A field experiment provides evidence that AI investment plans focus on the rivals next door.
Listen to this article:
We may be mistaking AI momentum for inevitability. AI capabilities seem to be advancing faster than ever, but technologies don’t diffuse automatically. They spread through thousands of individual decisions made by firms that often have an incomplete view of what is happening around them. That forms the starting point of the new working paper, “Innovation Without Borders? The Geography of Technological Diffusion,” co-written by 2026 HBS AI Institute Associate Zoë B. Cullen. Cullen and her co-authors conducted a massive field experiment involving 3,316 firms across twelve European Union countries. They measured how corporate leaders perceive rival AI investment choices, and their findings suggest that AI’s spread is more fragile, and far more geographically sticky, than the inevitability story assumes.
Key Insight: Firms Have No Idea What Their Competitors Are Doing
“Most firms underestimate both domestic and foreign competitors’ AI investment, often by substantial margins.” [1]
When asked to estimate AI adoption among similar peers (same sector and size), firms’ guesses fell far short of reality. While actual AI investment averaged 28% across twelve surveyed nations, ranging from 20% in Greece to 48% in Austria, firms underestimated domestic competitor investment by an average of 14 percentage points and foreign investment by 7 points. This gap was most severe in high-adoption markets, where firms assumed low investment levels regardless of national reality.
This misperception was driven by a strong “center-periphery” bias. In every country studied, firms wrongly assumed that the “Big-3” economies (Germany, France, and Italy) were out-investing their local market. In truth, Big-3 adoption averaged 25%, and six of the twelve countries actually out-invested the Big-3 average. Ultimately, firms had internalized a narrative about where innovation occurs that ran completely counter to the data.
Key Insight: Facts Change the Forecast
“The information intervention changed not only firms’ beliefs about competitors, but also their own intended AI investment over the next twelve months.” [2]
In another part of the experiment, half of the firms received the actual figures about AI adoption, and all firms were subsequently asked what share of domestic and foreign peers they expected would invest over the next year and what percentage of their own total investment they planned to allocate to AI. Because most firms had originally underestimated their rivals’ activity, the treatment increased expected domestic adoption by an average of 4.3 percentage points and expected foreign adoption by 3.3 points.
Most importantly, corrected misperceptions altered planned capital allocation. Firms receiving the information expected to allocate 10.13% of their total investment to AI over the following year, compared with 8.33% among the control group. This behavioral shift was strongest among firms whose original estimates had fallen furthest below the survey benchmark.
Key Insight: Global Signals, Local Reactions
“[I]deas can travel across borders, but their behavioral impact weakens with economic, cultural, and informational distance.” [3]
Although firms updated their beliefs about both domestic and foreign competitors, only the domestic information meaningfully affected their own investment plans. The researchers estimate that a one-percentage-point increase in expected domestic peer adoption raised a firm’s intended AI investment rate by 0.570 percentage points. The corresponding effect of foreign expectations was statistically insignificant.
Two mechanisms may explain the gap. Domestic firms may be perceived as more immediate competitive threats, making their investment harder to ignore. They may also offer more relevant evidence: a similar firm operating under the same institutions, labor market, customer expectations, and business practices may provide a more credible signal about whether AI will work locally. The experiment cannot cleanly separate those competition and learning effects.
Why This Matters
This research highlights that good AI strategy involves choosing the right reference points. A benchmark drawn only from global leaders might reveal what is technically possible, but fail to indicate what is competitively urgent or operationally relevant for your firm. Executives and business leaders need peer sets that reflect the markets in which they compete, the customers they serve, the talent they hire, and the conditions under which implementation must work. That means assigning different jobs to different comparisons: global examples can expand the possibility frontier, while closely matched domestic or regional peers may provide the stronger signal for timing, investment intensity, and execution. Finally, keep in mind that this study measured intentions, not spend. Revising a belief is the cheap part, converting it into success is harder.
Bonus
AI’s business value depends not only on whether firms adopt it, but also on how ideas, signals, and capabilities move between organizations. Competitive benchmarks can prompt firms to invest, while knowledge spillovers can help those investments generate new products, lower costs, and broader follow-on innovation, yet both processes weaken when information remains concentrated within particular firms, industries, or geographies. For a complementary look at how AI innovation creates measurable value through patents, profitability, and cross-industry spillovers, check out The Business Impact of AI Innovations: Driving Profitability and Growth.
References
[1] Baumann, Ursel, Zoë B. Cullen, Ester Faia, Annalisa Ferrando, Ricardo Perez-Truglia, and Judit Rariga, “Innovation without Borders? The Geography of Technological Diffusion,” NBER Working Paper 35314 (2026), 3. https://doi.org/10.3386/w35314.
[2] Baumann et al., “Innovation without Borders?” 4.
[3] Baumann et al., “Innovation without Borders?” 5.
Meet the Authors
Ursel Baumann is Deputy Head of Division, Capital Markets/Financial Structure, at the European Central Bank.
Zoë B. Cullen is Associate Professor of Business Administration at Harvard Business School and 2026 Associate at the HBS AI Institute.
Ester Faia is Professor at Goethe University Frankfurt.
Annalisa Ferrando is Senior Lead Economist, Capital Markets/Financial Structure, at the European Central Bank.
Ricardo Perez-Truglia is a Professor at UCLA’s Anderson School of Management.
Judit Rariga is an Economist, Capital Markets/Financial Structure, at the European Central Bank.