Build the way out before you need it.

When the registries were set up, addresses were plentiful and cheap, and a registry could look like an address book. That world ended with the free supply of IPv4. The turning point came when Nortel went bankrupt and its 666,624 IPv4 addresses were sold to Microsoft as assets in the proceedings. Since then addresses have become transferable, financeable, leased, litigated and subject to sanctions, and they sit inside live networks.

The language of the rules did not change with them. Rules that were harmless for free labels now act on something valuable, and every restriction has a price: buyers, lenders, lessees and insurers all discount addresses whose future depends on a registry’s discretion. The visible costs are fees, paperwork and delay. The larger ones are hidden: capital that cannot move, collateral worth less, investments never made.

An address is not just a number on a registry’s list. Customers’ firewalls and allowlists, payment systems, cloud deployments, contracts and years of security settings are built around it. The longer a network depends on one gatekeeper, the more of this accumulates, and the more a forced change costs: renumbering touches everything that has learned to trust that address.

Delay has a price as well. Suppose an operator has borrowed US$2 million for a purchase and a registry review adds sixty days: at 10% a year, that is about US$32,877 in interest alone, before any customer is lost. The figure is an illustration, not a measured loss, but the mechanism is real, and an approval that finally arrives does not refund a cost already paid.

The institutions are not steady enough to postpone the question. The registry for Africa has spent years in court, went without a quorate board from 2022 and was placed under a court-appointed receiver; its newly elected board then ratified a policy that keeps addresses inside the region.

At the same time, the rules that recognise the registries are being rewritten. ICANN has published a second draft of an RIR Governance Document that would replace ICP-2, the policy under which the registries are recognised, and set new rules for how they operate and how they can be derecognised. Whether the new rules give networks a real way out, or lock the present structure in place, will shape the coming decades.

Consent is thinning too. Turnout in registry elections is low, operators already treat their addresses as assets whatever the rules call them, and the United Nations has made its Internet Governance Forum permanent without touching any of the levers that actually control networks. A system that runs on voluntary cooperation rarely collapses in a day. It loses the trust of the people who run the networks, one decision at a time.

The path is predictable. More disputes will move from policy rooms into courts, which will be asked to freeze accounts, appoint receivers and decide who may act for a registry. States will stop treating the registries as harmless technical associations. Operators will route around registry authority, relying on contracts, court-backed transfers and their own records. The global bodies above the registries will be tempted to centralise and governments to nationalise, and the result could be conflicting registries and political pressure on routing.

The break may not be gradual either. If a registry can act against a major network or a country’s address space at almost no cost to itself, a single decision could split the one shared record the Internet relies on, leaving security systems and transfer records pointing at different versions of reality. Restoring a single record after that would be close to impossible.

The continuity everyone rightly worries about is the continuity of a function: unique numbers, accurate records, directory and security services, and running networks. It is not the continuity of any one institution’s power. The more critical the function, the less it should depend on the survival or goodwill of a single organisation.

That kind of continuity cannot be improvised in a crisis. The security system alone involves keys, certificates, repositories and the trust other networks place in them; moving it safely takes a plan, rehearsal and objective triggers agreed in advance. The transition has to begin early enough that the network survives the cure.

For AFRINIC I have proposed a timetable: an immediate halt to revocations, reclaims, transfer refusals and interference with security or reverse-DNS services on non-technical grounds; objective transfer recording and an end to regional export bans within 120 days; a design for portability and failover within 270 days; a rewrite of the registry’s whole role within a year; and an independent review body within eighteen months.

Operators need not wait for policy rooms. They can keep a record of every rule they relied on and every delay they suffered, hold registries to commitments already made, use ordinary law where a rule causes them loss, and organise with other holders now, while they still have leverage. Those who wait will organise after the break, under worse conditions and at a higher cost.

My own forecast, made in March 2026, is that by 2033 a distributed ledger will have replaced the registries as the Internet’s record. The direction matters more than the date: build the way out while there is still a choice.