Summary
- RIPE-826 §5.1 bars a further IPv4 request only once the sum of allocations to one LIR reaches 256 addresses; the registry's own request pages bar anyone who has ever received an allocation, of any size, and add two use conditions the quoted clause does not carry.
- The queue is published in requests — 806 LIRs, one /24 each, front of the queue at 487 days — and the transfer flow in addresses, 2,173,440 in September, while no page among the fourteen I read converts one unit into the other.
On 25 November 2019 the registry stopped selling. "In November 2019, we exhausted our remaining IPv4 pool," RIPE NCC's own run-out page states, dating the final phase to that day, when LIRs became able to enter a waiting list for one /24 from addresses returned in future. Everything a member can still be given from the registry's own pool is 256 addresses, and only after somebody else's space has come back: recovered blocks have their RIPE Database objects deleted and are held in quarantine "for six months or as long as the space is globally routed", according to the recovered-resources page.
The queue that distributes that trickle was last printed with numbers on 23 October 2026: 806 LIRs on the IPv4 waiting list, and 487 days for the first LIR in the queue, beside 20,717 LIR accounts in the same member update. Four hundred and eighty-seven days is about sixteen months, for the operator at the very front. The live waiting-list page draws its two counters, "LIRs in queue" and "Days that first LIR in queue has been waiting", in the reader's browser after the text loads, so the citable figures are the dated ones, and the page states plainly that the graph "should fluctuate over time".
The door to that queue is written twice.
The first version is policy. RIPE-826, published 20 June 2024, §5.1, in its own words: "All allocation requests are placed on a first-come-first-served waiting list. No guarantees are given about the waiting time." The unit is fixed — "The size of the allocation made will be exactly one /24" — and so is the bar: "The sum of all allocations made to a single LIR by the RIPE NCC is limited to a maximum of 256 IPv4 addresses (a single /24). If this allocation limit has been reached or exceeded, an LIR cannot request an IPv4 allocation under this policy." That is a balance. It can be read off a ledger, and a ledger can move in both directions.
The second version is the registry's request guidance. "How Does The IPv4 Waiting List Work?" sets out four criteria, including that "The LIR requesting addresses must not have received an IPv4 allocation from the RIPE NCC previously (including via the waiting list)", that "The allocation must be used to make assignments", and that "The address space must be used on a network with at least one active element in the RIPE NCC service region" (guidance page). The IPv4 landing page puts the same test in a harsher tense — a member may request only if it has "never received an IPv4 allocation from the RIPE NCC before (of any size)" — and says requests are "added to a waiting list and processed when we recover IPv4 addresses in the future". That is an event. It cannot be spent down, because it already happened.
The two texts also differ in verb and in scope. Policy says requests are "placed on" the list; the guidance page says they "are added to the waiting list automatically". The two use conditions appear on the pages and not in the clause as quoted. No page among the fourteen I read states which text governs when a member's history satisfies one test and not the other.
The divergence is not a proofreader's slip, because the two tests answer different questions about the same company. A balance asks what an LIR holds now. An event asks what an LIR has ever done. The registry's own material shows the balance can fall: §5.2 routes returned blocks of a /24 or larger under the §5.1 rules, and recovering space deletes the database objects that recorded it. So a member whose earlier allocation was recovered could present a ledger showing nothing outstanding and still fail a test phrased as biography.
My reading of those two texts is that this case exists; the pages do not tell me which reading the registry applies, and I make no claim about how any application has been decided.
One more sentence sits three clicks from the queue. The landing page says an IPv4 request "generally takes around one working day to complete". The interval that decides an applicant's outcome is not the one working day; it is 487 days for the first LIR in line, and no reviewed page prints an interval for anyone behind it. Both statements are current, both belong to the same registry, and no page I read reconciles them.
Then there is the counting.
The queue is published in requests. The market that moves the same scarce space is published in addresses. On 23 October 2026 the same page that printed 806 LIRs printed "IPv4 addresses transferred in September: 2,173,440". Run the two against each other and the units part company. Eight hundred and six requests at exactly one /24 each is 206,336 addresses of outstanding demand — my arithmetic from published figures, not a figure any reviewed page prints.
Two million one hundred seventy-three thousand four hundred and forty addresses is 8,490 /24-equivalents, roughly ten and a half times the entire queue — the whole queue's outstanding demand, at September's pace, in under three days. That comparison is mine. No reviewed page makes it, and the registry could not have made it from its own headline numbers alone, because no page in the set prints how many transactions made up September.
What that monthly figure includes is also unstated. The page gives no definition, no list of included or excluded transfer categories, no statement whether intra-RIR and inter-RIR transfers are combined, and no statement whether it counts transfers registered or transfers completed. It is a bare total.
That absence is not a limit of the registry's ability. The same registry publishes the rows. Its transfer-statistics index states that "the RIPE NCC publishes a list of all Internet number resources transferred according to the policy", and defines what each column means: Original Block, Transferred Blocks, From, From Country Code, To, To Country Code, Transfer Type, Date. Date is "Date on which transfers were processed". Transfer Type covers policy transfers or changes to business structure such as a merger or acquisition. The intra-RIR IPv4 tables carry all eight fields. The inter-RIR tables split incoming from outgoing by RIR and drop the Transfer Type column altogether, and the page states that only valid transfers are made public.
And it has counted before, more than once. The 2018 annual report printed a count beside a total: "IPv4 TRANSFERS 2,870 12,247,168 IPv4 addresses", and separately "155 IPv4 INTER-RIR TRANSFERS 2,899,712 ADDRESSES". In 2015 the registry's own analysts converted an address total into another unit: 14,526,720 addresses, "equivalent to 86.6% of a /8", between 17 October 2012 and 1 May 2015, with 80% of single transfers falling between a /24 and a /20 and a chart of transfers per month (2015 analysis). In 2017 the same channel published the scope — allocated PA, assigned PI and inter-RIR transfers in, legacy-resource transfers out (2017 analysis). The 2025 State of IPv4 report charts blocks per month rather than transfers, and says so (State of IPv4). Counting, scoping and converting are all inside this registry's demonstrated range. What the monthly series does not carry is the count beside the total, and the conversion into the unit the queue is published in.
So consider what an operator can actually do with these pages, because that is where the cost lands. It can verify a single transfer: parties, block, processing date. It can read the policy's allocation unit and its ceiling. It can compute a labelled ratio and check my arithmetic. It cannot learn how many transactions made up September, what that total includes, which of the two eligibility texts binds its own history, or how long a request filed today would wait. It can price the market and trace the queue, and it still cannot subtract one from the other.
The queue is a fact about the past; the transfer figure is a fact about a flow; and the row that would join them is published nowhere I read.
Yesterday BTW read this queue and found its rate missing — how fast recovered space leaves the line. This piece reads the door and the units instead. The finding is not concealment. Every row is public, and I allege no wrong figure and no favour to anyone. The finding is that an accurate ledger is not yet a reconcilable record. Two fields would close it, and this registry has printed both before: a transaction count beside the monthly address total with a one-line scope, and one canonical statement of the eligibility gate that says which test governs and whether a recovered allocation resets the balance or stands forever as the event.
That is a recommendation about a record, not a claim about conduct.
Policy text, published by the registry, states of its own queue: "No guarantees are given about the waiting time." That sentence is honest, and it is not the whole story. A guarantee is not merely withheld here; the arithmetic that would let a member state one — the count, the scope, the precedence, the interval behind the front of the line — is not assembled on any page I read. The RIPE NCC IPv4 Pool entry tracks the entity that operates this instrument. What the instrument publishes is an inventory, a queue and a flow, each true, each in its own unit, and none of them answering the other.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






