The Network Is Not the Owner: Co-ops, Peer-to-Peer and Who We're Ready to Be

I should declare a bias before I start. I grew up in regional Austria surrounded by agricultural co-operatives. The Lagerhaus sold the seed, the fertiliser and the diesel. The dairy co-op collected the milk. I’ve known them from the farm gate and, briefly, from inside the farmers’ league that runs alongside them. I’ve seen what they do well when their members actually run them, and I’ve seen what happens when governance gets comfortable and they start behaving like the corporations they were founded to keep at bay. They remain, perhaps sentimentally, a favourite of mine.

So when an online conversation about AI co-operatives recently ran into a familiar argument from the peer-to-peer side, I took it personally enough to think it through. The argument goes like this: co-ops always arrive second, trailing the corporations they react to, because they run on economic plumbing built for for-profit companies. Peer-to-peer systems build their own plumbing, so they sit outside both the corporate and the co-op world. On that reading, a co-op AI is just another flavour of centralised AI.

Parts of that argument are right, and they’re the useful parts. But it treats co-ops and peer-to-peer as two answers to the same question. They aren’t. One describes who owns a system, the other how it’s wired, and mixing them up is how so much of what gets sold as peer-to-peer turns out to be capitalism with distributed plumbing.

Credit Where It’s Due

The clearest example of a co-op AI right now is the Inference Cooperative, built by Metagov, the digital governance lab whose board includes Nathan Schneider. Schneider’s book Governable Spaces named the default for most online communities “implicit feudalism”: founders and admins rule as benevolent dictators because the software hands them the keys. The Inference Cooperative tries to fix that for AI access. One membership, from US$5 a month on a sliding scale, buys chat and API credits and a seat in governance. Even pricing is a member decision.

But the models still run in someone else’s data centre, and this is where the peer-to-peer critique lands squarely. The co-op governs access to a centralised stack. It doesn’t change the stack. Collective bargaining power and a democratic say over terms are worth having, but it is a co-op sitting on top of corporate infrastructure. The plumbing point holds too: co-ops struggle to raise capital in markets built around investor ownership, which is why Exit to Community, developed by Schneider’s Media Enterprise Design Lab with Zebras Unite, exists at all, as a way for startups to hand ownership to users and workers after conventional finance has got them off the ground.

Two Axes, Not One

The clearest map I know for this argument comes from Vasilis Kostakis and Michel Bauwens, the founder of the P2P Foundation. In Network Society and Future Scenarios for a Collaborative Economy (2014), they draw two axes. One runs from centralised to distributed control of the infrastructure. The other runs from accumulating capital to building up a shared commons. That gives four corners, not two.

The corner that matters here is the one they call distributed capitalism: distributed infrastructure, organised around capital accumulation. Here’s how they describe it:

“Any Commons is a by-product or afterthought of the system, and personal motivations are driven by exchange, trade and profit.”

They are specific about who lives in that corner: “the anarcho-capitalist/libertarian projects”, premised on “the elimination of the state in favor of individual sovereignty, private property and free/open markets”. Their slogan for the quadrant is blunt: “everyone can become an independent capitalist”. They also trace that corner’s design to the Austrian school of economics, which an Austrian who grew up among farm co-ops can only read with a wry smile.

That’s the libertarian version of peer-to-peer, and it has a track record. Bauwens’ own evaluation of Bitcoin calls its design “anarcho-capitalist, i.e. it is designed to favour the freedom of property owners, and the more you own, the freeer [sic] you are.” Bitcoin now sits in exchange-traded funds sold by the world’s largest asset managers. Peer-to-peer AI is following the same path. Bittensor, usually ranked the biggest decentralised AI network by market value, pays participants in its own token, and a 2025 analysis by Elizabeth Lui and Jiahao Sun found that its rewards are “overwhelmingly driven by stake”, with “a clear misalignment between quality and compensation”. The network is distributed. The rewards flow to whoever already holds the most. Regular readers will recognise the shape from Helium, which I covered in Cooperative by Design and again in The Decentralisation Paradox. Helium was never collaborative to begin with: it was corporate greed in a collaborative costume, and most DePIN (decentralised physical infrastructure) projects follow the same script.

So “peer-to-peer” on its own tells you how a network is wired. It tells you nothing about who owns what flows through it. A co-op answers the ownership question and is silent on wiring. Those are different axes, which means the claim that peer-to-peer escapes both corporations and co-ops doesn’t hold. Peer-to-peer escapes centralisation. Escaping capital accumulation takes a separate decision, and that decision is a governance one.

The Co-op in My Head and the Co-ops We Have

Fairness cuts both ways, and my own bias makes this part harder to write. Co-ops have a failure mode of their own, and it isn’t centralisation. It starts as complacency and ends as corporatisation.

First, the yardstick. The co-op in my head is the one the International Co-operative Alliance describes: “democratic organisations controlled by their members”, one member, one vote, and “autonomous, self-help organisations” that keep that control even when they deal with governments or outside capital. To that I’d add scale. A good co-op is small enough that members know each other, and it grows by multiplying and federating, not by swallowing. That’s the version I’ve spent the decades since Carinthia working towards.

Now the reality. Austria has the worst example I know, and I say that as an Austrian. The first Raiffeisenkasse opened in 1886 in Mühldorf, in the Wachau, so farmers could pool their savings instead of borrowing from moneylenders. Within a decade there were more than 600. Today the regional Raiffeisen banks that grew out of that movement own about 61% of Raiffeisen Bank International, a listed bank that has kept its Russian subsidiary operating through years of the war in Ukraine, over sustained protest from shareholders and activists. No village member voted for that, and no village member could have stopped it.

The corporate arm is only half of it. Austrian farmers are represented by law through the Landwirtschaftskammern, the statutory agricultural chambers, and in most states their elections are dominated by the ÖVP’s farmers’ league, the Bauernbund. In Salzburg in 2025 it took 73.3% of the vote on a turnout of 34.8%. In Lower Austria, the regional Raiffeisen bank has even had the right to send its own delegates into the chamber’s plenary. Raiffeisen, the chambers and the ÖVP have been entwined for generations: bank, chamber and party sit at the same table, and the Lagerhaus and the dairy co-op sit underneath them. That’s the autonomy principle turned inside out. For a small member, one vote at the annual meeting is a formality. The strings get pulled elsewhere. The co-op form survives on paper. Member control doesn’t.

Australia has its own version. Murray Goulburn was a dairy co-operative owned by its farmer suppliers. In 2016 it cut its farmgate milk price retrospectively and asked farmers to repay money for milk they had already delivered. The ACCC later took action over its pricing representations, and in 2018 its operating assets were sold to Saputo, a Canadian multinational. The co-op chased scale, partially listed on the stock exchange and ended up treating its own members like suppliers to be managed.

Size alone doesn’t decide it, though. CBH Group, the Western Australian grain handler and Australia’s largest co-operative, is enormous and still grower-controlled. In 2016 a consortium led by former CBH directors, with GrainCorp and First State Super as cornerstone investors, campaigned to corporatise it. Any such change needs a 75% vote of grower members. The board rejected the proposal unanimously, a grower survey found 78% backing that rejection, and the consortium withdrew. The protection was written into the rules before anyone needed it.

The lesson I take from all three: co-ops fail when members lose control of them, usually while chasing size or cosying up to power. The defence is governance, not a different form: rules that make capture hard, members close enough to notice, and growth by multiplying at community scale. That’s a reason to build co-ops better, not to write them off.

Both at Once

Once you see two axes, the obvious move is to combine them. Bauwens and Kostakis made exactly that case in a companion paper the same year with a memorable paradox: “the more communist the sharing license used in the peer production of free software or open hardware, the more capitalist the practice.” Open code gets built by commoners and harvested by multinationals. Their answer was open cooperativism: peer production of a shared commons, paired with co-operative ownership of the businesses that make a living around it, so value flows back to the people producing it.

This already exists for AI, in small form. CoCore, which I wrote about in The AI Stack Needs a Commons Governor, is a compute co-operative where members pool hardware they already own to run AI inference for each other. That’s peer-to-peer wiring under co-operative ownership. It isn’t “co-op AI” or “p2p AI”. It’s both.

Nondominium, from the Montreal open value network Sensorica (which I wrote about in The Commons Has No Ledger for This), takes a similar line. It runs peer to peer on Holochain and speaks Valueflows, built for resources that are “organization-agnostic, uncapturable, and natively collaborative”. Each resource carries its own governance, “embedded rules, transparent peer validation”, so the rules travel with the thing rather than living in a platform’s terms of service. Sensorica’s ventures also have a custodian, a legal entity that “represents them legally and acts as a trust for their tangible assets”, and Nondominium is pitched at helping traditional organisations collaborate while “preserving their current structure”. Peer-to-peer wiring, with an institution standing behind it, connecting the organisations people already have rather than replacing them.

GrowGood is being built on the same bet. Its economic language, Valueflows, works without a central platform: independent farms, processors and markets recording their own economic events and coordinating through offers and needs. hREA runs that same vocabulary on Holochain, peer to peer. But the project sits under a not-for-profit foundation, and the governance framework we choose has to be free and open source, self-hostable and affordable for small organisations, because the protocol alone won’t stop the commons being captured.

Short on Track Record

Here’s where my bias and the evidence agree. Peer-to-peer is the more fundamental change of the two, and I can see its advantages. What I can’t find is a long list of battle-tested successes.

As wiring, peer-to-peer has real wins. BitTorrent has moved files at enormous scale since 2001. Bitcoin’s network has kept running since 2009. But one is a protocol with no ownership model at all, and the other sits squarely in the distributed-capitalism corner.

As a way of owning and governing shared economic life, the record is thin. Decentralised autonomous organisations (DAOs, which run their governance through tokens and smart contracts) were meant to be the proof. A 2022 Chainalysis study of ten major DAOs found that fewer than 1% of token holders controlled 90% of the voting power. The original DAO, in 2016, lost about 3.6 million ether to an attacker exploiting its code. The fix wasn’t the code. It was people: the Ethereum community hard-forked the chain to return the funds, and those who refused split off to form Ethereum Classic. When “code is law” broke, humans had to step in, and they couldn’t agree.

The great commons success stories of the internet era, Wikipedia and Linux, are peer production, and they are genuinely remarkable. But they don’t run peer to peer, and they don’t run on protocol alone. Wikipedia sits on servers operated by the Wikimedia Foundation, which took in about US$185 million in 2024. Linux has the Linux Foundation. The commons that last have institutions behind them.

My reading of why: a lot of peer-to-peer designs are excellent in theory and stall in practice because not enough people share the vision to sustain them. That’s an observation, not a proof, and I’d be glad to be shown a list of counter-examples. But it points at people, not protocols.

Are We Ready?

That brings me to the real disagreement, and it isn’t about the technology. I don’t think most people are ready for peer-to-peer self-governance yet.

Forty years of neoliberal economics have taught several generations that society is a market and they are its individual consumers. That training shows up in the research. In 1993, Robert Frank, Thomas Gilovich and Dennis Regan found that economics students defected more often in a one-shot cooperation game, and that students whose instructor leaned heavily on game theory and the prisoner’s dilemma gave more cynical answers to honesty dilemmas by the end of the semester. Some of that is self-selection (selfish people choose to study economics), and the authors say so, but they also found evidence the training itself shifts behaviour.

The cynical read is that humans are simply lazy and greedy, and no protocol will fix that. The evidence points somewhere more interesting. When Joseph Henrich and colleagues ran the same economic games across 15 small-scale societies, from foragers to herders, the textbook self-interested model “is not supported in any society studied.” Behaviour tracked how each society organised its economic life. Selfishness, on that evidence, is learned. That’s bad news for anyone hoping for a quick switch, because what took decades to teach takes decades to unteach. It’s good news for co-ops.

A peer-to-peer commons asks everyone to govern themselves well, with nobody holding the keys. That’s a big ask of people who have never been trusted to govern anything. A co-op is where people practise that: one member, one vote, real budgets, real disagreements, a known way to resolve them. Elinor Ostrom spent a career showing that communities can govern shared resources well, and nearly every case she documented had rules, monitoring and a forum where people could argue. Co-ops are how we build that habit, not by growing into giants but by multiplying at community scale, where members still know each other.

Who Gets Left Behind

There’s a second problem with peer-to-peer, and it’s easy to miss when you’re fluent in the technology yourself. A peer-to-peer system quietly assumes every participant can take care of themselves: run their own node, guard their own keys, read the governance proposal, understand what they’re voting on. Kostakis and Bauwens call this assumption equipotentiality, the idea that anyone can participate on equal terms. Then they report what actually happens in the distributed-capitalism corner:

“In theory you have equipotential individuals (i.e., everyone can potentially participate in a project), but in practice what one gets is concentrated capital and centralized governance. Moreover, we see the emergence of oligarchies and aristocracies.”

Equal access on paper becomes rule by the technically fluent in practice. And self-sovereignty cuts both ways. If you lose your keys, nobody is coming to help, because helping you would mean someone having the power to override the system, which is the very thing it was built to prevent.

Now look at who that leaves out. The 2025 Australian Digital Inclusion Index found that one in five Australians (20.6%) are digitally excluded or highly excluded. For people aged 75 and over, the figure is 66.5%. For First Nations Australians, it’s 40.9%, and the gap widens sharply in remote areas. Those aren’t edge cases a better interface will fix. They’re a fifth of the country, and a large share of the farming communities I work with. A system that requires deep technical skill to participate fully hands power to those who have it, and the libertarian framing treats everyone else’s exclusion as their own problem.

This is where co-ops earn their keep, at least the ones I’d point to as examples. Caring for members who can’t do everything themselves is written into what a co-op is. The International Co-operative Alliance’s principles commit co-ops to “education, training and information” for members, so they “can contribute effectively”, and to “concern for community”. One member, one vote, whether or not you understand the database schema. A co-op can carry the technical complexity on its members’ behalf (someone runs the server, someone answers the phone when a password goes missing) while members keep the say over how it’s run. Not every co-op lives up to it, as the Austrian example shows, but the commitment is there to hold them to. A peer-to-peer protocol has no equivalent on its own, because there’s nobody in the middle to make the commitment. That obligation has to come from the people and institutions around it.

There’s an irony here, too. The ideal participant in libertarian peer-to-peer, a capable, well-resourced individual who looks after themselves, is pretty much the same person neoliberal economics assumed all along.

Where the Libertarians Fit

Peer-to-peer has a natural audience among libertarians, and that’s no accident: distrust of concentrated power, whether a state’s, a platform’s or a bank’s, is a shared starting point. But there are two libertarianisms in the room, split along exactly the axis peer-to-peer can’t settle on its own. One, represented by the Center for a Stateless Society and its best-known writer Kevin Carson, wants markets free of the state and of accumulated capital, and argues that peer production undermines capitalism rather than extending it. The other, which Bauwens calls propertarian, wants markets free of the state so property can do as it pleases. Both will happily run a node. Only one is building a commons.

Don’t Write Off the Co-op

So the useful question for any peer-to-peer project, AI or otherwise, isn’t “is it decentralised?” It’s Ostrom’s: who sets the rules, who benefits when they change, and what happens to the surplus. A peer-to-peer system with a good answer to that is a future I’d like to help build. Without one, it’s distributed capitalism, and we’ve already watched how that ends.

In the meantime, the co-op isn’t the thing to route around. The legal form is under two centuries old, but what it formalises is far older: neighbours agreeing on rules for shared things and holding each other to them, which I traced back through Alpine pastures and irrigation channels in The Commons Before the Combine. It has failed, often, and almost always in the same way, by forgetting its members. It is an ancient and deeply human form, built for people as they are, not as a protocol would like them to be.

Peer-to-peer is young. Some of its best ideas will prove themselves, and the projects most likely to get there are the ones pairing distributed wiring with co-operative ownership. Until it has a few more decades of battle scars to show, I’m not writing off the co-op. I’d rather put the code in the commons and the people in a co-op.

Sources

• Kostakis, V. & Bauwens, M. (2014). Network Society and Future Scenarios for a Collaborative Economy. Palgrave Macmillan. DOI: 10.1057/9781137406897. Working draft (PDF). The four scenarios, the “distributed capitalism” quadrant and its links to the Austrian school.
• Bauwens, M. & Kostakis, V. (2014). From the Communism of Capital to Capital for the Commons: Towards an Open Co-operativism. tripleC, 12(1). The “more communist the licence, the more capitalist the practice” paradox.
• Bauwens, M. (2014). Comprehensive Evaluation of Bitcoin. P2P Foundation Wiki. The “propertarian libertarians” critique.
• Schneider, N. (2024). Governable Spaces: Democratic Design for Online Life. University of California Press. “Implicit feudalism”.
• Inference Cooperative (2026). inference.coop. Made by Metagov.
• Sensorica. Nondominium. Venture page describing Nondominium’s Holochain and Valueflows architecture, embedded governance and custodian model.
• Lui, E. & Sun, J. (2025). Bittensor Protocol: The Bitcoin in Decentralized Artificial Intelligence? A Critical and Empirical Analysis. arXiv:2507.02951.
• Raiffeisen Bank International. History and Shareholder structure. The 1886 Mühldorf founding and the regional banks’ 61.17% stake.
• B4Ukraine (2025). Raiffeisen AGM 2025. Shareholder and activist protest over RBI’s continued Russian operations.
• Salzburg24 (2025). ÖVP-Bauernbund klarer Sieger von Landwirtschaftskammer-Wahl in Salzburg. The 2025 Salzburg chamber election result and turnout.
• Land Niederösterreich (2005). Wahl der NÖ Landwirtschaftskammer. Raiffeisen-Landesbank Niederösterreich-Wien’s four delegates to the chamber’s plenary.
• ACCC. Former Murray Goulburn MD Gary Helou to pay $200,000 penalty. See also the Senate Economics References Committee, Retrospective step-downs in 2015–16 milk pricing.
• Countryman (2016). Grains champion concedes defeat. The withdrawn Australian Grains Champion bid to corporatise CBH, the 75% vote threshold and the 78% grower survey.
• Chainalysis (2022). Dissecting the DAO: Web3 Ownership is Surprisingly Concentrated.
• Wikimedia Foundation. Form 990 filings via ProPublica. FY2024 revenue.
• Frank, R. H., Gilovich, T. & Regan, D. T. (1993). Does Studying Economics Inhibit Cooperation? Journal of Economic Perspectives, 7(2), 159–171.
• Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H. & McElreath, R. (2001). In Search of Homo Economicus: Behavioral Experiments in 15 Small-Scale Societies. American Economic Review, 91(2), 73–78.
• Australian Digital Inclusion Index (2025). The 2025 findings. RMIT University, Telstra and the ARC Centre of Excellence for Automated Decision-Making and Society.
• International Co-operative Alliance. Statement on the Co-operative Identity. The seven co-operative principles, including democratic member control (second), autonomy and independence (fourth), education, training and information (fifth) and concern for community (seventh).
• Carson, K. (2010). The Homebrew Industrial Revolution: A Low-Overhead Manifesto. Based on research papers written for the Center for a Stateless Society. Peer production and low-overhead manufacturing as an alternative economy “preparing to supplant” corporate capitalism.
• Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press.

• Peer-to-Peer
• Cooperatives
• Open Cooperativism
• Commons Governance
• Michel Bauwens
• Raiffeisen
• DAO Governance
• AI Governance