Innovation without borders

Europe has devoted substantial resources to fostering innovation and AI diffusion, through both centralised EU initiatives and national programmes. This column combines an experiment on 3,316 euro area firms with data from a large firm survey to demonstrate that that the decision to adopt AI is primarily driven by competition pressures rather than informational spillovers. If firms systematically underestimate their competitors’ AI investment – especially those abroad – informational gaps could be driving underinvestment. Reducing invisible barriers to knowledge sharing across regions may be just as vital as financial support for closing gaps in AI investment and innovation across Europe.

The global rollout of artificial intelligence (AI) has sparked intense debate over a widening technology divide, characterised by highly uneven adoption rates across firms and workers and between macro-regions. When foundational innovations fail to diffuse effectively across regions, the full benefits for long-term productivity growth, regional convergence, and cross-country knowledge spillovers cannot be reaped. If technology adoption remains clustered exclusively within localised high-skill ecosystems or advanced economies, lagging areas may experience prolonged growth stagnation, hindering aggregate productivity gains and stalling catch-up mechanisms. 

The European Single Market, launched in 1993, aimed at reducing spatial frictions, boost cross-border competition, and drive technological diffusion. Since its launch, firms have, in principle, been exposed to competition from both domestic and foreign markets within Europe. Yet this integration has not fully eliminated market fragmentation. Recent reports by Draghi (2024) highlight Europe’s persistent productivity and innovation gaps, emphasising the need for AI integration and addressing market fragmentation in knowledge, innovation, and research systems. As a result, the EU offers a uniquely salient testing ground for assessing whether any remaining barriers are impeding the diffusion of innovations. 

The issue of technology diffusion across borders has important implications for growth and development. A number of studies have examined the forces or impediments of international knowledge diffusion and its geographic frictions (Coe and Helpman 1995, Eaton and Kortum 2002, Comin and Hobijn 2010, Keller and Yeaple 2013), as well as to research on technology adoption and firms’ exposure to foreign firms and markets (Pavcnik 2002, Amiti and Konings 2007, Alfaro-Urena et al. 2022, Bilgin et al. 2024). 

In a new paper (Baumann et al. 2026) we designed a survey experiment to shed light on the question of whether perceived competition helps explain the spread of innovation across European firms. We conducted a large-scale randomised control trial experiment on 3,316 euro area firms, linked to a large firm survey (the SAFE survey conducted by the ECB). Results indicate that while information travels across national borders, the strategic behaviour necessary to drive the adoption of knowledge and technology often fails to extend beyond these boundaries, hindering the progress of lagging firms. 

Widespread misperceptions and the centre–periphery mirage

The experiment first elicited firms’ prior beliefs about the share of firms, in the same sector and size category, that had invested in AI both domestically and within the ‘Big Three’ euro area economies (Germany, France, and Italy). The treatment information derived from responses to specific questions in the SAFE questionnaire, collected during the spring 2025 survey. Comparing these perceptions with actual adoption rates reported by firms (also gathered from the spring 2025 survey wave) reveals a significant gap: firms largely underestimate competitors’ AI adoption (an observation that confirms past specific country evidence; see Cullen et al. 2025 and Menkhoff 2025).

On average, managers underpredict domestic competitor investment by 14 percentage points and foreign investment by 7 percentage points. This underestimation is particularly severe in countries where actual AI investment is high, indicating that firm perceptions are heavily compressed relative to true cross-country variation.

Interestingly, the survey responses also document a clear centre–periphery misconception. Firms in smaller or peripheral economies systematically overstate the relative AI investment of the Big Three, believing that advanced technologies are more concentrated in few leading hubs than they actually are. In reality, AI investment is much more widely dispersed across Europe. 

This misperception alone, both domestically and abroad, may already explain why firms are not properly responding to competition by increasing their own AI adoption rate. The perception that competitors have adopted less than they did reduces competition pressure. The perception that the distance between large and small countries is too big to close may also discourage innovation.

The power of information and asymmetric extrapolation

To test how these perceptions influence corporate behaviour, half of the surveyed firms were randomly assigned to a treatment group that received information regarding actual peer AI investment rates in their country and the Big Three. The information provision successfully shifted beliefs of competitors’ adoption, both domestic and foreign. Figure 1 provides a first visual test of this mechanism. The first four bars correspond to average domestic and foreign posterior beliefs, with the blue (red) bars representing firms in the control (treatment) group. The average domestic posterior belief rises from 23.6% in the control group to 27.9% in the treatment group, a difference of 4.3 percentage points that is statistically significant. A similar pattern emerges for foreign posterior beliefs: the average rises from 28.2% to 31.5%, corresponding to a statistically significant treatment effect of 3.3 percentage points.

Figure 1 Treatment effects on posterior beliefs about domestic and foreign AI investment and on the AI investment rate 

Notes: Figure 1 presents treatment effects on posterior beliefs about domestic and foreign AI investment and on the AI investment rate, comparing control and treatment groups. Confidence intervals are at the 95% level. The figure is based on the Q4–2025 wave of the survey.

Moreover, we find that firms that initially underestimated the share of AI investors among their peers updated their posterior expectations upward, while those who overestimated adjusted downward. The implied Bayesian learning weights indicate that managers placed roughly a one-third weight on the provided signal and a two-thirds weight on their priors, supporting that they viewed the survey data as credible and relevant.

Interestingly, firms exhibit an asymmetric cross-market extrapolation pattern. When managers learn that foreign competitors are investing heavily in AI, they revise their forecasts for domestic investment upwards, expecting domestic markets to eventually catch up to the regional leaders. However, the reverse does not hold; information about domestic investment has zero impact on what firms believe is happening in the dominant Big Three economies.

Strategic complementarities stop at the border

The ultimate objective of the experiment was to determine whether updating a firm’s beliefs about its competitors causes the firm to alter its own innovation policy. On average, being exposed to the information treatment caused a statistically significant 1.8 percentage point increase in a firm’s own expected AI investment rate over the subsequent twelve months (see the last two bars in Figure 1).

To separate the domestic channel from the foreign channel, a two-stage least squares (2SLS) model is implemented that leverages the exogenous variation in posterior beliefs induced by the information shock. The 2SLS estimates uncover a stark asymmetry in strategic complementarities: a 1 percentage point increase in the perceived share of domestic competitors investing in AI raises a firm’s own expected AI investment rate by 0.57 percentage points. This implies an investment elasticity of 1.6. Conversely, the effect of an increase in the perceived share of foreign peers investing in AI on a firm’s own investment choices is statistically indistinguishable from zero.

We conduct a heterogeneity analysis to disentangle the underlying mechanisms, showing that the domestic peer effect is significantly stronger among smaller firms, firms operating in less concentrated markets, and firms with low export exposure. These findings suggest that the decision to adopt AI is primarily driven by competition pressures rather than informational spillovers. In other words, firms are motivated to adopt AI primarily to stay competitive when they observe peers doing so, rather than to gain insights into the potential returns through observation of others’ experiences.

Policy implications for the Single Market

Europe has devoted substantial resources to fostering innovation and AI diffusion, through both centralised EU initiatives and national programmes. Yet our evidence suggests that financial incentives may not be sufficient on their own. If firms systematically underestimate their competitors’ AI investment – especially those abroad – informational gaps could be driving underinvestment. Low-cost information campaigns that raise awareness of peer innovation could help accelerate technology diffusion. More broadly, reducing invisible barriers to knowledge sharing across regions may be just as vital as financial support for closing gaps in AI investment and innovation across Europe.

Source : VOXeu