Something historically unusual is happening with artificial intelligence. For most general-purpose technologies, businesses led and households followed. Electric dynamos powered factories for roughly four decades before households electrified at scale, and the personal computer entered offices years before it entered living rooms (Comin & Hobijn, 2010; David, 1990). Generative AI has inverted that sequence: by August 2024, nearly 39% of the U.S. working-age population had already used it — a faster initial diffusion than either the PC or the internet at comparable points (Bick et al., 2024) — while firm-level surveys were still reporting adoption under 10% (Bonney et al., 2024). Workers, quite literally, dragged AI into their companies in their pockets.
But the consumer sprint does not mean firms can sprint. Turning a free browser tab into production use requires procurement, data governance, workflow redesign, and retraining — the slow, intangible "complementary investments" formalized as the productivity J-curve, which explains why new technologies so often arrive long before their payoff (Brynjolfsson et al., 2021). Firm adoption was always going to be slower.
The question is: how slow, and where? Fresh Eurostat data on the share of European enterprises using AI technologies give us the clearest answer yet (Eurostat, 2025) — and it reveals a continent splitting into distinct speeds.
What the data say
The figures below come from Eurostat's survey of ICT usage in enterprises (firms with 10+ employees), tracking the share reporting use of any AI technology — machine learning, text mining, natural language processing, image recognition, and similar tools (Eurostat, 2025).
The euro-area aggregate tells the headline story: 8.8% (2023) → 14.4% (2024) → 21.4% (2025). Firm adoption more than doubled in two years. The ChatGPT moment reached European boardrooms with a lag of roughly eighteen months, and the adoption S-curve — the canonical shape of technology diffusion since Griliches (1957) and Bass (1969) — is now in its steep phase. But that average conceals enormous variation.
| Group | Country | 2023 | 2024 | 2025 |
|---|---|---|---|---|
| Nordics + NL | Denmark | 15.2 | 27.6 | 42.0 |
| Finland | 15.1 | 24.4 | 37.8 | |
| Sweden | 10.4 | 25.1 | 35.0 | |
| Netherlands | 14.1 | 23.9 | 33.2 | |
| Norway | 9.2 | 20.8 | 28.9 | |
| Continental | Germany | 11.6 | 19.8 | 26.0 |
| France | 5.9 | 9.9 | 18.2 | |
| Poland | 3.7 | 5.9 | 8.4 | |
| Mediterranean | Spain | 9.2 | 11.3 | 20.3 |
| Italy | 5.1 | 8.2 | 16.4 | |
| Greece | 4.0 | 9.8 | 8.9 | |
| Reference | Euro area | 8.8 | 14.4 | 21.4 |
Source: Eurostat (2025). A methodological note before we proceed: several series carry break flags, and pre-2023 figures used a broader AI definition — which is why Denmark's anomalous 23.9% in 2021 should not be compared directly with later years. These numbers measure the extensive margin — whether firms use AI at all — not how deeply they use it (for intensive-margin evidence from usage data, see Chatterji et al., 2025; Handa et al., 2025).
The Nordics and the Netherlands: the digital vanguard
The top of the ranking is entirely Scandinavian, with the Dutch alongside. Denmark leads at 42% — more than four in ten firms — followed by Finland (37.8%), Sweden (35.0%), and the Netherlands (33.2%). Norway, outside the EU, tracks the same pattern at 28.9%.
The growth rates are as impressive as the levels. Sweden more than tripled adoption in two years (10.4% → 35.0%). These countries combine the classic preconditions: near-universal broadband, high digital skills, high wages that make automation attractive, strong English proficiency (an underrated advantage when tools are English-first), and institutional trust that lowers the friction of adopting new systems. The pattern is consistent with firm-level microdata showing AI adoption concentrated among larger, digitally mature businesses (McElheran et al., 2024). The Nordics are effectively answering the question "what does the middle of the S-curve look like?" for the rest of Europe.
The continental core: Germany pulls, France chases, Poland trails
Germany (26.0%) sits comfortably above the euro average, having more than doubled since 2023 — the Mittelstand is adopting steadily, if without Nordic speed. The real continental story, though, is France. For all its AI ambition — home to Mistral and an assertive national AI strategy — French firm adoption stood at just 5.9% in 2023, barely half the euro average. Since then it has tripled to 18.2%, one of the fastest growth rates in the table, and the gap with Germany narrowed from ten points to eight. Rhetoric and reality are converging, but from a low base.
Then there is Poland at 8.4% — less than half the euro average and barely a fifth of Denmark. For Europe's sixth-largest economy and a supposed beneficiary of supply-chain re-shoring, that is a strikingly weak position.
The Mediterranean: one sea, two stories
The "PIGS" label (Portugal, Italy, Greece, Spain — Portugal is not in this extract) implies a bloc, but the data show divergence. Spain (20.3%) is converging with the core, jumping nearly nine points in a single year after a sluggish 2023–24. Italy (16.4%) remains below average but doubled in one year (8.2% → 16.4%) — the fastest single-year acceleration in the table. Greece (8.9%) is the one genuine worry: after leaping from 4.0% to 9.8%, it stalled and even ticked down in 2025 — the only country shown where momentum broke. Whether that is survey noise or a real plateau, it marks Greece as the country most at risk of missing this technology wave entirely. The region's SME-heavy firm structure is a plausible drag: the firm-size adoption gradient documented in U.S. microdata (McElheran et al., 2024) implies that economies dominated by small firms adopt more slowly.
Central and Eastern Europe: highly uneven
This extract shows only Poland, but the full Eurostat dataset reveals the widest intra-regional spread in the EU: the Baltic states and parts of the former "cohesion" economies (Slovenia, Czechia) sit at or above the EU average, while Poland, Romania, and Bulgaria cluster near the very bottom (Eurostat, 2025). The drivers are structural — an SME-heavy firm population, sectoral mixes tilted toward manufacturing and agriculture, thinner digital-services ecosystems, and ongoing emigration of precisely the digitally skilled workers who would champion adoption internally. The risk is a two-speed Europe in which AI-driven productivity gains compound existing east–west income gaps instead of narrowing them — precisely the uneven-diffusion dynamic that historical work on general-purpose technologies warns about (Crafts, 2021).
Conclusion: the window is open, but not for long
Three facts summarize the moment. First, European firm adoption is real and accelerating — the euro area went from under 9% to over 21% in two years (Eurostat, 2025). Second, the dispersion is vast: Danish firms adopt at nearly five times the rate of Polish or Greek ones. Third, the consumer-first inversion means employees everywhere are ahead of their employers (Bick et al., 2024; Bonney et al., 2024) — and in the laggard countries, firms are squandering a workforce that is already fluent in these tools.
Three recommendations follow:
- For policymakers in laggard countries: the binding constraint is not access to models — those are cheap and global — but complementary capacity (Brynjolfsson et al., 2021). Prioritize SME digital advisory services, data-governance support, and workforce AI skills, and deploy EU cohesion and recovery funds toward exactly these bottlenecks. The French and Italian 2025 jumps show catch-up is possible; Greece shows it is not automatic.
- For firms: stop counting pilots and start redesigning workflows. The experimental evidence suggests gains come from reorganizing work around the technology, not bolting it on — and that the largest gains accrue to less-experienced workers, a rare skill-leveling pattern that argues for broad-based deployment rather than elite AI teams (Brynjolfsson et al., 2025; Noy & Zhang, 2023). And treat your employees' informal "bring-your-own-AI" use as market research, not a compliance violation.
- For statisticians and researchers: harmonize definitions across waves, report breaks transparently, and move beyond counting adopters to measuring intensity of use (Chatterji et al., 2025; Handa et al., 2025) — the extensive margin is where the data are, but the intensive margin is where the productivity will be.
The historical pattern of general-purpose technologies is that adoption gaps eventually close — but slowly, and the productivity gaps they open can persist for decades (Comin & Hobijn, 2010; Jovanovic & Rousseau, 2005). Europe's north is already mid-S-curve. Whether the south and east catch up on the steep part of the curve, or after it flattens, is one of the most consequential economic questions of the decade.
References
Bass, F. M. (1969). A new product growth model for consumer durables. Management Science, 15(5), 215–227.
Bick, A., Blandin, A., & Deming, D. J. (2024). The rapid adoption of generative AI (NBER Working Paper No. 32966). National Bureau of Economic Research. https://www.nber.org/papers/w32966
Bonney, K., Breaux, C., Buffington, C., Dinlersoz, E., Foster, L. S., Goldschlag, N., Haltiwanger, J. C., Kroff, Z., & Savage, K. (2024). Tracking firm use of AI in real time: A snapshot from the Business Trends and Outlook Survey (NBER Working Paper No. 32319). National Bureau of Economic Research. https://www.nber.org/papers/w32319
Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942.
Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The productivity J-curve: How intangibles complement general purpose technologies. American Economic Journal: Macroeconomics, 13(1), 333–372.
Chatterji, A., Cunningham, T., Deming, D. J., Hitzig, Z., Ong, C., Shan, C. Y., & Wadman, K. (2025). How people use ChatGPT (NBER Working Paper No. 34255). National Bureau of Economic Research. https://www.nber.org/papers/w34255
Comin, D., & Hobijn, B. (2010). An exploration of technology diffusion. American Economic Review, 100(5), 2031–2059.
Crafts, N. (2021). Artificial intelligence as a general-purpose technology: An historical perspective. Oxford Review of Economic Policy, 37(3), 521–536.
David, P. A. (1990). The dynamo and the computer: An historical perspective on the modern productivity paradox. American Economic Review, 80(2), 355–361.
Eurostat. (2025). Use of artificial intelligence in enterprises [Data set]. https://ec.europa.eu/eurostat
Griliches, Z. (1957). Hybrid corn: An exploration in the economics of technological change. Econometrica, 25(4), 501–522.
Handa, K., Tamkin, A., McCain, M., Huang, S., Durmus, E., Heck, S., Mueller, J., Hong, J., Ritchie, S., Belonax, T., Troy, K. K., Amodei, D., Kaplan, J., Clark, J., & Ganguli, D. (2025). Which economic tasks are performed with AI? Evidence from millions of Claude conversations. arXiv. https://arxiv.org/abs/2503.04761
Jovanovic, B., & Rousseau, P. L. (2005). General purpose technologies. In P. Aghion & S. N. Durlauf (Eds.), Handbook of economic growth (Vol. 1, pp. 1181–1224). Elsevier.
McElheran, K., Li, J. F., Brynjolfsson, E., Kroff, Z., Dinlersoz, E., Foster, L. S., & Zolas, N. (2024). AI adoption in America: Who, what, and where. Journal of Economics & Management Strategy, 33(2), 375–415.
Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.