Summary
- RIPE NCC's 2025 Annual Report lists 19,863 active members, 20,647 active LIRs and 120 member countries. These measure legal relationships, administrative accounts and geographic reach. They are not three versions of the electorate.
- The October 2025 General Meeting analysis identifies 19,665 eligible members, 1,004 registrations and 801 cast votes. This is a genuine participation funnel, but it still does not show active operating organisations, common corporate control, staff capacity or downstream users.
- Other RIRs use materially different denominators. ARIN distinguishes a large Service Member population from a much smaller General Member electorate. APNIC assigns tier-based vote weights. LACNIC's 2025 board election reported authorised organisations, organisations voting and weighted votes separately.
- Cross-registry turnout comparisons are unreliable unless every numerator is paired with the correct legal denominator, date, weighting rule and unit. A ballot carrying 64 entitlement units cannot be compared directly with one organisation under one-member-one-vote.
- A common public standard should publish a denominator ladder, stage-to-stage losses, distributions, uncertainty and privacy protections. Number Resource Society should adopt the same standard from inception so claims of broader representation can be tested rather than asserted.
Exhibit one: three true RIPE NCC totals
The RIPE NCC Annual Report 2025 places several headline facts close together. At 31 December 2025, it reports 19,863 active members, 20,647 active LIRs and members in 120 countries. It also reports 15,273 members with an IPv6 allocation.
Each figure answers a useful question. How many active contracting member entities are recorded? How many active LIR accounts do those members hold? Across how many countries are members registered? How many members have a particular category of allocation?
None answers how many organisations were eligible to vote at a specific meeting. The year-end member count may differ from the meeting-date eligible population. Candidate members, suspended members or tenure rules may affect eligibility. A member can hold more than one LIR. Country is based on a registration relationship, not necessarily the location of every network or user.
The figures are not defective because they differ. They become misleading when a report or reader slides between them. “Twenty thousand LIRs are represented” can imply one account, one network and one voice. “The membership spans 120 countries” can imply territorial representation. “Most members have IPv6” can imply operational deployment.
The report itself warns that one member can hold more than one LIR. That warning should be treated as a general rule of statistical reading: the units do not collapse merely because they appear in the same panel.
Representation begins with a defined principal. Is the institution claiming to represent legal members, LIR accounts, operating networks, resource holders, countries, technical entities or affected users? More than one answer may be legitimate, but each requires a different measure.
Exhibit two: an actual voting funnel
RIPE NCC's October 2025 analysis provides a more precise chain. It reports 19,665 eligible members, 1,004 members registered to vote and 801 members that cast ballots. Eligible-member turnout was 4.1 percent.
This is better than dividing meeting attendees by the annual member total. The numerator and denominator use the same legal unit: member organisations. Registration and completion are separate stages. The date and meeting are clear.
The funnel still contains unknowns. Eligibility says that a member could vote under the rules, not that it had a current internal mandate or effective notice. Registration says that someone acting for the member completed a step, not that the company considered the resolution. A cast ballot says that the organisation's voting right was exercised, not how many networks, employees or customers supported the choice.
The analysis improves further by examining six meetings. It identifies 400 members registered at all six and 1,932 registered exactly once. That distribution shows whether the electorate is recurrent or episodic. At a low-turnout meeting, frequent entities account for a larger share of registered votes.
It also compares groups across allocated IPv4 and IPv6 resource bands. The distributions broadly align, with some large outliers more likely to appear frequently. This is useful because it tests one possible concentration mechanism rather than assuming it.
But resource allocation bands are not a complete operator measure. They do not reveal whether addresses are active, how many staff can follow governance, whether several members share corporate control, or how many downstream organisations depend on a vote. The funnel measures formal participation well. It does not yet measure representative incidence.
Exhibit three: ARIN's population change between service and vote
ARIN's membership model illustrates why a headline member count can be far from the electorate. Current membership rules distinguish Service Members from General Members. General Members in Good Standing with a properly designated Voting Contact can vote. After the 2023 election and each election thereafter, a General Member that has not cast a ballot in any of the previous three elections reverts to Service Member status, though it may reapply.
At ARIN 56 in October 2025, the meeting transcript described a population of about 26,000 Service Members and a little over 1,600 General Members eligible to vote, subject to good standing and voting-contact conditions. The exact figures are a snapshot stated in a meeting, but the order of magnitude demonstrates the denominator problem.
If an election reports ballots as a percentage of General Members, the result measures participation among the qualified electorate. If a public statement compares those ballots with all Service Members, it measures something else: the combined effect of qualification and turnout.
Neither is inherently illegitimate. ARIN adopted active-participation requirements to define its constitutional membership. The analytical requirement is to show every transition. How many service organisations could seek General Membership? How many did? How many were in good standing? How many had a valid Voting Contact? How many voted? How many reverted after three missed elections?
Without the ladder, a rising turnout percentage may reflect a shrinking denominator rather than more organisations voting. The inverse can also occur: expanding eligibility can lower the rate even if ballots increase.
Representation reports should therefore publish counts and rates together. “Forty percent turnout” is incomplete. Forty percent of which organisations, selected by which rule, on which date?
Exhibit four: APNIC counts entitlement units
APNIC adds a different complication: vote weight. Its 2026 by-law reform voting information states that voting entitlement varies by membership tier, from one vote for an Associate member to 64 for an Extra Large member. Each member submits one ballot carrying all its entitlement units; a proxy submits a separate ballot for each represented member.
Three numbers may therefore be reported. One is the number of members entitled to participate. Another is the number of member ballots submitted. A third is the total weighted votes carried by those ballots. They are not interchangeable.
Suppose one hundred small members cast four votes each and ten extra-large members cast 64 each. There are 110 participating organisations and 1,040 entitlement units. Saying “1,040 members voted” would be false. Saying “110 votes were cast” would erase the constitutional weighting. Both dimensions are necessary.
NIR relationships add another layer. APNIC's headline membership includes National Internet Registry members in the overall figure. The organisations and users served through an NIR may not stand in the same direct relationship as an APNIC member. A regional total can therefore contain direct and indirect structures.
This does not make APNIC statistics unusable. It makes a labelled data model essential. Reports should distinguish direct members, NIRs, customers represented through NIR arrangements, voting-eligible members, ballots, proxies and entitlement units.
Cross-RIR comparison becomes particularly fragile here. RIPE NCC generally gives one vote to each eligible unsuspended member. APNIC's tiered ballot measures constitutional influence differently. Comparing “number of votes” without converting back to participating organisations would compare unlike units.
Exhibit five: LACNIC publishes three denominators in one election
LACNIC's 2025 Board of Directors election report provides a compact example of good labelling. It states that 12,501 voters were authorised, 1,207 organisations voted and participation was ten percent. It also reports 3,127 total votes, because voting weight varies.
The authorised-voter phrase must be read as authorised member organisations in the election context, not 12,501 individual humans. The 1,207 figure is the organisational numerator for participation. The 3,127 figure is the weighted decision total distributed among candidates and abstention.
These figures support different analyses. Organisational participation was about one in ten. The average entitlement carried by a participating organisation was greater than one, but an average would conceal the tier distribution. Candidate shares describe weighted electoral support, not shares of organisations.
A complete report would add the distribution of authorised and participating organisations by voting category, without identifying ballots. It would show whether high-weight members participated at a different rate and what share of total entitlement was exercised.
Again, this is not an argument against weighting. It is an argument against statistical ambiguity. A constitutional design may intentionally give larger organisations more votes. Legitimacy then requires showing both breadth and weight.
LACNIC's publication of authorised organisations, participating organisations and weighted votes is a useful baseline. The remaining challenge is comparability across years and institutions, where labels and translations may differ.
Countries are reach, not representation
Country counts are attractive because they communicate regional breadth. A registry with members in 120 countries appears diverse. The fact is relevant to service obligations, language, sanctions exposure, currency risk and meeting geography.
It does not prove that 120 national constituencies are represented in governance. Members are legal entities, not delegates appointed by countries. Several thousand members may be concentrated in one country while another has one. A company's registration country may differ from where it operates networks or serves users.
Country-level turnout can reveal disparity. RIPE NCC publishes registrations by country and has analysed participation rates. Such tables need minimum-cell rules and careful interpretation. A country with three eligible members and one registration has 33 percent turnout, but that rate is not more representative than hundreds of ballots in a larger country.
Population comparisons introduce further assumptions. National population is not the RIR membership denominator. Internet use, market structure and the number of operators vary. A state-owned national network and a fragmented commercial market will produce different member counts.
Countries also contain unequal speech and participation conditions. Sanctions, banking restrictions, visa access, language and political risk can affect members. A simple map may hide these institutional barriers.
The correct claim is modest: membership has a geographic distribution. To assess territorial representation, the report must show eligible organisations, participating organisations and operational presence by country, along with uncertainty and concentration. Even then, it should not imply that companies vote as national representatives.
Attendees are people, ballots are organisations
Meeting statistics often report registrations, attendees, remote entities, speakers and countries. These are valuable measures of community activity. They are particularly relevant to open policy forums where individuals contribute in their own capacity.
They do not automatically measure member governance. Several employees from one company may attend. One person may represent several member organisations by proxy. Many attendees may be non-members. An individual can contribute to a policy discussion without authority to cast the corporate ballot.
The distinction between the RIPE community and RIPE NCC membership is institutionally important. The community's open technical discussion is not reducible to corporate membership. The General Meeting's legal decisions are not reducible to room attendance. Both can occur near the same event.
A report should therefore maintain separate tables. Community participation counts people and contributions. Association governance counts eligible member organisations, representatives and ballots. Cross-links may show how many voting organisations had someone attend, but privacy and role definitions matter.
Speaker diversity is another distinct measure. Ten countries on a programme do not establish that member preferences were geographically balanced. A widely viewed webcast does not increase the formal electorate. Conversely, a low physical attendance does not imply low voting if electronic ballots are widely used.
Using attendance as a representation proxy tends to favour visible participation. Quiet members that vote remotely disappear. Active non-members appear to possess constitutional authority they do not have. The solution is not to choose one number but to name both institutional spaces.
Members are not necessarily active operators
The title “member” can suggest an organisation actively operating a network. That is often true, but the public membership count alone cannot establish it. A legal member may be a holding company, consultant, enterprise, public body, university, network operator or organisation winding down an LIR account.
Resource records provide clues. An active autonomous system, recent registry changes, routed address space or certification activity may indicate operation. None is conclusive. A network can rely on another organisation's ASN. A resource can be legitimately held without appearing in a chosen routing snapshot. A company can outsource operations.
Active-operator distribution should therefore use multiple indicators and publish bands. For example: evidence of routed resources, recent authenticated account activity, direct service operation, downstream sponsorship and current technical contacts. The method should state false-positive and false-negative risks.
Corporate control also matters. Ten eligible legal entities may belong to one group. One legal member may contain several autonomous operational divisions. A representation analysis that counts legal persons correctly can still miss control concentration.
Beneficial ownership data are difficult and sometimes unavailable. The registry should not publish speculative group maps. It can require members to disclose ultimate control confidentially for conflict and concentration analysis, with aggregate results and a challenge route for errors.
The objective is not to replace one-member-one-vote with one-operator-one-vote. “Operator” is too contestable for automatic constitutional use. It is to show whether the formal electorate resembles the operational population affected by decisions.
Users are the widest and weakest denominator
End users and downstream organisations depend on networks but do not generally vote in RIR association elections. Counting them is difficult. Retail subscribers, enterprise customers, hosted services, public institutions and people sharing connections cannot be added as if they were identical units.
Customer numbers are commercially sensitive and definitions vary. A wholesale customer may represent millions of users. A cloud service may host many organisations. Traffic volume can be dominated by a few applications. Population served can overlap across providers.
These weaknesses make user counts unsuitable as a direct voting weight. They do not make user impact irrelevant. High-consequence decisions should include ranges for downstream organisations, critical services and plausible affected populations. Methods and uncertainty should be explicit.
Representation can then be evaluated in layers. The member vote supplies corporate authorisation. Operator indicators show where implementation risk lies. User-impact evidence shows external consequence. No layer impersonates another.
This layered approach prevents two common errors. The first is plutocracy: giving the largest commercial network formal control because it claims more users. The second is corporate closure: treating all effects beyond member entities as politically invisible.
Users are not the legal electorate, but a legitimate institution should know when a narrow legal electorate imposes broad external costs.
Rates fail when dates do not align
Even correctly labelled units become misleading if dates differ. An annual report may give membership at 31 December. An election eligibility list may be fixed forty-five days before voting. Ballots may close weeks later. Companies join, close, suspend or change status between each point.
A turnout calculation should use the official eligibility denominator for that contest, not the nearest annual total. If that denominator is unavailable, the report should not manufacture precision from a snapshot.
Longitudinal charts need stable definitions. If ARIN changes General Member qualification, the series should mark the break. If APNIC changes tiers or proxy rules, vote-unit comparisons need restatement or caution. If RIPE NCC changes candidate-member tenure, eligibility changes even with stable member totals.
Cross-sectional comparisons need a common reference period. Comparing RIPE NCC May turnout with LACNIC October election participation can be informative only if issue type, weighting and electorate definitions are shown. It should not become a ranking of democratic quality.
Machine-readable data should include observation date, effective date, publication date and revision history. A corrected election report should not silently overwrite the number used in prior analysis.
Time is part of the denominator. A membership population is not a timeless fact.
A common denominator ladder
Every RIR should publish the following ladder for each corporate election or resolution.
Service population: all organisations receiving defined registry services, separated by direct member, NIR or sponsoring relationship.
Constitutional membership: legal persons recognised as members on the relevant date, with membership categories.
Eligible organisations: members entitled to vote after tenure, good-standing, suspension and contact rules.
Authorised representatives: organisations with a valid person or role able to exercise the vote.
Registered organisations: eligible organisations that completed election registration where required.
Participating organisations: organisations submitting a ballot, including an explicit blank or abstention under the rules.
Ballots: the number of organisational ballots, separated from proxies.
Entitlement units: total weighted votes carried by ballots where weights exist.
Decision units: votes counted for each option and abstentions under the applicable rule.
Operational exposure: privacy-protecting distributions for account count, active operation and downstream dependence.
Each stage should report the transition from the preceding stage. Reasons for exclusion should be aggregated: tenure, suspension, unpaid status, no contact, inactive class or another rule. The ladder makes institutional narrowing visible.
Distribution is more important than another total
Once the ladder exists, the next question is concentration. How many participating organisations are repeat voters? How many share common control? Which membership ages, regions and capacity bands are present? How much entitlement weight is exercised by each group?
Totals can rise while concentration worsens. An election may attract more ballots entirely from one well-organised category. A country count may increase because one multinational adds entities. A high-weight electorate may remain narrow even as low-weight participation grows.
Distribution tables should use medians, percentiles and bands rather than named rankings. Corporate groups can be reported in concentration ranges. Small cells should be suppressed. Analysts should be able to reproduce aggregate calculations from released data.
The report should include both absolute counts and shares. A region's participation rate can fall while its ballot count rises because eligibility expanded. Showing both prevents false stories.
Uncertainty belongs in the table. Active-operation indicators are estimates. Control mapping may be incomplete. Downstream reach may be self-reported. A confidence field is more honest than a polished total assembled from incompatible sources.
The desired result is not one universal score. Representation has several dimensions that can move differently. Breadth, equality, continuity, operator exposure and user consequence deserve separate indicators.
Privacy is a design constraint, not an excuse for blindness
Member-level voting behaviour is sensitive. Publishing that a named organisation registered or repeatedly voted may be allowed in some systems and inappropriate in others. Linking it to resources, country and corporate control can create security and commercial risks.
The denominator standard can operate with aggregate data. Minimum cell sizes, coarse bands, delayed publication and independent statistical review reduce identification risk. Ballot choices must never be joined to operational profiles.
An auditor can receive confidential member-level data, test the joins and publish certified aggregates. The institution should state which fields were unavailable and which transformations were applied. Researchers can inspect methods without receiving the underlying identities.
Privacy should be applied consistently. A registry should not publish a country table granular enough to identify the only member in a state while refusing harmless region-level capacity bands on privacy grounds. A documented disclosure framework is preferable to ad hoc decisions.
Members should be told what governance analytics use their data. Use should be limited to institutional evaluation, not marketing, enforcement or political targeting. Retention periods and access controls matter.
The choice is not between named surveillance and no measurement. Mature institutions routinely publish meaningful distributions while protecting individuals and organisations.
What the comparison cannot prove
The available figures establish structural differences. RIPE NCC distinguishes members from LIRs and publishes eligible-member turnout. ARIN distinguishes Service and General Members. APNIC uses tier-based voting entitlement. LACNIC reports authorised organisations, participating organisations and weighted votes.
They do not prove that one RIR is more representative than another. The decisions, legal environments, service populations and reporting dates differ. A ten-percent organisational turnout under weighted voting is not directly superior to a five-percent turnout under equal voting. Quality also depends on agenda formation, reasons, review and affected interests.
The evidence does not provide a complete five-RIR dataset. AFRINIC's institutional circumstances and election history require separate treatment with current, verified definitions. A common standard should include it, but absence of comparable data should be recorded rather than filled with assumptions.
Nor can the figures determine the ideal electorate. Some institutions intentionally condition voting on activity. Others preserve broad membership rights. Some weight by tier. Measurement reveals consequences; it does not settle normative design by arithmetic.
The present conclusion is narrower: existing headline statistics cannot support broad claims of representation unless the missing transitions and distributions are supplied.
Representation begins where the footnote becomes the table
RIR reports are often careful in footnotes. One member can hold more than one LIR. Vote weight depends on category. Only General Members can vote. Registrations are based on a particular country field. These caveats contain the real constitutional information.
The reform is to move them into the main table. Readers should see the legal unit, account unit, person unit, weight unit and exposure unit before encountering a participation percentage. Every rate should name its denominator in the label.
RIPE NCC's recent analyses provide a strong foundation because they publish eligible members, registrations, ballots, repeated participation and resource bands. The next step is a joined, privacy-protecting representation report that covers contact readiness, operational activity, control concentration and decision type.
The other registries should publish the same ladder using their own constitutional categories. Comparability does not require identical voting systems. It requires identical honesty about units.
Number Resource Society should adopt the ladder at inception. If it claims that operators have clearer standing or that choice improves legitimacy, it should publish service users, eligible principals, representatives, entities, decision weight and operational exposure as separate measures. A new institution is not representative because its architecture sounds participatory.
The basic rule is simple: do not use countries as members, members as accounts, accounts as networks, attendees as voters, voters as ballots, or weighted votes as organisations. Representation cannot be recovered from a pile of true numbers after their labels are removed.
The country denominator is the easiest to misuse
Country counts are attractive because they look public and intuitive. A registry can say that members come from many economies, that entities attended from many jurisdictions or that a meeting included voices from across the region. These statements may be true. They do not measure representation unless the decision itself is territorial and the territorial units have comparable standing.
In number-resource governance, countries are uneven containers. One country may contain a handful of small networks; another may contain hundreds of members, large carriers, cloud regions, public-sector networks and enterprise holders. A jurisdiction with few members may still be strategically important because it hosts critical infrastructure or underserved communities. A jurisdiction with many members may be represented by a small subset of repeat voters. Counting countries cannot reveal any of this.
The country denominator can also hide internal exclusion. If one entity attends from a country, the country appears present even if most affected operators there did not know the issue existed. If a capital-city carrier participates, rural providers may still be absent. If a national registry, ministry or large incumbent speaks, independent operators may have different interests. Geographic presence is a floor for diversity evidence, not proof of representation.
A useful report would therefore place country counts beside eligible members, voting members, active networks, resource relationship, disclosed operator type and participation rate by decision. A country with low voting participation but high operational exposure should be visible. A country with high attendance but low ballot use should prompt a different question: did attendees lack authority, or did the meeting discussion not translate into corporate voting?
Attendance is not authority
Meeting attendance is another useful but limited denominator. People who attend can ask questions, shape discussion and signal concern. But attendance does not mean authority to bind a member, vote in a general meeting or represent customers. A meeting room can be lively while the formal electorate remains quiet. Conversely, a member can vote without attending the deliberative event.
This distinction is central to the difference between community and corporation. Open technical communities often value contribution, persuasion and operational expertise. Corporate associations value legal membership, ballots and formal authority. RIRs live across both worlds. Reporting must therefore show which world a statistic belongs to. A policy meeting attendance figure cannot be used as if it were a shareholder turnout figure. A corporate ballot count cannot be used as if it were an operator-consensus figure.
The danger grows when institutions combine terms such as community, membership and region without denominators. A statement that "the community supported" a change may refer to a mailing-list rough consensus, meeting-room sentiment, board decision, member vote or absence of objection. Each is a different evidentiary entity. Some are valuable precisely because they are open to non-members. Others are valuable because they are legally bounded. Blurring them weakens both.
A representation report should list channels separately: mailing-list entities, meeting attendees, eligible voters, registered voters, ballots cast, weighted votes where applicable, board or committee approvals and affected non-member submissions. The same person or organisation may appear in several channels. That duplication should be acknowledged rather than hidden.
Resource weight can measure exposure and still distort representation
Address holdings, ASN counts and LIR accounts are tempting measures because they seem closer to operational reality than membership. A member with many resources may have more at stake. A large operator may serve more customers. A network carrying critical services deserves attention. But resource weight is not representation.
Historical allocation can dominate resource weight. Legacy holders, incumbents and early entrants can appear more important because they received resources when scarcity rules were different. A cloud platform, access provider and public institution can hold similar resources while facing different continuity duties. A small operator can be disproportionately exposed if its entire business depends on one registry relationship. A large internal enterprise may hold substantial space with little external-customer dependency.
Resource metrics are therefore best used as exposure indicators, not voting substitutes. They can tell a board that a policy affects large holders, small holders, transfer-market entities, new entrants or sponsored end users differently. They cannot by themselves decide who should prevail. If used politically, they can convert past allocation advantage into future governance dominance.
The reporting standard should present resource weight in bands and pair it with role categories. A high-resource internal enterprise, high-resource access provider, low-resource critical-service operator and low-resource reseller are different cases. Aggregated bands protect sensitive data while showing whether a decision was supported by those likely to bear the largest operational consequences.
Repeated participation is a separate legitimacy asset
The recurrent voter is not merely a statistical oddity. Repeated participation creates knowledge. Members who vote and attend regularly understand procedures, personalities, budget cycles and policy history. Their presence can stabilise governance. A system made entirely of occasional voters would have shallow institutional memory.
The danger is to confuse memory with representativeness. A dedicated core can be well informed and unrepresentative at the same time. It can save the institution from apathy while also narrowing agenda. It can detect procedural mistakes while missing costs borne by members that rarely appear. A fair analysis should credit the core without allowing it to stand for everyone.
This means repeated participation should be reported as a cohort. What percentage of eligible members vote repeatedly? What resource and country profiles do they have? Which decisions bring occasional voters back? Are new members joining the core or does it age in place? Are small operators absent from the repeated group because they trust it, cannot afford it or see no chance of influence?
Those questions convert participation data into institutional diagnosis. They also prevent a lazy reform response. If the same core participates because everyone else trusts them, the institution can formalise delegation or representation. If the core participates because everyone else is priced out of attention, the institution must reduce fixed costs and create issue-specific entry points.
A good denominator table should make overclaiming difficult
The final test of representation data is whether it prevents a bad press sentence. A reader should not be able to say "the region supported this" when the data show only a small eligible-voter turnout. A board should not be able to say "members approve" when most affected account holders or operators were outside the ballot. A critic should not be able to say "nobody participated" when a specialised but relevant group did.
The table should therefore be designed against overclaiming. Its labels should be annoyingly precise: eligible legal members, registered voters, ballots cast, weighted votes, meeting attendees, mailing-list entities, disclosed direct operators, disclosed parent or holding entities, sponsored end users noticed, customer-impact submissions and unresolved denominator gaps. Precision makes rhetoric harder.
It also makes improvement measurable. If translated summaries increase participation by newer members, the table will show it. If remote voting helps ballots but not deliberation, the table will show it. If a proposal affects non-members who submit evidence through a new channel, the table will show it. Representation becomes something the institution can improve rather than a word it invokes.
The same table would protect institutions from unfair criticism. If a low-turnout decision nevertheless received strong evidence from the specific operator class most exposed to the change, the record would show that. Good denominators do not only constrain institutions; they make valid legitimacy claims more credible, especially when participation is uneven but substantively relevant to the decision's real operating surface and consequences.
The practical consequence is a more disciplined kind of accountability. Institutions should not be punished for every uneven turnout figure, because voluntary governance will always involve attention gaps. They should be judged on whether they can explain which denominator they used, why that denominator matched the decision, which affected groups were missing and what was done to reach them. That form of disclosure is harder than publishing a participation headline, but it is also more useful.
It lets members, operators, customers and outside observers distinguish thin legitimacy from narrow but relevant participation, and it gives boards a record they can improve rather than a slogan they must defend.
Directory links
- RIPE Network Coordination Centre (RIPE NCC)
- American Registry for Internet Numbers (ARIN)
- Asia Pacific Network Information Centre (APNIC)
- Latin America and Caribbean Network Information Centre (LACNIC)
- African Network Information Centre (AFRINIC)
- Number Resource Society