Archetype 04 — The cooperative alternative

Archetype 04 — The cooperative alternative

A cooperative-alternative deep-dive for policy, cooperative, and academic audiences.

Header

What this talk is for

The audience is interested in alternatives — they’re already past the vendor-sweep question. They want to know: what does a non-vendor, non-state, non-corporate AI / data infrastructure look like in practice? What does it take to build? Where has it worked, where has it not? This talk walks them through three concrete cooperative / commons deployments (JoinData, GAIA, NAPDC), names the structural conditions that support them, and engages honestly with the limits of the model.

Run-of-show

1. Opening — the funding-model critique (5 min)

Frame. Open with the funding-model question, because that’s what makes the cooperative alternative necessary. Venture-style agritech investment has structural limits in agriculture. Hervé Pillaud (co-founder, La Ferme Digitale; farmer-livestock producer, retired from active farming):

“Unicorns do not exist in agriculture. The French agri-tech industry is not doing well, and funding from private investors remains insufficient.”

quotes/institutional-leaders/pillaud-herve-unicorns-not-in-agriculture.md. INRAE report on La Ferme Digitale / agritech AI (April 2025; 7-part dossier).

The structural point. The VC funding model — invest early, scale to billion-dollar valuations, exit — doesn’t apply to agriculture because the underlying market doesn’t support the returns. This is not a failure of effort; it is a structural feature of the market. Pillaud’s argument: the response is cooperative / commons infrastructure, not unicorn-scaling rhetoric.

Worth distinguishing (don’t conflate). Pillaud’s funding-model critique is structurally distinct from IPES-Food’s corporate-concentration critique. Different diagnosis, both real. Worth saying so because the cooperative alternative responds to both, but the mechanism is different for each.

2. Segment one — JoinData (12 min)

The world’s first agricultural data cooperative. Founded 2017 in the Netherlands. Independent, non-profit. Mission: “any farmer can pool, control, connect and share data — in a safe, secure and fair way — with agribusiness and innovation partners, and to make sure the data and benefits flow back to the farmer.”

Scale (Development Gateway / USAID case study, Feb 2023):

• 16,000+ farmer members
• 260 parties sharing data via JoinData
• 70 parties using data for the farmer
• €50/year farmer membership fee
• Companies pay fees for “data transport” — JoinData reinvests

Unit: units/joindata-netherlands.md.

Founding coalition. FrieslandCampina (largest Dutch dairy cooperative, multi-billion-euro), Agrifirm, LTO Nederland (Dutch farmers’ union), EDI-Circle (accountants), Rabobank (cooperative bank). The Dutch cooperative cultural tradition is structurally important — FrieslandCampina itself emerged from cooperative dairy traditions going back to the 19th century.

Producer voices (anchor the talk).

• Djessie Donkers (ZLTO): the vendor-fragmentation problem JoinData solves. quotes/producers/donkers-djessie-joindata-vendor-fragmentation.md.
• Mathé van den Bosch (dairy farmer, JoinData member): the milk-data case. quotes/producers/van-den-bosch-mathe-joindata-milk-data.md.

What JoinData is not (speak it).

• Not a data repository (sharing platform; doesn’t store farmer data)
• Not a vendor (non-profit, farmer-controlled)
• Not global yet (Netherlands-only as of source date; international expansion is an ambition)
• Not a substitute for Ag Data Transparent (control vs contract transparency)

Engage the limit. Per the case study, farmer-controlled data sharing infrastructure works in a cultural context that supports cooperation. This is a real, substantive observation worth surfacing: the model is replicable, not unique. The Dutch context (cooperative cultural tradition, FrieslandCampina scale, Rabobank presence) is structurally important. A copy-paste deployment in a non-cooperative cultural context will not succeed by default.

3. Segment two — GAIA and the French commons approach (10 min)

Move from data to AI. JoinData is data-cooperative. GAIA (La Ferme Digitale) is AI-cooperative. Different layer, complementary logic.

La Ferme Digitale. French agtech industry association. Co-founded by Hervé Pillaud. Mission is to coordinate French agritech actors around shared infrastructure rather than letting them compete for vendor-positioning.

Unit: units/la-ferme-digitale-gaia.md.

Anchor quote. Pillaud on the commons-and-mutual model: quotes/institutional-leaders/pillaud-herve-commons-and-mutual-model.md. On the culture of probability — the epistemic posture of working under uncertainty rather than claiming certainty: quotes/institutional-leaders/pillaud-herve-culture-of-probability.md.

GAIA — the AI layer. GAIA is the AI-infrastructure project of La Ferme Digitale. Open core, designed to be deployed cooperatively. Distinction from JoinData: GAIA operates at the model layer (foundation-model-style infrastructure for French agritech), JoinData at the data layer.

The structural point. The French and Dutch approaches are parallel implementations of the same logic — common infrastructure, member-controlled, non-vendor — at different layers of the agrifood AI stack. This is what cooperative / commons AI infrastructure actually looks like in practice.

Worth knowing — INRAE framing. The French national agricultural research institute (INRAE) frames La Ferme Digitale as the agritech hub. Unit: units/inrae-france-ai-agriculture.md. The state-research / cooperative / industry triangle is structurally important — INRAE provides research depth; La Ferme Digitale provides industry coordination; GAIA provides shared AI infrastructure.

4. Segment three — NAPDC and the US federally-funded cooperative development (8 min)

Move across the Atlantic. JoinData is bottom-up; NAPDC (Ag Data Cooperative / National Agricultural Producer Data Cooperative) is federally-funded top-down. Different model, important to know.

Unit: units/napdc-national-ag-producer-data-cooperative.md. US federally-funded framework-development cooperative. As of source date, framework-development phase; deployment is the next step.

Why this matters. The US model is public investment in cooperative framework development. The implication: government can fund the conditions for cooperative infrastructure without running it. This is structurally different from the Dutch and French models (where the cooperative tradition is the precondition) and from a fully state-stewarded model (where government runs the data).

Five cooperative / commons / state-stewarded models compared:

The cooperative / commons / state-stewarded model is not monolithic. Different countries reach it through different institutional pathways: industry-led (NL), industry-coordinated (FR), federally-funded framework (US), state-stewarded DPI (Japan), state-anchored cluster (Korea), cooperative-federation institutional substrate (Spain / Mondragón), cluster-with-three-structures cross-region pattern (Spanish cooperative AI + Brazilian seed AI). Worth saying this to the audience — the “alternative to vendor capture” is plural, not singular.

A new seventh, recently added — cluster-with-three-structures cross-region pattern (July 19 2026). The cluster-with-three-structures pattern operates across two regional contexts: Brazilian seed AI (peer-reviewed-led Tier-1 + multinational-corporate-pipelined Tier-2 + substantially-empty-at-Brazilian-origin Tier-3) and Spanish cooperative AI (peer-reviewed-led Tier-1 + Mondragón-federation-institutional Tier-2 + emerging-single-coop deployment Tier-3). The same cluster-with-three-structures pattern operates across two regional contexts with structurally-distinct Tier-2 substrates — corpus-valuable observation surfaced across six regional cycles.

A sixth, new addition — multilateral-state coordination (WAICO, July 2026). units/waico-alliance-china-multilateral-ai.md. The World Artificial Intelligence Cooperation Organisation is a treaty-based intergovernmental AI governance body headquartered in Shanghai, with 29 founding member states. WAICO is not a data-stewardship model — it is a governance layer above data-stewardship models, where vertical application governance (including agriculture) gets specified by member-state priority over the next decade. The institutional design choice is sovereignty-and-development-led, structurally distinct from IDSov (rights-and-Indigenous-sovereignty-led) and from cooperative-substrate-framed models (member-controlled-led). Treat the institutional design claim as V1; treat any operational / agricultural-deployment claim as V0 (announced, not documented). Worth naming in any audience concerned with how data-stewardship choices will be settled multilaterally — because the multilateral layer is where sovereign-rights-vs-development-priorities gets resolved.

Worth knowing what WAGRI isn’t. WAGRI is operational but it is not cooperative-governed; it is state-stewarded. Per the corpus’s pattern observation (units/japan-korea-agrifood-ai-pattern.md), Japan’s state-stewarded substrate operates above an equipment-vendor industrial-automation layer (Spread, Yamaha, Kubota, Yanmar). The state-DPI substrate complements vendor-layer deployment rather than replacing it. This is structurally different from JoinData, where the cooperative substrate is the operational layer.

Worth knowing what Korea’s Smart Farm Innovation Valley isn’t. Korea’s cluster is not a DPI substrate; it is a deployment cluster — a physical infrastructure of greenhouses, rental smart farms, test centres, startup incubation centres (SIC), and distribution centres anchored at four provincial sites. The unit units/korea-smart-farm-innovation-valley-rda.md carries the FAO Digital Villages profile for Sangju. The substrate (data layer) inside the cluster is vendor-proprietary at the site level; the cluster itself is state-stewarded at the programme level.

5. Segment four — OADA and the open-standards layer (6 min)

The infrastructure underneath. Open Ag Data Alliance (OADA) is the open-source interoperability standards project that cooperative / commons infrastructure can run on. Without open standards, even cooperatives end up with bespoke data formats that don’t interoperate.

Unit: units/oada-open-ag-data-alliance.md. Open-source project. Mission statements: quotes/institutional-mission-statements/oada-mission-statements.md.

The structural point. Cooperative / commons AI / data infrastructure is not just a funding model. It requires interoperability standards. OADA, Ag Data Transparent certification, IEEE 2890-2025 (Indigenous data provenance), and CARE Principles are the standards-layer of the cooperative / commons ecosystem. None of them are sufficient alone.

Mozilla framing. Mozilla State of Open Source AI 2026: units/mozilla-state-of-open-source-ai-2026.md, quotes/researchers-and-experts/mozilla-krikorian-open-source-turning-point.md. The open-source AI infrastructure is at a turning point in 2026 — Mozilla’s quantitative report (July 14, 2026) anchors the claim that open-source is becoming the default in some categories, not a marginal alternative.

6. Segment five — limits and contested claims (8 min)

Don’t end on boosterism. A 60-min academic talk earns its keep by engaging the limits honestly.

Contested claim C-029 (per taxonomy): “Data cooperatives are a global alternative to vendor capture.” Counter: cultural fit matters; replication requires cooperative tradition. The Netherlands model works because of 19th-century cooperative cultural infrastructure. France has it through La Ferme Digitale. The US has partial cooperative infrastructure (land-grant universities, USDA cooperatives, NAPDC development) but it’s a different texture. Canada has no deployed equivalent. Brazil, China, SSA, South Asia, Oceania: mostly absent.

Engage IPES-Food / critical voice. The cooperative alternative is a real response to corporate concentration in agrifood AI. But it is not the only response, and it is not the whole answer. Indigenous data sovereignty (units/indigenous-data-sovereignty.md) is a parallel framework with its own logic — CARE Principles, IEEE 2890-2025 — that is not reducible to farmer-cooperative frameworks.

The limit, named. The cooperative / commons alternative works in cultural contexts that support cooperation. The replication question — can this be built in contexts that don’t have cooperative traditions? — is genuinely open. NAPDC is one attempt; the answer is not yet in.

7. Close — what’s at stake (3 min)

Frame. Three things are at stake when cooperative / commons infrastructure is or isn’t built:

1. Who captures the value of aggregated farm data — farmer-controlled cooperatives keep the value with the farmers; proprietary platforms capture it externally.
2. Whether AI deployment reflects farmer priorities or vendor priorities — JoinData’s milk-data case (Mathé van den Bosch) is a substantive example of farmer-priority deployment.
3. Whether AI’s role in food sovereignty is structural or instrumental — if AI deployment runs on cooperative / commons infrastructure, food sovereignty is structurally enabled; if it runs on proprietary vendor infrastructure, food sovereignty is at best instrumental.

The single sentence. The cooperative alternative is real, deployed, and replicable in specific cultural conditions. The structural question is not whether it works (it does); it is whether the cultural conditions can be supported where they don’t exist.

Q&A handles

• “What about blockchain-based agricultural data platforms?” → Worth knowing: several blockchain / distributed-ledger proposals exist (e.g. JD Farm in China uses blockchain traceability). They are not the same as cooperative / commons infrastructure. Blockchain is a technical layer; cooperative is a governance layer. The two can combine but are not interchangeable.
• “What about the developing world?” → Cooperative / commons infrastructure in SSA, South Asia is thin. The replication question is genuine. Indigenous data sovereignty frameworks (units/indigenous-data-sovereignty.md) are a parallel and partially-overlapping framework — worth naming for any audience concerned with global food sovereignty.
• “What about Mondragón / Spanish cooperatives?” → scans/2026-07-spanish-cooperatives-ai.md for the cluster scan; three anchor units (units/mondragon-corporation-cooperative-federation.md — federation of worker cooperatives since 1956, €11.213B 2024 revenue, 70,000+ workers; units/spain-cooperative-agrifood-ai-cluster-pattern.md — Spanish cooperative AI cluster-with-three-structures observation; units/spain-cooperative-covap-ai-deployment.md — corpus’s first Spanish cooperative-led AI deployment-of-record with 5 named AI tracks). Worth surfacing: Mondragón is federation-substrate; JoinData is data-rights-substrate; both are cooperative-statutory frameworks but structurally distinct (multi-sector federation vs. single-sector data-rights).
• “Is Mozilla / open-source AI sufficient?” → Mozilla State of Open Source AI 2026 shows open-source is becoming default in some categories. But open-source models running on proprietary data infrastructure is a real pattern; cooperative / commons is a stricter requirement than open-source alone.

Freshness check

• JoinData figures: Feb 2023. Verify on annual cadence. Likely higher in 2026.
• La Ferme Digitale / GAIA: April 2025 INRAE report. Verify on annual cadence.
• NAPDC: development phase. Verify deployment status annually.
• OADA: open-source project. Verify activity annually (GitHub commits, mailing list activity).
• Mozilla State of Open Source AI 2026: July 2026. Confirm next-year report cadence.

Substitutions

• If audience is farmer co-op, swap segments 2-4 for the Canadian context (haven-greens, fcc-root-ai, the gap that there is no Canadian JoinData equivalent). Roughly archetype 02 with more depth.
• If audience is policy / leadership, shorten segments 4-6 and lead with the funding-model critique. Roughly archetype 03 with the cooperative spine.
• If audience is academic STS / political-economy, deepen segment 5 with contested claims and the parallel-but-distinct Indigenous data sovereignty framework.

What this archetype is doing in the methodology

This is the alternative-imaginary talk — the one a presenter gives when the room is full of people who already agree there’s a problem and want to know what the structural responses are. It’s the most demanding of the five archetypes because it requires the audience to hold multiple models in mind at once (JoinData / GAIA / NAPDC), engage with cultural-fit limits honestly, and not retreat to either boosterism or nihilism. The 60-min duration matters: it gives room for both the substantive deployments and the limit-engagement. The structural framing — real, deployed, replicable in specific cultural conditions — is the load-bearing claim. If the audience leaves thinking cooperative / commons is either a panacea or impossible, the talk failed.