Europe’s applied-AI push starts on the farm.
On 24 June 2026 the European Commission opened the sectoral phase of its Apply AI Strategy with agriculture. The choice says less about farming than about how Europe now intends to make AI pay at the scale of a small business.
Gli Insights sono pubblicati in inglese.
On 24 June 2026, the European Commission’s directorates-general for agriculture (DG AGRI) and for communications networks, content and technology (DG CNECT) held the first structured sectoral dialogue under the Apply AI Strategy, adopted on 8 October 2025. Around 180 experts from industry, research, farming organisations and public authorities took part.
The event report, published on 3 July 2026, records a consensus notable for its restraint: “AI should support, rather than replace farmers’ decision-making”, and it must deliver measurable value at farm level. Trustworthiness, security and usable design were listed as conditions, not aspirations.
The choice of sector is the informative part. Agriculture is not where Europe’s most advanced AI research sits. It is where the distance between what the technology can do and what is used is easiest to measure, and where that distance is economic rather than technical.
The barriers named at the dialogue are integration problems, not capability problems
The overview presented by the OECD set out four obstacles: limited digital infrastructure, uncertain return on investment, insufficient interoperability, and difficulty integrating AI tools into existing farm management information systems. The event report adds the absence of connectivity between platforms, for instance between farm management systems and public subsidy systems.
None of them concerns model quality. They describe the last metre of the chain: whether an output reaches the system where the decision is taken, and whether anyone can prove it was worth producing.
McKinsey’s Global Farmer Insights 2024, a survey of roughly 4,400 farmers across nine countries in Europe, North America, Latin America and India, quantifies the result: 27 per cent of European farmers use digital yield monitoring, against 60 per cent in North America. Reviewing the same survey, the Global Agricultural Productivity Initiative at Virginia Tech records unclear return on investment and high implementation costs as the barriers farmers cite consistently, and farm size as a critical determinant.
Farm structure, not scepticism, sets the ceiling on European adoption
Eurostat counted 8.8 million agricultural holdings in the European Union in 2023, averaging 17.8 hectares. A provisional 62.8 per cent work fewer than five hectares, while the 7.6 per cent above 50 hectares account for 68.3 per cent of the utilised agricultural area. Between 2005 and 2023 the EU lost 5.6 million holdings, about 39 per cent of them, while the area actually farmed fell by 0.6 per cent: farms have merged without the land base changing.
Against that structure, McKinsey’s figures read differently from the way they are usually quoted. The average surveyed European farm covers about 450 acres, roughly 180 hectares — ten times the EU average, because a commercial survey samples commercial operations. Its American counterpart averages some 3,500 acres, about 1,400 hectares, and the Brazilian one about 3,700. Even Europe’s commercial farms therefore operate at roughly an eighth of American scale, above a long tail of holdings smaller again by an order of magnitude.
The mechanism is arithmetic. Most of the cost of an agronomic AI system is fixed: sensors, connectivity, integration work, licence, and the time spent learning to trust the output. The return is variable and accrues per hectare. Divide the same fixed cost by 180 hectares rather than 1,400 and the payback period changes category. A third of European farmers in the survey say their farms lack the scale to justify agtech; in the United States, farms above 2,500 acres are 45 per cent more likely to adopt agtech than farms below 100.
The European response is to move fixed costs off the farm balance sheet
The instruments discussed at the dialogue share one logic: mutualise whatever a single holding cannot amortise.
- The common European agricultural data space, one of fourteen sectoral data spaces under development, is intended to allow trustworthy pooling and sharing of agricultural data between private stakeholders and public authorities.
- agrifoodTEF, the testing and experimentation facility for AI and robotics in agri-food, is co-funded by the Digital Europe Programme under grant agreement 101100622 and runs ten nodes across ten countries, letting suppliers validate technologies on real farms rather than in a laboratory.
- AI factories provide compute on EuroHPC supercomputers with priority access for start-ups and SMEs: nineteen selected since December 2024, plus thirteen antennas, within combined investment of some €10 billion over 2021–2027.
- The Horizon Europe partnership Agriculture of Data and the CAP’s investment measures complete the set; CAP support for digital farming technology is expected to reach more than 274 000 farms.
The sector’s size explains the effort: the Commission’s Vision for Agriculture and Food, of 19 February 2025, puts the EU agri-food system at €900 billion of added value and some 30 million jobs, with 12 per cent of farmers under 40.
None of this buys a farm a sensor. It lowers the cost of finding out whether the sensor pays.
Swiss farm structure makes the same argument, and Swiss evidence complicates it
Switzerland counted 47 075 farms in 2024, 1.3 per cent fewer than in 2023, averaging 22.1 hectares — a little above the EU figure and a fraction of the American one — and employing 147 876 people, according to the Federal Office for Agriculture’s Agrarbericht 2025. On the scale argument alone, Swiss agriculture should adopt slowly.
Swiss evidence suggests the argument is incomplete. An Agroscope survey of Swiss farm managers conducted in 2021, reported in the same Agrarbericht, examined acceptance of two concrete technologies: virtual fencing in livestock and autonomous weeding robots in arable farming. About 40 per cent of respondents were favourably disposed towards digitalisation, and only a small minority rejected it. The determinants were financial situation, age, education and self-assessed digital skills. Farm size, number of workers and workload were not significant.
Read together, the two data sets describe two constraints. Scale determines what a farm can afford; capability and confidence determine what it will use. A policy addressing only cost moves the first and leaves the second. Agroscope’s own recommendation — peer networks, model farms and test facilities — is in substance the instrument list the European Union is now funding at scale.
What a company can decide without waiting for the policy
Five decisions sit inside the firm, not the programme.
- Treat machine data as a contractual right, not a vendor courtesy. The Data Act, Regulation (EU) 2023/2854, has applied since 12 September 2025. Users of connected products — the Commission lists farmers among them — may access the data their use generates and have it shared with third parties, and connected devices placed on the EU market must be designed to allow that sharing. Buyers should specify the access route in the purchase contract; for a Swiss manufacturer, the obligation follows the product onto the EU market.
- Specify interoperability before purchase. The barrier named at the dialogue was integration with existing systems, not model performance. Export formats, documented interfaces and the right to extract historical records belong in the tender, where they cost nothing, rather than in year three, where they cost a migration.
- Use the shared facilities that already exist. agrifoodTEF tests agri-food AI and robotics under real conditions, and AI factories give priority access to SMEs and start-ups. Independent validation is cheaper than a lost pilot.
- Join the sectoral forum. The Apply AI Alliance stakeholders’ catalogue is open to organisations established in the EEA and in Switzerland, and the Alliance supports annual sectoral workshops. Swiss firms are inside this process by design.
- Measure value per unit of production, not adoption. The consensus recorded on 24 June was that value must be measurable at farm level. The discipline elsewhere is the same: define, before the pilot, the denominator by which the fixed cost will be divided.
The dates that will carry information
Two dates belong in the calendar. The EU AgRI 2040 conference takes place in Brussels on 24–25 September 2026, and stakeholders have been invited to the Apply AI Summit in November 2026. The agenda, the event report and the presentations delivered on 24 June, the OECD’s included, are already published on the dialogue’s page — the fullest record of what was argued.
The wider lesson is not about farming. Where the fixed cost of a technology exceeds what a single small operator can amortise, adoption does not wait for better models; it waits for someone to carry that cost collectively. That condition describes much of the European economy, in which the median firm is small and the technology is priced as though it were not.