The Center is pleased to publish a piece by Trebor Scholz, in anticipation of the upcoming international conference at Chulalongkorn University from November 12 to 15, 2026.
*
SOLIDARITY AI
R. Trebor Scholz
Solidarity AI begins with a proposition. Workers and communities should own and govern AI, from the land and data centers it runs on to the data, models, and labor that make it work. We propose connecting cooperatives, unions, and public institutions so they can govern these layers together. This essay takes the layers one at a time, asking what democratic control requires at each and what these allies can build.
We are told AI is coming whether we like it or not. That story leaves people out, as though they had no say. So let’s put them back. If you love AI, we’re not here to crash the party. But as Arundhati Roy warns, “Colonisation always comes bearing shiny gifts.” The question is who holds the steering wheel, and who’s just along for the ride.
Our answer: the people whose labor makes AI possible and who live with its consequences. We call this Solidarity AI, not Cooperative AI, because cooperatives cannot do this alone. It is built by institutions where capital does not have the final say, and it includes the right to refuse.
What worries us
By AI, we mean an ecosystem of models, data, data centers, and human labor, from the annotator training a system to the driver whose livelihood it governs. A few companies control these systems; workers and affected communities have little say. Their growing use in surveillance and war makes that concentration of power more dangerous. AI companies burn through capital while the firms supplying minerals, chips, cloud computing, and energy make money selling the picks and shovels of the AI gold rush.
Workers need collective power to bargain over algorithms and govern the platforms that organize their work. The Majority World supplies AI’s materials, labor, and water while capturing little of its value: cobalt from Congolese mines, low-paid data work linked to lasting psychological harm in Kenya and the Philippines, and water consumed by data centers in Querétaro and Santiago.
Who gets to say no? Those who live with the consequences must be free to refuse, limit, or reverse AI’s deployment in their workplaces, industries, and neighborhoods.
Even the Pope wants these decisions open to challenge. In his encyclical Magnifica Humanitas (2026, §164), Pope Leo XIV insists that algorithmic decisions be “understandable, contestable and subject to oversight, so that individuals are not reduced to mere profiles.”
But moral appeals don’t enforce themselves. You can’t optimism your way out of a database that already has your name in it. That takes regulation, and governments with both the will and the capacity to act.
In June 2026, the International Labour Organization adopted the Decent Work in the Platform Economy Convention, 2026 (No. 193), the first global standard specifically for platform workers. It addresses algorithmic management through transparency and human review of automated decisions, and protects workers against unlawful account deactivation. This was a real victory for platform worker organizing, but its impact depends on national ratification and enforcement. The ILO cannot enforce its conventions directly. A decade after its landmark Domestic Workers Convention, for instance, more than a third of domestic workers still had no labor law protection at all.
And time is short. As Janjira Sombatpoonsiri argues, Silicon Valley and the far right have struck a pact for a collapsing world where AI companies are becoming mighty enough to seize the state and hand it the machinery of surveillance and deportation. What takes shape is not only an extraction stack but an authoritarian one, secured through military procurement, sanctions regimes, and the tightening alliance between firms like Palantir and OpenAI and state security institutions.
Working people cannot wait a lifetime for laws to improve their lives. They have the right and the need to build power and institutions of their own. Cooperatives and unions must become anti-fascist networks of solidarity that make democracy an everyday practice of deciding together and holding one another accountable. They must also deliver dignified work, dependable incomes, affordable rent, and more renewable energy; bad news for merchants of fear whose promises depend on economic insecurity and powerlessness. When the AI bubble bursts, cooperatives could jointly hire tech workers to run independent, affordable AI services rooted in shared values.
What role should AI play, and where should it have no place? Its roadmap belongs to the night-shift nurse, the delivery rider, and the kid on a cracked phone. Could it help a care cooperative plan fair schedules, farmers track a changing climate, or a minority language survive online? Could AI agents owned and governed in common help people pursue these shared purposes?
What is Solidarity AI?
When we first set out to explore what cooperatives could do with AI, we learned that cooperatives alone are not enough. Running AI democratically also takes municipalities, unions, social movements, public institutions, and alliances across borders. Solidarity AI is AI built, owned, and governed by the people who do its work and live with its consequences, including their right to say no. Its builders are institutions where capital does not hold ultimate control: worker and multi-stakeholder cooperatives, democratically governed nonprofits, democratic ESOPs, solidaristic DAOs, and new partnerships between public institutions and cooperatives. It ties AI to the wider solidarity economy and asks how AI can serve people’s needs such as a livable planet, dignified work, affordable essentials, living languages and cultures.
It draws on Asian and majority-world traditions, from Buddhist interdependence and Thailand’s sufficiency economy to Gandhian decentralization and feminist, queer, and decolonial critique, that have long pushed back against extraction and the belief that every social problem has a technical fix. Buddhist interdependence asks us to count the costs imposed on exploited data workers, mining communities, and people living near water-intensive data centers. The Thai philosopher Soraj Hongladarom, in The Ethics of AI and Robotics: A Buddhist Viewpoint (2020), argues that AI should be designed and governed around compassion and the reduction of suffering, not efficiency or profit alone. His idea of “machine enlightenment” holds that technical capability is not enough: AI should be judged by its ethical character and its effects on human flourishing. Thailand’s sufficiency economy favors using only as much computing power as a task requires, rather than treating scale as progress. Gandhian decentralization calls for giving affected communities the power to decide whether to adopt an AI system, control the data it collects, and end its use.
The solidarity stack makes this real by treating AI as an ecosystem of interdependent resources, institutions, and people.
Solidarity Stacks
AI is not one thing. It is a “stack.” The land, water, energy, and computing that power it; the data that feeds it; the models built on that data; and the human labor that runs through every layer. Today a few companies control the whole stack. A solidarity stack puts each layer under democratic ownership and governance, so that workers and communities have power over the systems they depend on, from how they are built to what they are used for. We introduced this approach in “Building a Solidarity Ecosystem for AI.”
Parts of a solidarity stack are already being built. Communities, universities, cooperatives, city governments, and public agencies build them from the ground up, often for jobs too small or too unprofitable for the big companies to bother with. The idea is not new. In the 1910s to 1930s, wholesale networks linked hundreds of local co-ops through shared purchasing and pooled capital, an ambition called the cooperative commonwealth. Today’s initiatives flicker like fireflies, scattered points of light that have yet to find one another. The question is whether they stay isolated or can be assembled into a system strong enough to hold.
Infrastructure. Communities must have power over the use of their land, water, and energy, including the right to refuse computing infrastructure. Shared ecological limits become part of procurement and everyday decisions.
No cooperative can run this layer alone. Credit unions, worker cooperatives, unions, universities, and public agencies can pool finance, computing, and security expertise, keep reliable backups, and offer mutual support across borders, so that services survive when individual projects face disruption. Energy and connectivity cooperatives already exist, from renewable energy co-ops across Europe and the US to community networks like Guifi.net. Cities and states can support them through preferential procurement. Argentine technology cooperatives Farox, Cambá, and Nayra are developing GenIA, an open-source suite designed to help organizations deploy and manage AI with greater control over their data and infrastructure. In Greater Manchester, Tameside council turned a former bathhouse into the Ashton Data Centre, whose servers support the council, the local NHS trust, cooperatives, and local firms. It is a small, greener alternative to hyperscale cloud, with costs and governance shared among public institutions and cooperatives. Vietnam is pushing to build its own AI stack. But national sovereignty is not the same as democratic control; a state-owned stack can serve surveillance as easily as it serves people. The open question is who should own and govern the physical base of AI, and who gets to say no to it.
Data. Workers who build, maintain, and are managed by AI should help decide what data it collects about them and how that data is used. Communities should govern their own data, with room for minority languages, local knowledge, and dissent. Farmer-led data cooperatives in India show how this can work. So does Origins Co-op, a regenerative rubber farming network in Thailand whose members own and govern their farm data. Farox, the Argentine tech co-op, has developed GenWay, a tool that masks sensitive data before requests reach external AI models. In Spain, Mondragon has piloted a data-sovereign pipeline that shares factory sensor data with outside analysts while setting enforceable rules for how it is used. The open question is how that control can work in practice, through data cooperatives, community language data, and worker data rights.
Models. Scale and pace are choices too. AI hype makes decisions about investment, work, and public resources seem inevitable; solidarity means building the collective power to challenge those decisions. Smaller, slower, locally governed AI may serve a community better. Sometimes AI can help build a simpler tool that people then maintain without continued dependence on it. Sometimes no AI is needed. The question is what serves a shared purpose. As the Decentralization Research Center’s report Alternatives to Concentrated AI argues, alternatives need not match the largest models to be valuable: systems built for particular languages, places, and community needs can succeed on their own terms. Chip manufacturing and frontier-scale training nevertheless remain concentrated.
Public and cooperative models already exist. In Switzerland, the federal government and public universities released Apertus, a fully open language model trained in more than a thousand languages. In Singapore, a public research program built SEA-LION so that AI understands Thai, Bahasa, Vietnamese, and other Southeast Asian languages, and multilingual systems are emerging in Indonesia and Malaysia. InkubaLM does similar work for African languages on modest computing resources. Rather than a race between nation-states to build “sovereign AI,” these projects point to a distributed ecosystem of smaller, interoperable models. Public LLMs focused on local languages are one example of how alternative AI systems could help communities sustain their languages and decide how their cultures are represented. Transkribus, an AI co-op, is used by libraries around the world. In Toronto, Hypha Worker Co-operative builds with open-weight models that run locally; its RooLLM is a co-operative AI assistant that reviews code, checks vacation sheets, and analyzes consensus in chat.
Bread Cooperative, a worker-owned “crypto leftist” collective that builds open-source financial tools for mutual aid, also builds AI agents and messaging bots through messaging platforms such as Signal for cooperatives and solidarity groups. Movement collections held by HEC Montréal, Desjardins, the Kerala State Cooperative Union, the Co-operative University of Kenya, and the Organization of Brazilian Cooperatives could together form the training corpus for a cooperative language model rooted in movement knowledge. Together these projects show what small, open, and community-run models already do, and where they fall short.
Labor. Workers should help decide how AI organizes their work, and when to refuse or reverse its use. That commitment must extend across the whole supply chain. A cooperative’s gains cannot rest on exploited data labelers or communities left to bear the costs of extraction. Where unions are absent, data labelers can organize as cooperatives to bargain collectively and arrange insurance or savings schemes. In Kenya, content moderators, data labelers, and engineers formed the Gamayyar African Tech Workers Cooperative to replace extractive outsourcing with work they own and govern.
The seven cooperative principles provide a democratic foundation, but cooperative ownership alone cannot guarantee fair conditions for workers elsewhere in an AI supply chain. A cooperative can still depend on models trained by underpaid data workers or contractors with little power to challenge their treatment. Fairwork’s new certification for AI supply chains offers independent scrutiny of pay, conditions, contracts, management, and representation, drawing on worker interviews as well as documentary evidence. Together, these approaches can help ensure that a cooperative’s democratic commitments reach everyone whose labor makes its services possible.
Platform cooperatives have been answering this question for a decade: what happens when workers own the algorithm that manages them? In Mexico City, delivery workers formed SIRApps, a cooperative running on shared CoopCycle software. In Thailand, the Thai Therapy Spa at Home collective books and runs its own home-massage service. Riders, annotators, care workers, and moderators belong at the center of this discussion, as protagonists, not solely case studies.
The Politics and Practice of Building Solidarity
Thinking in stacks makes the plumbing of power visible: who supplies the infrastructure, who does the work, who decides, and who gets the benefits and carries the costs. Answering these questions takes durable institutions with clear responsibilities, accountable leadership, financial oversight, ways to resolve conflict, and protection of shared assets for future members. CoRE Stack, a climate-resilience platform in rural India, shows what that means. Five hundred landscape stewards are turning it into a cooperative they run themselves. It takes capital, expertise, and customers. Universities and public agencies can help by committing part of their technology budgets to cooperatively owned providers; people whose work, data, taxes, and public infrastructure sustain AI deserve a say in its direction and a share in its benefits. It also takes alliances across a cooperative movement that too often works in isolation, and with feminist, climate, and digital rights movements beyond it. Together they can bargain with suppliers and commission tools that meet shared needs. Practical partners could include REScoop.eu, Patio, the Global Indigenous Data Alliance, UNI Global Union, and the International Alliance of App-Based Transport Workers.
In Brazil, UniTI demonstrates one practical form of cooperation: 22 cooperatives in Paraná pool purchases of hardware, software, and licenses to lower individual costs, alongside shared work on infrastructure and cybersecurity. The same logic could extend to employment. Cooperatives could jointly hire developers, system administrators, and support staff to run AI services that none could sustain alone.
Cooperatives and other organizations can also start with an AI policy that sets standards for which systems they buy or use, how data are protected, what suppliers must disclose, how environmental harm is limited, and how workers retain control over important decisions.
Across communities, federations, sectors, and borders, cooperatives can build AI systems and knowledge commons together by selectively sharing authorized knowledge and dividing responsibilities for data stewardship, technical development, infrastructure, and deployment, while retaining control over what they share. And it takes radical pedagogies. Worker-led research that sets the questions and judges the results, with time and resources to experiment and build the capacity to govern. Practical responses range from collective bargaining and shared infrastructure to worker buyouts and conversions to cooperative ownership, and at the far end, expropriation. James Muldoon once asked what it would mean to expropriate Google. With AI, the target would no longer be one company. It would be the stack. We build step by step, on small successes. But the horizon is the stack itself. Here, ideas turn into institutions.
Stacks also need roots. Thailand’s cooperative movement goes back to 1916, when Prince Bidyalongkorn helped found the first cooperative at Wat Chan in Phitsanulok. Today the Cooperative League of Thailand represents more than 7,700 cooperatives with over eleven million members. Alongside them, solidarity-economy initiatives show the cooperative idea in everyday life: SAFETist Farm, an urban farming collective promoting food security and environmental learning; Zero Baht Plus Shop, where residents exchange sorted waste for goods; Dignity Returns, a worker-recovered garment factory, run by the workers it once laid off; TechTransThai, a community that builds free software and public-interest digital projects; and, from Indonesia, Aqsa Farm, a cooperative strengthening local food systems.
What it adds up to. Run democratically, the stack should deliver dignified work, dependable incomes, affordable essentials, and a collective say over technological change. Its wider result is a Solidarity AI commons: models, data, tools, and records of workers’ and communities’ resistance that they create, maintain, use, and contribute to under rules they shape together. Some will be open to everyone, others will belong to their members. These institutions build on people’s capacity for mutual care and help sustain it.
The point is to turn these ideas into practice. Across every layer of the stack, people building alternatives need to find collaborators, test proposals, and make commitments they can carry out together: what will we share, what will we build or refuse, who will fund it, and who will take responsibility?
How we got here
Solidarity AI didn’t come out of nowhere. It grows from more than a dozen annual conferences organized by the Platform Cooperativism Consortium (PCC) at The New School that have rallied the global platform co-op community in New York City, Hong Kong, Berlin, Rio de Janeiro, Thiruvananthapuram, Mombasa, and Istanbul.
From exploring collective ownership and democratic governance of AI at PCC’s 2025 global Cooperative AI conference in Istanbul, to research exposing the hidden labor and inequalities of AI supply chains at INDL in Geneva; from rank-and-file tech worker organizing at Circuit Breakers in New York, to challenging the growing entanglement of state power and corporate data systems at the Data Justice Conference in Wales; from radical, anti-fascist pedagogies and building a university for the people through Nova Schola in Italy, to developing a shared roadmap for cooperatively governed AI at the Cooperative AI Bootcamp in Rome; from resisting data colonialism through Tierra Común in Latin America to strengthening regional digital-rights networks at the Digital Rights Asia-Pacific Assembly in Manila, scattered allies are connecting, learning from one another, and building alternatives. They share a stubborn conviction. Surveillance capitalists shouldn’t get to decide where AI goes next. We do.
What comes next
In 2025, young people across Asia took to the streets against leaders who govern for themselves. In Nepal, they brought down a government. In Jakarta, the protests grew after police ran over Affan Kurniawan, a 21-year-old delivery rider working for an app. The future of work is already being fought over in the streets.
As Arundhati Roy writes of the men who run the AI industry, “This entitlement, this wealth, this power, this presumption, was not given to them by God. It has been seized from us.” What was seized can be reclaimed, and governed together.
We were told AI is coming whether we like it or not. That was never a forecast. It was capitalist fiction. A demand to step aside. We won’t. To the night-shift nurse, the delivery rider, the annotator, and the kid on a cracked phone. What will you share? What will you build? And what will you refuse? Come with your questions, your stubbornness, and someone who disagrees with you. Solidarity isn’t agreement.
–
R. Trebor Scholz is a professor at The New School and a faculty affiliate at Harvard University’s Berkman Klein Center for Internet & Society whose research and organizing advance worker ownership and democratic control of digital platforms and AI; he is the author of Uber-Worked and Underpaid (Polity) and Own This! (Verso), and co-convener of Solidarity AI at Chulalongkorn University November 12-15, 2026.
