Summary

  • IPv6 growth is substantial and valuable, but adoption measurements describe particular users, services and vantage points; they do not establish that every material counterparty can be reached without IPv4.
  • The decisive test of transition is counterfactual withdrawal: what breaks, who becomes unreachable and which contracts, security controls or public services fail when IPv4 is removed from a real network?
  • Translation and dual-stack designs reduce the amount of IPv4 required, yet preserve IPv4 dependency at shared gateways, public endpoints, allowlists, cloud edges and application destinations.
  • Cloud pricing, bring-your-own-address services and active transfer markets show that clean IPv4 authority remains an economic option even where the underlying compute estate is increasingly IPv6-capable.
  • Registry power survives through holder recognition, transfer completion, contact records, reverse DNS and route-origin authority. Those functions require stronger accountability, not neglect justified by protocol optimism.
  • Public-sector continuity and uneven regional adoption make aggregate milestones a poor basis for retiring compatibility before agencies, suppliers and citizens can complete essential transactions.
  • The Number Resource Society should support IPv6 abundance through research, member education and advocacy while pressing RIRs to keep IPv4 registration thin, portable and independently reviewable for as long as evidence shows material dual-stack dependence.

A milestone is not a repeal

In April 2026, Google's global measurement briefly placed IPv6 above 50% of users reaching its services. APNIC marked the event as evidence that IPv6 had become a mature protocol deployed at global scale. Both observations are warranted. A protocol that carries roughly half of traffic to one of the Internet's largest service families cannot be dismissed as marginal.

The same APNIC analysis also recorded a lower worldwide capability estimate from its own measurement system. In July 2026, the APNIC Labs public table showed a 30-day global average of about 42% IPv6 capability and about 41% preference, while Google's page displayed a different share based on access to Google. The difference is not a scandal. APNIC weights per-economy observations and tests retrieval of small web elements. Google observes users of its own services. Each instrument answers a defined question from a defined vantage point.

Governance goes wrong when an observed share becomes a declaration that the older address family has ceased to matter. Fifty per cent of Google access over IPv6 does not mean that fifty per cent of organisations can operate without IPv4. It does not identify the banks, identity providers, government services, suppliers, embedded devices or business partners that remain IPv4-only. It does not show whether a capable user chose IPv6 for every destination, or whether IPv4 remained available as a rapid fallback. Nor does it show how many IPv6-originating sessions reached an IPv4 destination through translation.

The milestone should therefore be read as an achievement and a denominator warning. It proves that large-scale IPv6 deployment works. It does not prove universal substitutability. The institutions governing scarce IPv4 holdings are not repealed by a line on a chart any more than property records are repealed when a newer transport system carries half of journeys. Their scope may shrink. Their obligations may change. Their continued power must be justified by live dependencies rather than habit. But until withdrawal is safe, the old rights and chokepoints remain operational facts.

The measurement history since 2012 describes coexistence

The time horizon matters. The IETF's RFC 6540, published in 2012, said that IPv6 support should no longer be treated as optional. That position was justified by depleted IPv4 supply and the limitations of indefinite workarounds. In the same period, APNIC Labs began the daily end-user measurements that now provide one of the longest comparable views of deployment.

APNIC's account of its measurement method separates capability from preference. A tested client is capable if it can retrieve an IPv6-only element. It is preferred if the dual-stack test is actually fetched over IPv6. The distinction is central. Capability measures an available path. Preference records a choice made under the conditions of the test. Neither says that IPv4 can be removed from the client's wider set of transactions.

Growth over the period is nevertheless striking. APNIC's earlier history places global end-user capability at roughly 0.3% when measurement began in late 2011, above 1% only in 2013, around 18% by the end of 2017 and around 30% by 2020. The Asia Pacific aggregate exceeded a 50% 30-day capability average in April 2025, according to APNIC's regional review. These are not cosmetic gains. They represent capital investment, equipment replacement, mobile-core redesign, operational learning and content deployment.

Yet the shape of the evidence is coexistence. The same regional review described wide differences among economies and slower change where fixed-broadband equipment has long replacement cycles. A regional average combined India above three quarters, Japan above half and Indonesia below one fifth at the stated observation point. A global service may see a high IPv6 share because large mobile populations and a few major content providers are highly capable, while a specific enterprise still confronts an IPv4-heavy set of counterparties.

The correct historical conclusion is therefore more demanding than either triumph or failure. IPv6 moved from negligible to mainstream between 2012 and 2026. It did so without making IPv4 institutionally irrelevant. The coexistence period became long enough for IPv4 scarcity, transfers, cloud pricing and translation services to mature alongside IPv6. By 2027, those parallel systems should be judged by the dependencies they serve and the power they confer, not by the expectation that one of them must have disappeared already.

Transition is not the same as substitutability

The word transition suggests a sequence: deploy the successor, move use, retire the predecessor. That sequence works only when the new service substitutes for the old at the points that matter. Address families do not make that decision alone. Reachability is bilateral. An IPv6-capable customer gains little from a destination that publishes only an IPv4 address unless a translator stands between them. An IPv6-enabled server does not help a customer on an IPv4-only access network unless the service retains an IPv4 edge or uses an intermediary.

This creates a coordination problem with an unusually stubborn tail. The value of keeping IPv4 is not proportional merely to the average share of IPv4 traffic. It depends on the value of the users and transactions that cannot complete without it. A payment processor, emergency information service or public benefits portal cannot dismiss ten per cent of users as an acceptable residue simply because ninety per cent can connect another way. The commercial and public cost of exclusion may be concentrated in the remaining minority.

Dual-stack resolves the coordination problem by making both paths available. Happy Eyeballs, standardized in RFC 8305, lets applications attempt suitable routes without forcing users to understand address families. The user sees a service that works. The operator sees two reachability domains, more possible failure combinations and a continuing need to know which path carried a disputed or failed transaction.

Translation changes where the burden sits, not whether the dependency exists. A mobile access network may operate IPv6 internally and use NAT64 to reach IPv4 destinations. A content provider may place dual-stack load balancers in front of an IPv6-heavy service estate. A cloud customer may use private addressing behind a small set of public IPv4 egress identities. In each case, the quantity of IPv4 can fall sharply. But the scarce addresses at the boundary become more consequential because many sessions, workloads or customers depend on them.

Substitutability must therefore be tested at the service boundary. Can the user complete the transaction? Can the operator attribute abuse accurately? Can the organisation move providers without losing a trusted public identity? Can the public buyer accept the service under its compatibility terms? IPv6 growth improves the answers. It does not answer them by itself.

The remaining IPv4 user is not a rounding error

Aggregate percentages flatten the distribution of dependency. A network can report a high share of IPv6 traffic because video, software updates and search are delivered by a few large dual-stack content networks. That traffic is voluminous and important. It may say little about the long tail of lower-volume destinations used for payroll, customs, healthcare, supplier administration, industrial maintenance or government reporting.

Application criticality and byte volume are different measures. A quarterly tax filing may move fewer bytes than a minute of video while carrying much greater consequence for the user. An identity callback from a legacy service may be one short exchange without which a whole session fails. A security appliance may consult an IPv4-only reputation service infrequently, yet block production traffic when the consultation cannot complete. A transition claim based only on traffic share systematically discounts such dependencies.

The same problem appears in customer counts. The final users without reliable IPv6 are not randomly distributed. They can cluster among small enterprises with old routers, public institutions with slow procurement, fixed networks with long-lived customer equipment, economies with low deployment and industrial environments where replacement is costly. Their value to a service and their vulnerability to exclusion may exceed their statistical share.

A defensible retirement decision needs a dependency inventory rather than a slogan. It should identify top counterparties by transaction value, not only bytes; failure-sensitive public services; inbound and outbound allowlists; third-party callbacks; devices with literal IPv4 configuration; customer networks that lack IPv6; and incident tools whose evidence is keyed to IPv4. Tests must include real geographic and access-network diversity rather than a laboratory where both paths are modern.

This does not mean every IPv4 dependency deserves indefinite preservation. Some are obsolete and should be removed. Some vendors use compatibility as an excuse to avoid investment. The inventory makes that visible too. It allows an organisation to distinguish a remediable application defect from a market-wide reachability requirement. What it prevents is the political convenience of treating unmeasured users as negligible. A modern institution should demand evidence of the tail before declaring that the tail no longer merits service.

Application compatibility preserves IPv4 at one end of the bridge

Apple's application rules offer a useful example because they were designed to accelerate IPv6 readiness without pretending the public Internet had already become IPv6-only. Since June 2016, applications submitted to the App Store have had to support IPv6-only networks. Apple's developer guidance tells developers to avoid IPv4-specific assumptions and test in an IPv6-only environment.

The same guidance explicitly says that an Internet server does not need to be updated immediately because an IPv6-only device can reach an IPv4 server through DNS64 and NAT64. That accommodation is practical. It allows an access network and client application to modernize while the destination remains on IPv4. It also illustrates why client success is not proof that the destination's IPv4 role has vanished. The translation service consumes and depends on IPv4 reachability on the far side.

Apple's more detailed DNS64 and NAT64 documentation lists persistent obstacles: address literals in protocols and configuration, older socket interfaces, small address containers and applications that behave differently when a server's direct IPv6 path is used. These are engineering defects worth correcting. Their continued existence also means that a policy declaration cannot create compatibility by itself.

Translation can conceal the boundary from the end user, which is often its purpose. Governance should not conceal it from the operator. The operator still needs capacity planning for translator state, logging sufficient for abuse and legal requests, monitoring of asymmetric failures and a plan for destinations that embed addresses or use protocols poorly suited to translation. The IPv4 addresses used by the translator or destination remain part of the service's effective infrastructure.

The lesson is neither that translation failed nor that direct IPv6 is unnecessary. Translation has been one of the mechanisms that allowed IPv6-first access networks to scale. Its success demonstrates that transition has frequently meant architectural intermediation rather than immediate endpoint replacement. Institutions deciding whether IPv4 records, rights and continuity can be neglected must count those mediated dependencies instead of classifying every successful client session as evidence of complete substitution.

Cloud pricing is a live valuation of compatibility

Cloud platforms translate scarcity into a visible product decision. In July 2023, AWS announced that from February 2024 it would charge USD 0.005 per hour for every public IPv4 address, whether attached or idle. Its announcement said the acquisition cost of a public IPv4 address had risen by more than 300% over five years and described the charge as both cost recovery and an incentive to use addresses efficiently and accelerate IPv6.

The important evidence is not the precise price. It is that a leading cloud provider found public IPv4 sufficiently scarce, demanded and measurable to meter across services. The charge applies in an environment with extensive IPv6 capability, private networking, load balancing, managed translation and global engineering resources. IPv6 did not make the public IPv4 product disappear. It changed the architecture and price at which customers consume it.

AWS also launched Public IP Insights so customers could see address use across accounts and regions. It later made it possible to remove an auto-assigned public IPv4 address from a running interface without rebuilding the instance. Those features create a conservation response: inventory the public boundary, remove accidental exposure, consolidate egress and reserve public IPv4 for workloads that require it. That is economically rational. It is different from saying no workload requires it.

The platform's VPC documentation describes the resulting hierarchy. Most resources use private IPv4. Direct Internet reachability over IPv4 uses public addresses. Customers can also bring their own public IPv4 and globally unique IPv6 space into the platform. In 2024, AWS expanded cross-account use of customer-owned IPv4 for Global Accelerator, reinforcing the value of a portable public identity that can be attached to cloud service rather than abandoned at migration.

Pricing therefore supplies a sharper institutional signal than broad transition rhetoric. Customers reveal which public identities they retain when every hour is billed. Platforms reveal that address authority has value when they build metering, audit and bring-your-own-address capabilities around it. Registries remain implicated because the customer-owned option depends on recognized holding, valid route authorization and accurate registration. IPv6 growth changes demand intensity, but it has not dissolved the chain of authority.

A market transfer remains an institutional event

The exhaustion of general IPv4 pools did not end distribution. It moved a larger share of distribution from administrative grants to transfers, returns, limited residual pools and commercial arrangements. APNIC's exhaustion guidance states that all RIRs have either limited supply or exhausted their ordinary pools. APNIC limits eligible new or existing members to a maximum final-pool delegation of a /23 and directs organisations needing more space toward transfers.

That market cannot function on possession alone. A buyer needs confidence that the seller is the recognized holder, that the block is eligible to move, that no unresolved dispute prevents transfer and that the receiving record will be accepted by networks and service providers. APNIC's transfer policy requires the source to be the currently registered holder and provides for a public transfer log. Its transfer conditions say that, once complete, the source no longer has rights to the resources and the recipient becomes the registered holder.

These are institutional acts. The commercial agreement may identify a price, warranties and delivery date, but the registry changes the widely relied-upon account of authority. If it delays, rejects or misrecords that change, the asset's practical usability and the parties' risk change. If it recognizes conflicting claims, routing and security services may become contested. If it imposes opaque conditions, it can redistribute bargaining power without owning the address space in an ordinary commercial sense.

IPv6 abundance does not remove this authority while organisations continue to buy or carry IPv4. It may reduce the quantity a growth network needs, encourage sharing and cap long-term demand. It may also make the quality of the remaining blocks more important. A clean registration history, usable reputation, acceptable geolocation and coordinated route-security state can differentiate one block from another.

The governance question is consequently not whether a registry should preserve scarcity. It should not manufacture scarcity or obstruct IPv6 to defend its relevance. The question is whether the registry performs the narrow acts required for voluntary exchange accurately, quickly and impartially. If real dual-stack dependence sustains transfer demand, neglecting those acts in the name of transition would not accelerate IPv6. It would make a still-needed market less reliable.

Registration authority outlives allocation authority

The original allocation era gave RIRs a visible gatekeeping role: organisations applied, demonstrated eligibility and received address space from a pool. Exhaustion reduces that role for IPv4 but leaves registration authority intact. The registry still records the holder, applicable status, contact points and transfer history. It still supports reverse DNS and the resource certificates used to make route-origin statements. Those functions can matter more when scarcity and transfer increase the number of contested or changing holdings.

APNIC's description of resource registration services calls its database the official record of information about organisations holding number resources in the Asia Pacific. Its guidance on network assignments distinguishes portable delegations made directly to members from non-portable downstream assignments that customers must return when they leave a provider. That distinction affects customer exit, bargaining power and continuity even when the packet forwarding is technically sound.

The institution therefore controls more than an address lookup. It maintains a set of recognized relationships. A direct holder may choose upstream providers because its delegation is portable. A downstream customer may be tied to renumbering when it changes provider. A transferred block gains a new recognized holder only when the record changes. A court, security team, cloud platform or counterparty may rely on those records when deciding whose instructions to accept.

This is why claims that IPv4 is an obsolete technology can be politically useful to an unaccountable institution. If the public treats the record as a dying administrative remnant, scrutiny weakens while the registry continues to exercise material discretion over assets and services. The opposite error is to inflate registry authority because IPv4 remains valuable. Continuing relevance does not justify a broad mandate over price, network design or competition.

A legitimate settlement keeps the function narrow and the evidence strong. The registry verifies identity and authority, records changes, preserves history, supports route security and exposes reviewable service performance. It does not turn protocol transition into a reason either to abandon holders or to govern every commercial use of their resources.

Route security makes the old record operationally current

Registration is sometimes described as paperwork separate from the live Internet. Resource Public Key Infrastructure makes the relationship harder to dismiss. A route-origin authorization states which autonomous system is permitted to originate a prefix. Networks performing route-origin validation can classify an observed announcement in relation to that authorization. A stale or incorrect statement can therefore affect how a route is treated.

The mechanism applies to IPv4 and IPv6. That symmetry does not make IPv4 institutional power disappear. A holder moving IPv4 between providers, transferring a block, changing an origin or bringing space into a cloud service needs the authority to update the relevant statements. If access to the certificate service is interrupted by an account dispute, governance failure or mistaken suspension, a technical transition elsewhere in the network offers no substitute.

The Number Resource Organization's RIR statistics page publishes common reports for delegated resources and RPKI adoption across both address families. The existence of separate coverage measures is itself useful discipline. It prevents an aggregate claim about route security from hiding different deployment and risk profiles for IPv4 and IPv6.

Reverse DNS and contact records create similar operational links. An enterprise may use reverse names in diagnostics, mail handling, reputation or customer controls. Abuse desks and investigators use registration contacts to identify the organisation responsible for an observed address. These systems are imperfect and should not be treated as proof of end-user identity. They remain part of the evidence used to coordinate incidents across networks.

Dual-stack increases the need for coherent records because one service can have two address families, different routes and different security states. An organisation may fix the IPv6 route while leaving an obsolete IPv4 authorization, or update a holder name on one resource family but not the other. Institutional quality is visible in whether the registry lets holders reconcile those states quickly and whether relying parties can distinguish current authority from history. Protocol growth raises that standard; it does not make the record ceremonial.

Procurement can preserve IPv4 long after engineering moves

Public-sector technology changes at the pace of contracts, accreditation and service continuity as much as at the pace of protocol implementation. The United States Office of Management and Budget's M-21-07 memorandum illustrates an ambitious response. Issued in 2020, it called for agencies to move toward IPv6-only operating environments, set staged targets and tied acquisitions to IPv6 capability.

Such purchasing rules can change vendor incentives. A supplier that previously treated IPv6 as optional must demonstrate support to compete for government work. Shared testing and explicit acceptance criteria reduce the excuse that no customer has asked. Public-sector demand can therefore accelerate deployment beyond what isolated technical teams could achieve.

The same contracts also preserve IPv4 where continuity clauses, legacy dependencies or citizen access require it. A government service cannot infer that every resident has reliable IPv6 from a global traffic share. An agency cannot retire an IPv4 interface if a critical supplier, field device, intergovernmental partner or emergency channel still depends on it without an accepted bridge. IPv6-only targets typically require inventories, milestones and exceptions precisely because withdrawal has consequences.

The institutional danger lies in confusing aspiration with observed readiness. A registry or policymaker may point to a public target as proof that IPv4 governance can be downgraded. An incumbent supplier may point to residual dependencies as proof that modernization should be deferred indefinitely. Both positions avoid measurement.

Public-sector continuity needs a service-by-service evidence set: external user reachability, supplier readiness, translation capacity, security accreditation, incident attribution, accessibility and tested fallback. Exceptions should have owners and expiry reviews. Success should be measured by safely retired dependencies, not by an announcement or an address-family share alone. Until that evidence supports withdrawal, the IPv4 registrations and public identities attached to essential services remain current public infrastructure.

Uneven geography defeats a single transition date

Global averages can conceal differences of several multiples among economies. APNIC's 2025 regional analysis recorded high capability in India, Japan and Viet Nam, much lower levels in several other large economies and a very low aggregate across Africa. By July 2026, APNIC Labs still showed substantial variation among regions and subregions. The exact values move daily, but the structural conclusion is stable: there is no single global deployment state.

This variation matters to organisations whose customers, workers or suppliers span regions. A service headquartered in a high-adoption economy may still need IPv4 to reach customers in a lower-adoption one. A cloud region may offer strong IPv6 features while a private connection, payment partner or local access provider requires IPv4. A multinational cannot retire compatibility based on its home-network statistics.

Geography also interacts with market power. Large operators can spread dual-stack engineering and public IPv4 acquisition across millions of customers. Small providers in low-revenue markets face the same need for reachability with less capital and a smaller technical team. They may rely more heavily on upstream translation, leased addresses or provider-assigned space. Their customers then inherit the provider's choices and its continuity risk.

This is not a reason to slow IPv6. Economies with scarce IPv4 and rapid user growth can gain greatly from IPv6-first mobile and access networks. The evidence from India and Viet Nam demonstrates that deployment at population scale is possible. The governance lesson is that successful IPv6 deployment and continuing IPv4 compatibility can coexist for a long period, with the burden distributed unevenly.

A global institution should therefore avoid one date that declares the older address family institutionally irrelevant. It should publish disaggregated measures, support local evidence and make services portable across different stages of adoption. The relevant end state is not uniform ideology. It is a world in which networks can choose an efficient architecture without losing customers or surrendering recognized resource rights because their region moves at a different pace.

Operators carry a dual obligation, not two identical networks

The phrase dual-stack can imply complete duplication. Real operators use a range of architectures. Some run both address families to the customer. Some provide IPv6 with shared IPv4. Some keep IPv4 inside a legacy domain while moving new services to IPv6. Some concentrate public IPv4 at gateways and use private addressing internally. The costs and dependencies differ.

What remains common is a dual obligation: preserve broad reachability while moving toward a more scalable address system. Operators must decide where IPv4 is indispensable, where translation is acceptable and where direct IPv6 improves efficiency or performance. They also need records that identify which organisation controls each public boundary and which network may originate it.

Scarcity can sharpen incumbent advantage. An operator with ample clean IPv4 space can add customers, offer dedicated public addresses or avoid some translation constraints. A new entrant may buy space, lease it, use carrier-grade translation or depend on an upstream. Each choice affects capital, abuse attribution, customer experience and exit. IPv6 can reduce that disadvantage, particularly for internal growth, but cannot erase it if customers and counterparties still demand an IPv4-reachable edge.

Registry conduct influences the options. A slow transfer adds timing risk. Unclear holder records raise due-diligence cost. Non-portable downstream space increases renumbering cost at provider exit. Inaccessible route-security services can turn an account dispute into a reachability problem. None of those issues can be answered by telling the operator that IPv6 is the future.

The operator's responsibility is also real. Scarcity is not permission to hoard unused space, maintain inaccurate contacts or externalize all translation costs to users. A credible holder can document use, consolidate waste, deploy IPv6 and maintain security state while preserving the IPv4 compatibility its customers still need. Institutional accountability should reward that evidence rather than force a false choice between modernization and continuity.

Cloud platforms inherit and amplify registry power

Cloud platforms are often described as an escape from physical network constraints. In practice they reorganize those constraints into service boundaries. A customer may deploy thousands of private workloads behind a small number of public endpoints. That architecture conserves IPv4, but it also gives the platform substantial control over public identity, egress, logging and migration.

Platform-provided addresses are easy to attach and difficult to carry elsewhere. A customer that changes provider may need to replace allowlist entries, rebuild reputation and coordinate counterparties. Bringing a customer-owned prefix can improve continuity, but it requires sufficient address space, accepted registry authority, platform eligibility, routing coordination and valid route-origin statements. The outside option is strongest for organisations that already possess clean portable space.

IPv6 can reduce dependence on platform-owned IPv4 by giving workloads stable global addresses and making large-scale internal planning easier. Yet customer adoption depends on whether external services accept IPv6 and whether the platform exposes feature parity across load balancers, security products, managed databases, containers and observability. A headline statement that the platform supports IPv6 says less than a service-by-service matrix tested against the customer's design.

The platform also becomes a private allocator of scarce public IPv4. It decides product availability, charges, quotas, account-level permissions and which services can use customer-owned space. These decisions may be commercially reasonable. They still shape who can obtain a stable public identity and at what price. The underlying registry record determines which customer claims can be brought into that environment, while the platform determines how useful the claim becomes.

Governance must recognize the combined chain. Registries should not regulate cloud architecture beyond their competence. Platforms should not be treated as neutral pipes when their product terms determine portability. Customers should measure both direct address charges and the cost of changing public identity. IPv6 progress weakens some forms of dependence, but the remaining IPv4 boundary can become more concentrated and institutionally valuable as internal networks modernize.

The withdrawal test is stricter than an adoption target

A credible claim of completion should be framed as a controlled counterfactual: if IPv4 were withdrawn from this service, what would fail? The test can be run at a workload, customer segment, agency, network or economy level. It cannot be answered by a global average alone.

The first measure is reachability. What share of intended users can complete representative transactions over native IPv6 or an approved translation path? The test should weight critical and high-value transactions, not merely bytes. It should include lower-adoption economies, fixed and mobile access, enterprise proxies and old customer equipment.

The second measure is application behavior. Does name resolution work without literal addresses? Do identity callbacks, webhooks, voice protocols, file transfer, monitoring and vendor licensing work? Are direct IPv6 paths as reliable as the translated paths used during earlier testing? Can incident teams distinguish failures by address family?

The third measure is institutional continuity. Can the organisation preserve a trusted public identity, route authorization, reverse DNS and abuse contact during provider change? Do cloud and security suppliers support the intended IPv6 design without requiring an IPv4 exception? Have public buyers and regulated counterparties accepted the architecture?

The fourth measure is residual concentration. How many customers, sessions or services depend on each remaining public IPv4 address or translator? Consolidation can reduce the number of addresses while increasing the consequence of one failure. A small inventory is not the same as a small dependency.

Finally, the test needs reversibility. A temporary withdrawal should produce observable evidence and permit safe restoration. Results should distinguish defects the organisation can fix from counterparties it cannot control. Over time, the inventory should shrink as remediation and market adoption proceed. A transition target becomes credible when repeated withdrawal tests show that the cost and exclusion risk have become immaterial, not when a calendar or chart says they ought to have done so.

Institutional power should shrink with measured dependency

The survival of IPv4 dependence is not an argument for preserving every old rule. Institutions should lose power as the functions that justified it disappear. A registry that once allocated from a large free pool should not reinvent broad discretionary rationing merely to remain central after exhaustion. A provider should not use a customer's need for compatibility to block transfer or retain control of portable space. A platform should not portray avoidable lock-in as a technical necessity.

Measured dependency provides a way to reduce power without denying reality. If an organisation needs IPv4 only for a few public edges, its registry and platform relationship should be limited to those edges. If a block moves through a voluntary market, the registry's role should be identity, authority, accurate recording and route-security continuity rather than deciding the buyer's business model. If a customer can use IPv6 for internal scale, policy should not require scarce IPv4 for every workload.

Transparency is equally important. Service statistics should expose transfer times, contested cases, correction times and security-service availability. Cloud customers should be able to inventory public address use and understand which services prevent IPv6-only operation. Operators should report significant compatibility dependencies and progress removing those they control. Public agencies should publish exceptions without exposing sensitive network detail.

This approach avoids two forms of mandate laundering. The first says IPv4 is disappearing, so holders need less protection from registry error. The second says IPv4 remains essential, so the registry deserves wider control over markets and networks. Both are false. Continuing dependency justifies reliable narrow services and strong remedies. Declining dependency justifies a smaller footprint. Neither justifies unreviewable discretion.

The practical aim is graceful institutional contraction. As IPv6 becomes genuinely substitutable, fewer transactions should require IPv4, fewer public identities should be retained and less value should sit behind each registry decision. Institutions should facilitate that reduction while remaining accountable for the rights and continuity that have not yet disappeared.

A Number Resource Society can advocate abundance without denying scarcity

The Number Resource Society is most credible when it refuses the forced choice between IPv6 advocacy and protection of members affected by IPv4 decisions. IPv6 abundance is desirable because networks need room to grow, design cleaner end-to-end connectivity and reduce dependence on increasingly concentrated IPv4 boundaries. IPv4 remains scarce and valuable because compatibility is incomplete. Both statements can be true without turning the Society into a registry operator or a defender of the status quo.

Its IPv4 role should be advocacy, research and authorized member representation. It should document whether the responsible RIR maintains a trustworthy holder record, portable authority, transfer history, contacts, reverse delegation and route-security capability. It can press RIRs and other lawfully authorized operators to provide predictable handoff when a holder changes service provider or platform, and it can represent a member in a dispute only under a valid power of attorney. NRS does not maintain the record, execute transfers, preserve the authoritative evidence or decide remedies.

Its IPv6 role should also respect limits. The Society can publish deployment evidence, explain member experience and advocate simpler allocation and registration procedures, stronger route-security support and removal of administrative barriers. The responsible RIRs make allocations, maintain registration and operate the relevant security services. NRS should not claim institutional credit for traffic it did not deploy or use IPv6 promotion to excuse poor treatment of IPv4 holders. Technical abundance does not make institutional accountability optional.

The Society should publish a dual-stack dependency report built from observable measures: capability and preference by economy, transfer volumes, public address utilization, portability events, route-security coverage, customer withdrawal tests and unresolved compatibility classes. The report should explain denominators and avoid converting one service's traffic into a universal conclusion.

Most importantly, members need remedies that work during transition. A holder should not lose an IPv4 asset because the registry considers it obsolete, nor be forced to retain a provider because a transfer is administratively difficult. An IPv6 holder should not be trapped by a hosted security service. Continuity belongs to the resource record and the network user, not to one institution. NRS earns a positive role by documenting failures, advocating independent review and supporting or representing members acting under valid authority, rather than by declaring old power dead, making it permanent or taking that power for itself.

The 2027 horizon requires scenarios, not prophecy

No defensible analysis can know a single global transition state for 2027. The useful approach is to define scenarios and the evidence that would distinguish them. In a faster-adoption case, major access networks, cloud services and public agencies remove remaining application barriers, direct IPv6 reachability improves and more organisations pass withdrawal tests. IPv4 demand becomes concentrated in legacy compatibility, and address holdings can be reduced more rapidly.

In a continuation case, global capability rises but remains uneven. Large content flows prefer IPv6 while commercial and public-service tails preserve IPv4 edges. Translation, public IPv4 pricing and transfers remain ordinary. Registry power narrows in quantity but stays material at each change of holder, origin or provider.

In a fragmentation case, geopolitics, platform differences and uneven investment create separate compatibility zones. Some networks operate IPv6-first with translation, while others remain heavily IPv4. Cross-zone services require intermediaries whose public address inventory and logging become concentrated chokepoints. A global average improves while practical portability worsens.

The evidence to watch is not only the IPv6 percentage. It includes the number and importance of IPv4-only destinations; direct versus translated IPv6 sessions; cloud public IPv4 use and charges; transfer volumes and completion times; customer-owned address use across platforms; RPKI coverage and service availability; public-sector exceptions; and verified withdrawals by large services.

Each scenario supports IPv6 investment. None supports ignoring current IPv4 authority. The faster case requires reliable transfer and continuity while inventories contract. The continuation case requires durable thin registration. The fragmentation case requires stronger portability and independent evidence. The institutional response should adapt to observed dependence rather than bet customer rights on a date.

The watchpoint is where compatibility becomes authority

The most important places to watch are boundaries where a technical requirement becomes someone else's power. A translator that serves millions of users controls a concentrated IPv4 exit and the logs needed to interpret it. A cloud platform that supplies public addresses controls price, attachment and migration terms. A registry that recognizes a transfer controls whether a commercial transaction becomes a usable public claim. A public buyer that writes compatibility requirements shapes which suppliers can compete.

None of these actors is illegitimate merely because it has power. The question is whether the power is bounded, evidenced and contestable. Can the customer see the dependency? Can it choose another provider? Can a disputed registry action receive independent review? Can a service demonstrate why an IPv4 exception remains necessary? Does a platform let a qualified customer carry its own public identity?

IPv6 growth should improve these answers. It gives customers an alternative address family, lets networks move internal scale away from scarcity and reduces the quantity of IPv4 needed at the edge. Its success can make concentrated IPv4 authority easier to identify. That is a governance benefit if institutions use the visibility to reduce lock-in rather than to dismiss the remaining dependency.

The conclusion is deliberately unspectacular. IPv6 works. IPv4 still matters. The proportion and location of that importance are changing, and good evidence can show where. A serious institution neither mocks the transition nor declares victory before users can leave the older system safely. It protects current rights, accelerates viable alternatives and allows its own authority to contract as the evidence warrants.

The dual-stack era is not an excuse for permanent duplication. It is a test of whether technical progress can reduce institutional power without sacrificing customer continuity. By 2027, the strongest proof of progress will not be another rounded percentage. It will be a growing number of networks that can remove IPv4 from defined services without losing users, and a shrinking set of IPv4 institutions that remain powerful only where measured dependence still makes them necessary.