We used to own our tools; we now rent our means of thinking. Software, music, films and artificial intelligences are used by subscription, on distant machines that belong to others. I propose to read this situation across two long spans of time, that of the centralisation cycles of computing and that, far older, of the ownership of land, from medieval tenures to the enclosures and on to lands taken back. Libraries, whose rights are attached to the physical media they lend, are already living through this shift, and new machines now make domestic computation possible, though it remains reserved for specialists. The stakes, which I propose to name with an ancient figure, computational land, are therefore doubled by a question about the future: democratising the ownership of one’s means of computation.
From the egg beater to the subscription
An object bought in the twentieth century belonged to the person who bought it: you could repair it, modify it, lend it, resell it. The digital industries have silently undone this regime. We no longer buy a record but access to a catalogue, no longer a piece of software but a revocable licence, no longer an encyclopedia but a subscription, and the artificial intelligence companies are carrying this logic to its conclusion: the tool of intellectual work is rented by the month, runs on the landlord’s machines, and can be modified, restricted or withdrawn without our consent. The ownership of use has been separated from the ownership of control, and it is this separation I want to examine: we cultivate fields whose soil, seeds and rules we do not own.
The cycle of the terminal
We have lived through this moment before. The Minitel, whose history Valérie Schafer and Benjamin Thierry have traced (Le Minitel. L’enfance numérique de la France, Nuvis, 2012), was a terminal with no intelligence of its own: all the computing happened on central servers, and the little beige box did nothing but display. The personal computer represented a historic decentralisation: computation entered people’s homes, programs ran in their houses, and they owned them. Then the online service providers, and today the big artificial intelligence companies, have turned our powerful machines back into mere display terminals, a return to the Minitel that the activist Benjamin Bayart had anticipated as early as 2007 when he spoke of the “minitelisation of the Internet”. The biologist Olivier Hamant describes the same movement at another scale (Antidote au culte de la performance. La robustesse du vivant, Gallimard, “Tracts”, 2023): with high-performance globalisation, the world has shrunk through hypercentralisation, and this concentration, optimal as long as nothing fluctuates, becomes the weak point of everything as soon as something does. But these cycles are not inevitabilities. A counter-movement already exists, that of open models you can download and local machines able to run them, and the question is political before it is technical: where do we want computation to live?
The two economies of use
Olivier Hamant nevertheless argues for a shift from an economy of ownership to an economy of use, and this position seems to contradict mine. In his analysis, the individual ownership of fragile, disposable objects feeds overconsumption, while the shared use of repairable objects, the rented car, the tyre sold by the kilometre rather than as an object, frees us from the burden of possessing; he imagines the village of the future equipped, alongside its town hall and its village hall, with a citizens’ repair workshop and a fleet of shared cars. The contradiction with the digital subscription is only apparent, and dispelling it brings out the criterion that matters. In the economy of use that Hamant describes, the object remains locally repairable, if possible by the citizens themselves, know-how circulates, and control moves closer to the people who use. In the platform subscription, the object ceases to be repairable by anyone other than its owner, know-how concentrates, and control moves away. The two economies resemble each other, since in neither do we own, and they are opposed on the essential point, which is not the title of ownership but maintenance: the first distributes the capacity to look after the tool, the second confiscates it. The same word, use, thus covers two opposite political regimes, and to know which one you are in, it is enough to ask who has the right and the means to repair.
The library, testing ground of the two economies
This divide can be verified on the shelves of an institution older than any platform: the public library. Before any theory, the library is an economy of use in the full sense Hamant gives the phrase: a single object serves hundreds of people, an institution maintains it, repairs it, catalogues it, keeps it moving, and nobody needs to own it in order to live with it. This economy rests on a legal detail that is now taking on the value of a principle: the rights are attached to the physical medium. A DVD acquired with its lending rights belongs to the media library as the egg beater belonged to the household; the library lends it for as long as the object holds, resurfaces it when it gets scratched, gives it away or passes it on when weeding the collection. The shift to video on demand dissolves this foundation: the rights are no longer attached to an owned object but to a rented licence, and the library tips from one economy into the other.
In the article The future of audiovisual media in libraries, an undertaking that is just beginning I described the national project that has taken the measure of this shift. The figures it gathered measure the impasse: in 2023, 446 French libraries subscribed to a video-on-demand offer, for a total budget of 4.8 million euros and 1.3 million viewings, and since the media chronology reserves recent films for pay-per-view pricing, the more an offer finds its public, the more it costs the local authority. A public service is thus punished by its own success, a sign that the economic framework was not designed for it. Bundle offers, for their part, make librarians lose part of their control over the choice of titles and their mediation, which is to say precisely their profession. And meanwhile, on the terrain of the owned medium, libraries remain a decisive economic actor: in segments such as documentary, children’s films or auteur cinema, their acquisitions represent 30 to 50% of the volumes sold in physical video, and up to 70% of the revenue of certain titles. The lesson goes beyond the audiovisual field: when a public institution carries the economy of use, it keeps whole sectors of creation alive; when it is summoned to switch to the economy of the licence, its means do not follow and its function erodes. The avenues opened by the national project, asking the State to negotiate sustainable models with the platforms, allowing donations of physical media between local authorities and with individuals, creating shared territorial reserves, are attempts to re-establish, in the world of licences, the freedoms that ownership of the medium used to give without anyone thinking about it.
Dominium utile, dominium directum
The separation between ownership of use and ownership of control has a thousand-year history, and that history is the history of land. Medieval jurists, rereading Roman law from the twelfth century onwards, forged a distinction that remained famous: on one and the same piece of land coexist a dominium directum, the lord’s, and a dominium utile, that of the peasant family who cultivates it. The peasant tenure was often hereditary, transferable, almost a property; but it owed dues, rents and labour services, and the lord kept the last word. Around the tenures stretched the commons, lands of which nobody was the owner in the modern sense and where collective rights of use were exercised, common grazing for the herds after the harvest, wood-gathering rights, gleaning for the poorest. That world was no paradise, it rested on subjection; but its law recognised something we have unlearned: use constitutes a right, distinct from ownership and enforceable against it.
The industrial era was built on the destruction of that right. In England, landowners fenced off the commons, an enclosure movement begun in the sixteenth century for sheep and completed in the eighteenth and nineteenth centuries by thousands of acts voted in Parliament; the economist Karl Polanyi, who described in The Great Transformation (1944) the invention of land as a commodity, called this movement “a revolution of the rich against the poor”. Deprived of the commons that supplemented their subsistence, the peasants flowed towards the towns and the factories: Marx named this founding expropriation primitive accumulation, which supplied nascent industry with its workforce along with its land (Capital, Volume I, 1867). France took another path, the night of 4 August 1789 and then the complete abolition of feudal rights in 1793 turned tenants into owners; but the commons were whittled away there too, sold, divided, and the overall result is comparable: by the twentieth century, full and exclusive ownership, buyable and sellable, had become the normal form of the relationship to land, and collective rights of use a curiosity for the archives. I tell this story because we are replaying it: the digital platforms are reconstituting a dominium directum, that of the owner of the servers who sets the rules and collects the dues, and are leaving us a precarious, revocable dominium utile, without the heredity or the enforceability that medieval law, at least, granted to tenures.
Computational land
The figure of land therefore seems to me the most accurate name for what is at stake. For centuries, those who cultivated the fields did not own them, handed over the bulk of their production to the owner, and the conquest of the land by those who cultivated it was one of the great struggles of history, long, hard, never completely won. I call computational land the whole of the machines, models and rules on which intellectual work is now cultivated: we pour into it our work, our data, our writings, and the harvest goes up to the owners of the servers. Pierre Dardot and Christian Laval have shown that the alternative is not reducible to the choice between private property and State property: the common is a political principle in its own right, instituted by the activity of those who share a resource and give themselves its rules together (Commun. Essai sur la révolution au XXIe siècle, La Découverte, 2014). Free software, open models, pooled infrastructures are the already existing forms of this computational common, unknowing heirs of the commons that preceded the enclosures. The economist Elinor Ostrom, whose principles of self-organisation I presented in the article Leading Collective Intelligence in the Face of Crises, adds a condition that free software communities know from experience: commons last when an outside entity recognises them. The role of States is therefore not to build yet another national cloud; it is to recognise, fund and network the computational commons that exist, as they eventually came to guarantee the lands of those who cultivated them.
Lands taken back
This guarantee is not wishful thinking: land, precisely, has been taken back several times, and these experiences form a repertoire of available methods. On the Larzac plateau, the French State had announced in 1971 the extension of a military camp over thousands of hectares of farmland; ten years of non-violent struggle by the peasants and their allies led to the abandonment of the project in 1981, then to a lasting legal invention: the State, which remained the owner of the land it had bought, entrusted it in 1985 through a long-term lease to a civil society created by the peasants, the Société civile des terres du Larzac, which has since allocated it collectively to farming projects, outside any land market. Public ownership and collective use combine there instead of excluding each other, exactly along Ostrom’s lines. In 2003, activists of peasant farming founded Terre de Liens, a land trust fed by citizens’ savings which buys farms threatened by speculation or disappearance, withdraws them permanently from the market and leases them to farmers under contracts that commit them to soil-respecting practices: several hundred farms have thus been removed from the logic of the fictitious commodity Polanyi described. In Scotland, where the concentration of land ownership inherited from the enclosures remains among the highest in Europe, the inhabitants of the small island of Eigg bought their island back in 1997 from its private owner, through public subscription and within a community trust, and the Scottish land reform act of 2003 then gave every rural community a right of pre-emption over the lands that sustain it. These three experiences resemble each other on one point of method: none of these collectives waited for ownership to change hands by miracle; all of them built an institution of use, a civil society, a land trust, a community trust, which public authority eventually recognised and equipped. That is exactly the road still to be travelled for computational land.
The tractor and the server
This road opens at the precise moment when the material conditions are changing. Since 2025, NVIDIA has been selling a computer the size of a thick book, the DGX Spark, for around four thousand dollars, whose 128 gigabytes of unified memory run language models of the order of a hundred billion parameters locally; manufacturers such as ASUS, MSI, Dell and Lenovo offer their own versions, Apple sells desktop machines whose memory reaches 512 gigabytes, AMD equips comparable mini-computers. In the same movement, open models of high quality, those of Mistral in France, of Meta, of DeepSeek and others, can be downloaded freely. In other words, the counter-movement of the cycle of the terminal now has its hardware: a powerful artificial intelligence can run on a kitchen table, without subscription, without connection, without anyone being able to modify or withdraw it remotely. The irony deserves to be noted in passing, these machines of decentralisation come out of the most concentrated industry in the world, a few chip designers, a handful of foundries; domestic computational land remains sown with industrial seeds, as the most autonomous farm buys its tractor from a large manufacturer.
But the machine on the table is not enough, and this is where the question becomes political. Installing a model, choosing its size and quantisation, updating it, securing it, understanding what it does with our data: all of this remains today the work of highly specialised people, and ordinary mortals have no access to it. And here my own criterion turns against enthusiasm: I wrote above that the dividing line between the two economies of use runs through maintenance, through the answer to the question of who knows how to repair. A sovereign machine that I can neither install nor maintain does not take me out of the regime of dependency; it merely changes the dependency’s address, from the server landlord to the friend who works in IT. The ownership of one’s means of computation will only be democratic if the capacity to maintain them is distributed, and this distribution is a project for collective intelligence, not a technical problem.
The method already exists, and it comes once again from agriculture. In 1945, a tractor cost far more than a farm could pay, and its maintenance required skills nobody in the village had; farmers did not wait for everyone to become rich and a mechanic, they created the farm machinery cooperatives, the CUMA, instituted by a French ordinance of 1945, where machines are bought together, where their maintenance is pooled and where know-how is learned from farm to farm; thousands of these cooperatives are still running. Nothing forbids founding CUMA of computation: an association, a village, a network of cultural organisations buy a machine, install open models on it, and train each other to maintain it. The free software movement has been practising this transmission for thirty years in its install parties, where you come with your computer and leave with a free system installed and the skills to keep it going; the citizens’ repair workshop Hamant places at the centre of his village of the future is the same place, extended to computation. And public authority already has its relays: the digital advisers deployed across French territories since 2021, the public digital spaces, and first of all the media libraries. A media library that hosts a server of open models for its territory, and makes it an object of mediation as it made Internet access an object of mediation twenty-five years ago, is extending its age-old function exactly: giving access to what is too expensive for one person alone, and accompanying that access with human knowledge. For democratising does not mean turning everyone into a specialist: nobody asked everyone to become a printer in order to democratise reading; libraries were built instead.
Sovereignty as co-resilience
This detour through ownership joins the reflection on collective resilience I opened in the article Building Co-Resilience. A territory whose schools, care, administration and creation run entirely on machines rented from distant powers is fragile twice over: fragile because any single point of failure suspends it, and fragile because the loyalty of its tools resides elsewhere. I described a neighbouring mechanism at the scale of individuals in the article Mononormative Delegation: entrusting all one’s functions to a single system means espousing its fragility. Conversely, a share of computation owned locally, open models installed in schools, media libraries and town halls, distributed skills to maintain them, all of this constitutes an infrastructure of robustness, in the sense I specified in the article Robustness, antifragility, resilience, on the same footing as water reserves. Hamant describes the energy equivalent of this choice: rather than a few giant wind turbines whose breakdown suspends a territory, a multitude of small, removable, repairable machines, installed on dwellings along with the training that makes it possible to maintain them; the output is lower, the viability in times of turbulence is incomparable. The inhabitants of Eigg, once owners of their island, made this choice without waiting for the theory: since 2008, a small autonomous electricity grid, fed by water, wind and sun and maintained by the islanders, has replaced individual generators. Land taken back and distributed energy go together, and computation will follow the same road. To assess where a territory, an organisation or a person stands, I propose asking three questions:
- where does the computation live, that is, on which machines, in which country, under which law do the tools my activity depends on run;
- who can modify, restrict or withdraw these tools, and what recourse do I have if that happens;
- who, around me, knows how to install, repair and maintain an open equivalent of these tools.
The gestures that improve the answers are modest, I detailed them in the article Sovereign Digital, a Micropolitical Choice, and it is their modesty that makes them political: they are within the reach of local authorities as well as individuals, as were a subscription raised by islanders or a civil society of peasants.
