been created at public universities and 1,200 predoctoral fellowships funded for AI. It is a start, but it does not resolve the structural deficit. The problem is not training researchers (Spain does that well) but retaining them. Without competitive salaries and a business ecosystem that absorbs AI talent, we train researchers for export.

Ecosystem: Acceleration programs (INCIBE for cybersecurity with AI, CDTI for R&D projects) have worked reasonably well. The Spanish government’s AI regulatory sandbox, one of Europe’s first, has processed 28 pilot projects with a 43% conversion rate to product. That is a positive signal.

Adoption: The Kit Digital program has distributed digitalization vouchers to over 800,000 SMEs. But the specific AI allocation within Kit Digital is modest, and accredited solutions are basic. For most SMEs, Kit Digital has served to get a website and a CRM, not to adopt AI. It is a necessary prerequisite, but it is not AI adoption.

Projections: what to expect from 2025-2027

Projections are inherently wide-margin exercises. That said, available data support several estimates:

Enterprise adoption will reach 20-22% by end of 2027, driven primarily by AI embedded in SaaS tools, not by custom projects. Adoption among large enterprises will converge with the EU average (55-60%).

Investment will grow 15-20% annually, pushed by three factors: EU funds (which start executing seriously in 2025-2026), inference cost reduction (making previously unviable projects viable), and competitive pressure (companies that do not adopt AI watch their competitors do so).

Talent will remain the primary barrier. Unless there is a significant shift in skilled immigration policy or sector salary levels, Spain will not close its AI talent deficit before 2030.

Regulation will clarify. Full AI Act application in August 2026 will eliminate regulatory uncertainty as a barrier, though it will create new compliance costs for high-risk systems.

What these numbers mean

Spain does not have an AI problem. It has a scale problem. Research capability exists. Large enterprises compete at European level. The startup ecosystem grows. What is missing is for that capability to reach the 85% of the business fabric that consists of SMEs.

Solving that is not a technology problem. It is a business model problem. Whoever figures out how to make AI accessible, useful, and affordable for a 20-person company will capture an enormous market. Not just in Spain. Across southern Europe, with similar business structures.

The data say we are at the mean. The question is whether the mean is enough. For companies already adopting AI, probably yes. For the country as a whole, probably not. The difference between today’s 13% and the 25% of leading countries is not measured in percentage points. It is measured in productivity, competitiveness, and ultimately, living standards.

This report will be updated quarterly as new data are published. The numbers will change. The underlying conclusion probably will not: AI in Spain works where serious investment occurs. The challenge is expanding the “where.” For the more qualitative analysis of these same trends, see our article on Spain and AI adoption. And to translate these numbers into a concrete budget for your company, our breakdown of real AI implementation costs for SMEs details it with data from our own projects.