The familiar story about mesh and peer-to-peer networking is a story about radios. When central infrastructure fails, or when authorities block gateways, devices can relay traffic directly. The appeal is obvious: no single switch, no single operator, no single point of legal or technical failure. Yet by mid-2026 the more interesting question is no longer whether packets can find alternative paths. It is whether the surrounding layers of the system can remain decentralised once a network becomes socially useful, economically valuable and politically contested.
That distinction matters because transport resilience is the easiest part to romanticise and the hardest part to operationalise at population scale. Radios, local relays and distributed routing can keep bits moving. They do not by themselves solve naming, trust, moderation, software maintenance, abuse management, fraud, payment, or liability. In practice, these adjacent functions often drift back towards a handful of developers, directory operators, app stores, cloud relays or payment intermediaries. A network can be decentralised in topology while becoming centralised in the layers that decide who is recognised, who is updated, and who is excluded.
Resilience has moved up the stack
This is not a new problem so much as a neglected one. David Clark and colleagues described the internet as a site of constant institutional conflict in The Tussle in Cyberspace, arguing that technical architecture and social power are inseparable. That observation lands differently in decentralised systems. The old internet could tolerate considerable concentration in service provision because the basic transport layer remained open and globally interoperable. A mesh or peer-to-peer environment inverts the emphasis: transport may be improvised, local and adaptive, while coordination, safety and usability become the scarce resources.
For that reason, routing resilience is only one layer of sovereign resilience. The decisive question is who can impose updates, revoke identities or meter access at scale. If those powers sit with a narrow group, the network may survive physical disruption but remain vulnerable to administrative capture. The vulnerability is subtler than a shutdown, but not necessarily less consequential.
Routing resilience is only one layer of sovereign resilience.
Naming is where many networks quietly recentralise
Most users do not interact with routes; they interact with names. They remember a contact handle, a group identifier, a community directory or a searchable address. In decentralised environments, naming is hard because the functions bundled into ordinary internet platforms are usually split apart. Discovery, authenticity and reputation must be recreated without a universal operator. The temptation is to solve this with a lightweight directory, a popular resolver, or a small set of highly trusted maintainers. That works, until it becomes the real centre of gravity.
Most decentralised systems recentralise first in naming, not transport. A few registries or lookup services begin to shape discoverability. A few identity providers become de facto authorities for access. A few repositories become the accepted source of software truth. Once that happens, the network may still look distributed on a map, but power has consolidated around the indexes and trust anchors that make it legible to humans.
The analogy with the web is imperfect but revealing. The OECD has noted that decentralised web architectures redistribute functions across layers rather than eliminating intermediation altogether. Naming and discovery are among the functions that have proved especially sticky. In short, the topology can be plural while the user experience depends on hidden chokepoints.
Identity is the real bottleneck
Routing resilience is only one layer of sovereign resilience.
Identity in a mesh setting is more than authentication. It is a bundle of persistence, reputation and recoverability. A device key can prove continuity, but people lose devices, rotate credentials, share hardware and cross borders. Human trust does not map neatly to cryptographic permanence. Moreover, systems that make identity too cheap are vulnerable to spam, coercion and Sybil attacks, while systems that make identity too expensive often default to state documents, telecom credentials or platform logins. Either path undermines the political promise of decentralisation.
This tension increasingly mirrors broader digital policy debates. Zero trust principles, as set out by NIST, emphasise continuous verification rather than inherited trust from network location. That logic is useful for decentralised systems, but it also shifts operational burden onto identity and policy engines. In a mesh environment, if those engines become standardised through a narrow governance process, then security practice can end up reintroducing central control by another route.
Even privacy engineering adds complexity. IPv6 temporary addressing, standardised by the IETF, reduces straightforward device tracking on conventional networks. In peer-to-peer systems, however, continuity and reachability often require some stable identifiers. The very properties that make a node easy to find can make a person easier to profile.
The edge is expensive to maintain
There is also a stubborn economic problem. Centralised networks externalise maintenance onto large operators with predictable budgets and staffing. Decentralised ones distribute maintenance to volunteers, local administrators, community groups and small device owners. That can widen autonomy, but it can also diffuse responsibility until no one is clearly accountable for patching, replacing failing nodes, paying for backhaul or resolving disputes.
The history of resilient overlay networks showed that traffic can be rerouted around failures surprisingly well. What proved harder was sustaining the operational discipline around the system. Mesh deployments face the same reality. Firmware ages. Batteries degrade. Local champions move on. Security advisories require rapid uptake. The infrastructure that cannot be switched off by any single actor can still decay because too many actors each have only a weak incentive to maintain it.
That is one reason the most durable decentralised deployments are often not the most ideologically pure. They are the ones with a boring maintenance model: clear stewardship, funded support, regular key rotation, documented upgrade paths and a way to retire compromised nodes without a constitutional crisis.
Abuse management does not disappear when the centre does
A recurring misconception is that decentralisation removes the need for moderation or abuse handling. In practice it redistributes those tasks and makes them more granular. Spam, harassment, malware propagation and fraud travel efficiently across low-friction peer networks. If a decentralised system lacks ways to quarantine malicious peers, revoke compromised credentials or rate-limit abusive traffic, ordinary users tend to retreat to gated enclaves run by trusted curators. The network survives, but the social graph recentralises into islands of delegated authority.
This is where legal pressure enters. Regulators do not need to control the packet layer to influence behaviour. They can target app distribution, endpoint operators, payment processors, hosting providers for auxiliary services, or identifiable maintainers. Recent reporting in the Financial Times on European digital regulation and decentralised social media captures the broader pattern: distributed systems still meet concentrated legal obligations somewhere. The resulting compromise is often selective centralisation in compliance functions.
Most decentralised systems recentralise first in naming, not transport.
Interoperability can create its own elite
Most decentralised systems recentralise first in naming, not transport.
Standards are usually presented as the antidote to lock-in, and often they are. Yet mature interoperability can also produce a technical elite that controls the evolution of the protocol, the reference implementations and the release cadence. The internet has long depended on this kind of stewardship. The issue in decentralised networking is not whether stewardship exists but whether it can be contested, diversified and audited without splintering the ecosystem.
Messaging Layer Security, now standardised by the IETF, illustrates the point at a high level. Shared protocols can improve security and portability across implementations. But if only a handful of teams can correctly implement a complex standard, real power remains concentrated in those teams. Complexity itself can become a centralising force. A network that anyone may join is not necessarily a network anyone can meaningfully govern.
Localism helps, but only up to a point
Some of the most serious work in decentralised networking has shifted towards local-first assumptions: neighbourhood meshes, community backhaul, device-to-device messaging and opportunistic sync rather than a fantasy of replacing the global internet wholesale. This narrower scope is sensible. It reduces the burden of universal discovery, global moderation and transcontinental performance. It aligns architecture with actual failure modes, whether they involve disaster response, conflict disruption or chronic underinvestment in conventional access infrastructure.
Yet localism has a ceiling. A local network still needs bridges to external systems for software updates, identity recovery, security advisories, payment settlement or intercommunity federation. Each bridge is a governance question disguised as a technical one. The more successful the local network becomes, the more pressure there is to formalise those bridges. Formalisation, in turn, tends to create recognised intermediaries.
The result is not failure. It is a reminder that decentralisation is better understood as a distribution of bargaining power than as the absence of institutions.
Emergency use is not the same as civic use
Mesh systems are often justified by their value in emergencies, and that value is real. But a network designed for short bursts of crisis communication does not automatically make a good substrate for daily civic life. Emergency use tolerates rough edges: intermittent identity, reduced privacy, manual trust decisions, informal administration. Long-term civic use requires dispute resolution, accessibility, data portability, multilingual support, safeguards for vulnerable groups and dependable governance.
This distinction has become sharper as governments and international bodies frame digital resilience more explicitly. The UN’s Global Digital Compact and outcomes around internet governance stress inclusion, human rights and accountability alongside resilience. Those norms are easy to endorse and hard to encode in systems without central authorities. A decentralised network that works under duress may still struggle to support ordinary public life unless it builds institutions that are durable, legible and contestable.
Health and welfare cases reveal the hidden trade-offs
One underexplored angle is the use of peer and mesh architectures in health and welfare settings, where intermittent connectivity and local data exchange can be genuinely useful. Here the challenge is not merely throughput or latency. It is validation, provenance and duty of care. Medical and welfare information cannot be treated like casual chat traffic. Local exchange may preserve continuity during outages, but it raises difficult questions about consent, record integrity and synchronisation when systems reconnect.
Research in digital medicine repeatedly shows how small validation errors can have outsized human consequences. The lesson is broader than any single device class. Decentralised communication is not automatically emancipatory in high-stakes settings; it can shift risk onto users and local administrators who have the least capacity to manage it. In such domains, the strongest argument for decentralisation is often continuity of service, not libertarian autonomy.
The decisive question is who can impose updates, revoke identities or meter access at scale.
The geopolitics are less about shutdowns than dependency
For states and municipalities, the attraction of decentralised network architecture increasingly lies in strategic autonomy. The fear is not only that an adversary might switch off connectivity, but that key digital functions depend on remote operators, foreign standards processes, or legal jurisdictions beyond local control. Mesh and peer designs offer one route to reducing certain dependencies. But they do not abolish dependency; they rearrange it.
Dependency moves to chips, radios, cryptographic libraries, mobile operating systems, certificate practices, update channels and the governance of protocol changes. WSIS and wider digital cooperation debates have made this plain. Sovereignty in digital infrastructure is rarely achieved by topology alone. It is negotiated through supply chains, institutional design and the ability to replace components without losing the whole system.
That is why the simplistic contrast between centralised and decentralised internet models is now less useful than a more granular question: which layers are decentralised for whom, under which failure assumptions, and with what rights of exit.
The mature design principle is constrained recentralisation
If the first decade of enthusiasm treated decentralisation as a moral good in itself, the emerging lesson of the second is more restrained. Some functions benefit from strong distribution: local forwarding, multipath routing, store-and-forward messaging, replication and federation across administrative domains. Other functions may need bounded centres: software signing, vulnerability coordination, transparent registries, elected policy bodies or accountable appeals processes.
The challenge is to design these centres so that they are replaceable, plural and procedurally constrained. A central function is far less dangerous when users can verify it, fork it, or switch to alternatives without social exile. The goal is not to pretend that institutions can be removed. It is to prevent any single institution from becoming both technically indispensable and politically unanswerable.
The decisive question is who can impose updates, revoke identities or meter access at scale.
What endurance actually looks like
The practical test of a decentralised network is therefore not whether it can route around one blocked gateway in a laboratory demonstration. It is whether it can absorb ordinary governance stress without collapsing into either chaos or hidden hierarchy. Can users recover identity without pleading with a monopoly help desk. Can communities remove abusive actors without empowering permanent censors. Can maintainers push urgent fixes without creating an unquestionable sovereign at the update layer. Can local operators federate without yielding autonomy to a global directory.
Those are awkward questions because they replace romance with administration. Yet they point towards a more realistic definition of resilient network architecture. Endurance is not the absence of centres. It is the presence of multiple centres with limited scope, reciprocal checks and credible exit. In that sense, the future of mesh and decentralised networking may look less like a technical insurrection against the internet than like a constitutional redesign of the services built on top of it.
The radio still matters. But the harder part of a decentralised internet is learning where central authority remains necessary, then shrinking, auditing and contesting it before it hardens into the very dependency the network was meant to escape.
