Artificial intelligence (AI) is emerging as a double-edged sword in the labor market. On one hand, 40% of small and medium-sized enterprises (SMEs) are adopting it to address labor shortages, seeing it as a solution to recruitment difficulties that have become structural. On the other hand, an identical percentage of employers, across all sectors, hesitate to take the plunge, hindered by a glaring lack of internal skills. This contradictory dynamic reveals a profound divide within OECD economies: an AI skills gap that isolates large companies, which are agile and pioneering, from SMEs struggling not to be left behind. This article aims to analyze the multiple dimensions of this divide, examine the training policies implemented to address it, and evaluate the impact of this gap on overall productivity.
The Great Divide: A Multi-Faceted Skills Fracture
Data from the Organisation for Economic Co-operation and Development (OECD) paints an unequivocal picture of the situation. A 2025 report on AI adoption by SMEs highlights an alarming disparity: half of the SMEs surveyed in four G7 countries acknowledge that their employees lack the skills required to use generative AI. This finding is aggravated by the fact that approximately one-third of these same SMEs have faced labor shortages and a general lack of skills or experience within their workforce over the past two years [1].
This skills deficit directly impacts adoption rates. In 2024, while 40% of large companies (with 250 employees or more) had integrated AI into their operations, only 11.9% of small companies (10-49 employees) had done the same. This gap of nearly 30 percentage points illustrates a significant lag that is not limited to general adoption but extends to specific applications. SMEs are particularly behind in using technologies like autonomous robots, where the adoption gap with large companies is most pronounced (7.2% versus 0.7%). The gap is slightly less pronounced for tools like natural language generation (16.7% versus 4.6%), suggesting that SMEs prioritize the most accessible and least expensive AI applications [1].
Source: OECD, "AI adoption by small and medium-sized enterprises", 2025.
The purposes of AI use also differ. While large companies deploy AI in strategic areas such as R&D, logistics, or information systems security, SMEs more readily confine it to marketing or sales functions. A 2025 OECD survey shows that, although a significant portion of SMEs use generative AI, they do so primarily for peripheral tasks that support operations without radically transforming production processes. Only 29% of SME users report employing it in their core activities [1]. This deficit in adoption and deep integration constitutes a major obstacle to SME competitiveness and innovation.
Training Policies with Uneven Results to Bridge the Gap
Aware of the urgency, several governments have launched ambitious training programs to attempt to bridge this skills gap. The analysis of these initiatives, with their varying degrees of success, offers valuable insights.
The Singapore Model: SkillsFuture
Singapore has established itself as a reference with its SkillsFuture program. In 2025, the country recorded record participation with 606,000 people engaged in training supported by SkillsFuture Singapore (SSG), an increase from 555,000 in 2024. Notably, more than 105,000 of these training sessions concerned artificial intelligence, demonstrating the program's ability to align with the needs of a rapidly changing labor market. The program's success rests on an individual credit system that has encouraged more than half of eligible Singaporeans (30-75 years old) to train. However, this individual success contrasts with a decline in employer-initiated training, attributed to a more conservative economic climate [2].
Germany and the Qualifizierungschancengesetz: A Promise Awaiting Confirmation
Germany has opted for a targeted approach with the Qualifizierungschancengesetz (Qualification Opportunities Act). This mechanism aims to finance continuing education for employees whose jobs are directly threatened by digital transformation and automation. Support is particularly generous for very small companies (fewer than 10 employees), which can obtain full reimbursement of training costs. While the initiative is promising, its actual effectiveness is still difficult to assess. According to the Institute for Employment Research (IAB), the law has not yet caused a significant jump in the number of subsidized training programs, and no comprehensive official evaluation has been published to date [3].
The Personal Training Account (CPF) in France: A Tool at Two Speeds
In France, the Personal Training Account (CPF) has succeeded in democratizing access to training for millions of people. It has proven particularly effective in disseminating basic digital skills and mastery of common software. Nevertheless, the CPF struggles to stimulate the acquisition of advanced skills. In 2021, a paltry 90 trainees only followed artificial intelligence training through this system. The analysis of the CPF also reveals that it reproduces the gender imbalances present in the French educational system, with more specialized STEM fields remaining overwhelmingly male. The CPF thus seems to be an excellent tool for general upskilling, but it is not yet the appropriate lever for training the AI experts the country needs [4].
Beyond Skills: Other Barriers to AI Adoption by SMEs
While the lack of skills is the most frequently cited barrier, it is not the only one. The OECD report identifies three other major obstacles that hinder AI adoption by SMEs: connectivity, access to data and computing resources, and financing.
High-quality connectivity is the foundation of any digital transformation. However, significant disparities persist between urban and rural areas in many G7 countries, penalizing SMEs located outside major economic centers. Similarly, access to quality data and the computing power needed to train and deploy AI models remains a major challenge. Large companies have vast proprietary datasets and the means to invest in computing infrastructure, an overwhelming competitive advantage.
Finally, financing remains the crux of the matter. SMEs face structural difficulties in accessing bank credit, due to asymmetric information, lack of guarantees, and limited credit history. In a context of tightening credit conditions, financing long-term investments in emerging technologies like AI becomes a nearly insurmountable challenge for many, forcing them to focus on short-term financing needs for their immediate survival [1].
The Productivity Paradox and the Threat of a Two-Speed Economy
AI's potential to boost productivity is colossal. The OECD puts forward promising estimates, with annual labor productivity growth that could reach between 0.2 and 1.3 percentage points in G7 economies over the next decade. The highest gains are expected in the United States and United Kingdom, while Japan and Italy could experience more modest growth. Generative AI, in particular, is perceived as a general-purpose technology capable of reinventing entire swaths of the economy [1].
However, this potential remains largely theoretical for a large part of the economic fabric. The productivity paradox, where a revolutionary technology fails to spread its effects throughout the economy, threatens to create a two-speed economy. On one side, large companies, equipped with the necessary human and financial resources, capitalize on AI to increase their efficiency and market dominance. On the other side, SMEs, unable to overcome entry barriers, risk stagnation and loss of competitiveness. This scenario is not only detrimental to SMEs themselves; it threatens the resilience and dynamism of the entire economy, which relies largely on the vitality of its small and medium-sized enterprise fabric.
OECD research shows that the most productive companies are also those that adopt AI the most. In France, in 2018, the adoption rate of companies in the top productivity decile was 40% higher than that of companies in the bottom decile. This gap reaches 120% in Germany and even 240% in Italy in 2020. This correlation is partly explained by a selection effect: companies that are already more competitive and digitized are more inclined to adopt AI. But it also suggests that AI could become a multiplier of inequalities, widening the gap between leaders and others [1].
Moreover, AI-related productivity gains are not immediate. They often follow a J-curve: an initial drop in productivity due to investment and reorganization costs, followed by an increase once complementary investments (training, new processes) bear fruit. SMEs, with their shorter time horizon and more limited resources, are less able to absorb this initial dip, which constitutes an additional barrier to investment [1].
Conclusion: A Call for Coordinated Action for a Just Transition
The AI skills divide is not inevitable, but a market and public policy failure that requires a coordinated and multidimensional response. The examples of Singapore, Germany, and France, despite their imperfections, offer food for thought. It is clear that there is no single solution. Training policies must be both ambitious in their objectives and finely targeted in their implementation. They must aim not only to improve basic digital skills for the entire workforce, but also to cultivate cutting-edge AI expertise within a pool of specialized talent.
For SMEs not to be the forgotten players of the AI revolution, it is imperative to go beyond training alone. Resolute action is necessary to improve connectivity throughout the territory, to democratize access to data and computing infrastructure, and to create innovative financing mechanisms adapted to SME needs. This could involve creating dedicated investment funds, sectoral data-sharing platforms, or shared competence centers where SMEs could experiment with AI at lower cost.
The stakes are high. The goal is to ensure that AI's promise of productivity and innovation is a rising tide that lifts all boats, not a wave that only benefits the largest ships. The future of our economies' competitiveness and social cohesion depends on it. The time for observations is over; it's time for action.
References
- [1] OECD. (2025). AI adoption by small and medium-sized enterprises. https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/12/ai-adoption-by-small-and-medium-sized-enterprises_9c48eae6/426399c1-en.pdf
- [2] The Straits Times. (2026, February 9). Over 1 in 2 S'poreans aged 30 to 75 used SkillsFuture credit, surge driven by year-end deadline: SSG. https://www.straitstimes.com/singapore/parenting-education/1-in-2-sporeans-aged-30-to-75-used-skillsfuture-credit-surge-driven-by-year-end-deadline-ssg
- [3] Cedefop. (n.d.). Qualification Opportunities Act. https://www.cedefop.europa.eu/en/tools/matching-skills/all-instruments/qualification-opportunities-act
- [4] Bruegel. (2023, December 20). Promoting STEM skills: a brief assessment of French individual learning accounts. https://www.bruegel.org/analysis/promoting-stem-skills-brief-assessment-french-individual-learning-accounts
- Receive Journal analyses directly in your mailbox.
- China's CO2 emissions fell by 0.3% in 2025. It's a modest figure. But it extends a trend that has lasted for 21 consecutive months: since March 2024, emissions from the world's largest polluter have been "stable or declining."
- Does AI destroy jobs? Two studies published in March 2026 — one by Harvard Business School, the other by Anthropic — provide the first solid empirical data. And the answer is more nuanced than public debate suggests.
- Malaria killed 608,000 people in 2022, 95% of them in sub-Saharan Africa and 78% children under five years old. And for the first time, a vaccine deployed on a large scale shows measurable results in the field.
- Wanting to be 20 today. An independent media that documents progress with rigor, without naivety or catastrophism.
- Structured reading sheets: central thesis, key arguments, limitations, and verdict.
- JdP is an independent editorial project based on data, counter-narratives, and lucid optimism. Each article is sourced, nuanced, and open to discussion.
- The Journal of a Progressive uses cookies to improve the reading experience and understand how the site is used. No data is collected for commercial, advertising, or resale purposes. Cookies necessary for site operation are always active. Optional cookies are only activated with your explicit consent, in accordance with GDPR.