Summary

  • LACNIC's own record calls IPv4 exhaustion on 10 June 2014 and again on 19 August 2020, and only the second of those dates changed where the addresses come from.
  • A newcomer's ceiling is one block between a /24 and a /22 — 256 to 1,024 addresses — obtainable after an IPv6 assignment and locked against transfer for three years.
  • The wait for that block has been printed as about 7 years, about 9 years, more than 10 years and at least 18 years in four LACNIC documents since July 2023, and no document in this reading prints the recovery rate behind any of them.
  • Transfer is the route that moves volume, and its published series rises 9% and then 25% when counted in blocks while the address counter falls about half and then holds flat.

The word exhaustion, printed twice

On 10 June 2014 LACNIC announced the exhaustion of its IPv4 address pool, and it announced it with a number: the pool of available IPv4 addresses had reached the mark LACNIC printed as 4.194.302, using periods as thousands separators. That is a /10. The same release states what the exhaustion triggered — the Gradual Exhaustion policy, the New Entrants policy and the intra-regional IPv4 transfer policy at section 2.3.2.18 — and what happened to the space that remained: of the 4,194,302 addresses, 2,097,150 could be assigned during that phase in blocks of 256 to 1,024 addresses, with further requests only after six months, and once that space was used a /11 reserve for new members would activate as Phase 3, in which a new member could receive one assignment of between a /24 and a /22 (LACNIC, 10 June 2014).

One piece of arithmetic belongs here, and it is mine: a /10 is 4,194,304 addresses. LACNIC printed 4,194,302, and it printed 2,097,150 as the assignable half where half of 4,194,304 is 2,097,152. No document in this reading explains either two-address difference, so it stays unexplained rather than being tidied into a number that no source prints.

Then, on 19 August 2020, LACNIC announced that "the last available IPv4 address block has been reserved" (LACNIC). The same announcement sets out the sequence: the final phase began in February 2017; LACNIC and the National Internet Registries assigned more than 5.6 million IPv4 addresses during that phase; recovered IPv4 addresses must pass through a six-month quarantine, with the first quarantined blocks released in September 2020; approved requests are served according to their position on a waiting list active as of that date; and LACNIC had assigned 189.3 million IPv4 addresses to more than 11,200 organizations in the region over its lifetime.

LACNIC restated the sequence itself that October: the final exhaustion phase was triggered in February 2017; during Phase 3 assignments were limited to a maximum /22 and a minimum /24 and were available only to new members; the last available block was reserved on 19 August 2020; recovered space passes through a six-month quarantine (LACNIC, 7 October 2020). Five years later the registry put the same fact in the present tense: "Since 19 August 2020, LACNIC's IPv4 address pool has been formally exhausted. Currently, assignments are only possible from addresses that are returned or recovered, and are managed through a waiting list" (LACNIC, 12 August 2025).

Two dates, one word, two different events. The 2014 date changed which policy governs allocation while space remained. The 2020 date changed the source of the space itself. Every planning sentence that says "after LACNIC exhausted its IPv4" is quoting one of those two moments, and the words around it usually do not say which.

The ceiling is a policy object

What a new entrant can obtain is not a market outcome and not a price. It is a number written into policy text, and it is small enough to state in one line.

LACNIC's web rendering of section 11 of the Policy Manual — carrying its own warning that the PDF manual is authoritative and that this version may lag it — describes a special IPv4 reserve for new members built from post-exhaustion IANA space plus recovered and returned space. Under item 11.1, no IPv4 is distributed from that reserve to organizations that already hold IPv4 from LACNIC or its predecessors, except for critical infrastructure; allocations or assignments run from a /24 up to a /22; it is one block per organization; blocks received may not be transferred for three years; and an applicant with no LACNIC IPv6 block must request one (LACNIC, section 11 as published on the web). Item 11.2 creates a separate /10 reserve for gradual exhaustion, allocated in the same /24-to-/22 units, and item 11.3 directs that space distributed by IANA to LACNIC after 11.2 took effect may be distributed only under 11.1.

The reserve itself was resized once. LAC-2013-3 enlarged both reserves from a /12 to a /11, was ratified on 25 March 2014, and reasoned that /12 reserves were too small for a smooth transition to IPv6 in the region (LAC-2013-3). A later instrument, LAC-2019-9, sets the order in which returned capacity is handed out: IPv4 resources recovered under that policy may only be assigned under the guidelines of item 11.1; assignment of recovered resources begins when, for the first time, an assignment cannot be made from the special reserve pool; and resources with an earlier recovery date are assigned first (LAC-2019-9).

LACNIC's own request page closes the loop in plainer language. It states that the registry is "currently in IPv4 exhaustion Phase 3"; that in Phase 3 LACNIC will only assign blocks equivalent to a /24 up to a /22; that an IPv4 request enters the waiting list only after the organization has received an IPv6 assignment; and that organizations based in Mexico or Brazil must request resources from their respective National Internet Registry (LACNIC, obtaining IPv4 and ASNs).

So the ceiling, stated once: a new entrant's IPv4 outcome from LACNIC is one block of between 256 and 1,024 addresses, held under a three-year transfer lock, obtainable only after an IPv6 assignment. Across the documents read here there is no printed route to a second block from the reserve, critical infrastructure aside. That is a policy object with a date and a section number, and it is the thing to quote when someone asks how much space a new Latin American network can still get from its registry.

The line, as an instrument

The waiting list is not a metaphor and not an estimate. It has a creation date, an eligibility rule, a queue discipline and a maximum delivery.

LACNIC's English page describes the list as created on 19 August 2020, when the last available IPv4 address block was assigned, to establish an order among organizations requesting IPv4 addresses. To be included, an organization must be a LACNIC member and must already have been assigned IPv6 resources. Only initial requests are accepted, and they are evaluated under section 11.1 of the Policy Manual. Requests are handled first-come through LACNIC's ticket system; a position may change if an older ticket is approved; positions are not transferable and are lost in the event of purchase, sale or merger. Blocks assigned from the list have been quarantined for at least six months, and the recipient is responsible for rehabilitating the addresses. The maximum deliverable is 1,024 IPv4 addresses (LACNIC waiting list). The Spanish page adds a structural detail that changes every regional total: NIC.MX in Mexico and NIC.br in Brazil administer their own tickets independently of LACNIC (LACNIC, lista de espera).

Read that list of conditions as a mechanism rather than as a queue. Membership is a precondition. IPv6 assignment is a precondition. The request must be initial, which excludes an organization that already holds LACNIC IPv4. The position cannot be sold, and it does not survive the corporate transaction that changes who owns the applicant. The space arrives quarantined and needs rehabilitation. And the ceiling holds at 1,024 addresses — the same /22 that section 11.1 prints as its maximum.

Two consequences follow without any inference about anyone's conduct. First, the list is a queue over one block per organization, so its length and its wait describe how long it takes to hand out 256 to 1,024 addresses at a time. Second, the position is attached to the legal entity that holds it, which makes the queue an asset with a shape that cannot be transferred and that disappears in a merger.

Four numbers for one wait

Here is where the reading stops being tidy. Since July 2023 LACNIC has printed a waiting time in four documents, and the four do not agree.

Four numbers for one wait
Printed estimate Document Date of that document
about 7 years for an entrant then joining; last entrant projected 2030 LACNIC blog (es, en) 17 July 2023
about 9 years for new entrants LACNIC 2024 annual report (es) reporting year 2024
more than 10 years LACNIC blog (12 August 2025) 12 August 2025
at least 18 years for the last request on the list current waiting-list page (en) undated on the page

The instrument and the projection are different objects, and the difference is worth holding still. The 18-year figure on the waiting-list page is described there as resting on "historical resource recovery patterns", and the page calls the estimate variable. That is the registry telling a reader what kind of number it is: a projection from then-current recovery behaviour, not a measurement of a wait that anyone has completed.

No document in this reading prints the recovery rate — addresses recovered per month, or per year — that would let a reader recompute any of the four figures. Without that rate the estimates cannot be checked against each other, only dated. And a date is exactly what two of them lack or carry loosely: the waiting-list page is undated on its face, and the 2024 annual report gives a reporting year rather than a publication date.

What the documents do print is a headcount, twice, with a date attached to each. 820 organizations were on the list in August 2022 and 1,070 in July 2023, according to the July 2023 blog (LACNIC). The 2024 annual report states that the list grew 22% in 2024 while printing no headcount (LACNIC 2024 annual report). A percentage whose base is not printed cannot be applied to a count printed a year and a half earlier without inventing the base, so it is not applied here.

The one completed wait

Among all these projections there is exactly one completed-wait figure in the material read, and it is small enough to be quoted whole: organizations then receiving recovered space had waited on average about 825 days (LACNIC, 17 July 2023).

That number and the four projections are not the same kind of quantity, and the gap between them is the finding. About 825 days is what the registry says happened to organizations that received space. About 7, about 9, more than 10 and at least 18 years are what the registry says will happen to organizations still waiting. A planner who quotes the second set as though it were the first is quoting a forecast in the grammatical position of a measurement.

There is a reason to be careful about the direction of the series rather than only its size. The July 2023 blog projected the last entrant at 2030 on the strength of the queue as it then stood. Two later documents moved the number up, and the current page moves it further. A wait estimate that lengthens as the queue lengthens is behaving the way a queue model should; it is not evidence of a contradiction inside the registry. But it does mean that the number published in July 2023 has already been superseded twice by LACNIC's own later documents — and that any figure older than the document in front of you has a successor.

The counter that reverses the trend

Transfer is the other route, and it is the only one in this material that moves real volume. It is also the one whose published direction depends on which object is being counted.

Start with the policy. LACNIC Policy Manual v2.18, dated 27 April 2023, at section 2.3.2.18, permits IPv4 block transfers between LIRs and/or End Users and applies both to inter-RIR transfers and to transfers inside the LACNIC region. The minimum transferable block is a /24; no maximum transfer size is stated. A receiving organization inside the region must first justify its IPv4 resources before LACNIC; a recipient in another region is subject to that region's RIR criteria, verifications and requirements. Intra-RIR transfers require a signed copy of the legal document supporting the transfer from both organizations, and inter-RIR documentation is agreed between the RIRs. LACNIC verifies the holder of the resources and that the resources are not involved in a dispute. LACNIC keeps a publicly accessible transfer log recording the transaction date, the originating organization, the receiving organization, the transferred addresses and, for inter-RIR transfers, the source and destination RIRs. The transferring organization becomes ineligible to receive IPv4 allocations or assignments from LACNIC for one year from the transaction date; transferred addresses may not be transferred again, in whole or in part, for one year; addresses from LACNIC allocations or assignments may not be transferred for three years from the allocation or assignment date; and legacy resources transferred into the LACNIC region cease to be considered legacy resources (LACNIC Policy Manual v2.18).

The regime has its own paper trail, and one detail in it matters for anyone quoting the clause. LAC-2019-1 was ratified on 20 June 2019, implemented on 21 July 2020, and covers inter-RIR as well as intra-RIR transfers (LAC-2019-1). The 2014 release already cites section 2.3.2.18 for intra-regional transfers, which means that section number has carried more than one text in its life. The version read here is the 2019–2020 regime. Two conditions therefore sit on a reader: the section number alone does not date the rule, and the manual read is v2.18 of April 2023 — no later consolidated manual was obtained for this reading, so every transfer condition above belongs to v2.18 and a later revision could have moved a term. That possibility is left open rather than asserted in either direction.

Now the counts, and the stall they produce in any single-line summary.

For 2023 the English annual report says: "During 2023, a total of 159 address blocks were transferred, 9% more than the total number of transfers completed in 2022. However, the number of transferred IP addresses was significantly lower: approximately 50% of those transferred in 2022." LACNIC attributes the lower transferred-address volume to two 2022 corporate reorganization transactions between companies in Chile and Brazil involving more than 250,000 IPv4 addresses transferred to RIPE NCC. The same report states that 78,848 IPv4 addresses were assigned in 96 IPv4 assignments during 2023 (LACNIC 2023 annual report).

For 2024 the Spanish annual report says: "Durante 2024 la cantidad de bloques transferidos fue 197, un 25% superior al total de transferencias realizadas en 2023. El incremento en la cantidad de transacciones, sin un aumento en el volumen de IPs transferidas, obedece a una mayor cantidad de bloques /24 involucrados en las transacciones en relación con años anteriores." The same report states that 95 IPv4 blocks were assigned, equivalent to 77,824 addresses, and 659 IPv6 assignments — "por cada asignación de IPv4, otorgamos casi 7 de IPv6" (LACNIC 2024 annual report).

Blocks and addresses are different counters, and read separately they tell opposite stories. In blocks, transfer activity rises 9% and then 25%. In addresses, 2023 falls to about half of the 2022 volume and 2024 stays flat while block count grows a quarter. LACNIC explains both: the 2023 address drop traces to two large corporate reorganizations that moved more than 250,000 addresses out of the region in 2022, and the 2024 volume stayed flat because the added transactions carried more /24 blocks rather than bigger ones. Neither statement contradicts the other. They are counting different objects, and the same registry printed both.

Two pieces of arithmetic here are mine, and both are labelled because neither appears in the reports.

First: 197 ÷ 159 = 1.239, so the two printed block counts differ by about 23.9%, while the 2024 report states 25% above the 2023 total. The 1.1-point gap cannot be reconciled from the documents read — a restated 2023 base would explain it, and no restatement is printed — so it is recorded rather than resolved.

Second: 78,848 ÷ 96 = 821.3 addresses per IPv4 assignment in 2023, and 77,824 ÷ 95 = 819.2 in 2024. Both averages sit between a /24 of 256 addresses and a /22 of 1,024, which is consistent with a mixed population inside that band. Looking at the same two printed totals against the /22 unit is the sharper version: 78,848 ÷ 1,024 = 77 exactly, and 77,824 ÷ 1,024 = 76 exactly. Both annual assignment totals divide into whole /22s — 77 and 76 — which is consistent with the 2024 report's own word, "equivalentes", and is not a claim that either year consisted of uniform /22 blocks.

Neither report prints the block-size distribution behind its totals, so the mix cannot be reconstructed here, only bounded.

One more division, on the 2020 announcement: 5.6 ÷ 189.3 = 3.0%, using the printed floors, since the release says "more than 5.6 million" against 189.3 million lifetime. On LACNIC's own numbers, the final exhaustion phase from February 2017 to August 2020 delivered about 3% of the addresses the registry had distributed in the region since it began. That is my division of two printed figures, and it describes a phase, not a year.

Three perimeters, and a page that prints nothing

Every total above sits inside a perimeter, and confusing two of them is the easiest error available in this material.

The 2024 annual report presents its work as carried out together with the National Internet Registries, so its totals describe LACNIC plus NIC.br and NIC.MX — not LACNIC alone. The Spanish waiting-list page confirms that those two NIRs administer their own IPv4 tickets independently (LACNIC). A LACNIC-only figure and an annual-report total are therefore different perimeters, and any comparison across them has to say so.

The registry statistics portal is the third surface, and it is the one that does not answer. LACNIC's statistics site states that it presents the assignment and distribution of Internet number resources since LACNIC's creation in 2002, and that IPv4, IPv6 and ASN assignments over time can be visualised (LACNIC registry statistics). The retrieved text does not expose chart values, units, or whether NIR data are included. No number is taken from it in this article, and that absence is a reading limit, not a finding about the registry.

There is a larger perimeter question that no figure in this material answers at all. Assignment is not use. Nothing here measures whether any assigned or transferred address is announced, routed or in service. Every number above is a registry act: a block handed over, a transfer logged, a position in a queue. An organization reading this material for procurement should hold that distinction firmly, because a registry count cannot tell it whether the space it obtains will work on any network it does not operate.

Finally, the entity itself, for readers who want the record behind the argument: LACNIC in the BTW directory.

What a capacity plan has to carry

The practical conclusion is narrow, and it is the reason this material is worth assembling into one place.

A 2026 capacity plan that needs Latin American or Caribbean IPv4 has three routes and no fourth. It can enter the waiting list, where the deliverable is one block of 256 to 1,024 addresses, the queue position is not transferable, and the position dies with a purchase, sale or merger. It can obtain space by transfer, where the minimum block is a /24, the receiving organization inside the region must justify need, the transferring organization loses access to LACNIC IPv4 for a year, and transferred addresses are locked for a year. Or it can already hold the space it needs.

Every number the plan quotes about those routes should carry three labels beside it: the unit it is counted in, the date of the document that printed it, and the document that attests it. In this material, each of those labels changes the answer. The unit decides whether transfer activity is up 25% or flat. The date decides whether the wait is 7 years or 18. The document decides whether the figure is LACNIC alone or LACNIC with two NIRs — and whether the sentence carrying it is a policy ceiling, a queue instrument, a projection, or a count of registry acts.

The exhaustion headline is true and it is not the planning instrument. The instruments are a policy section that caps a newcomer at a /22, a queue whose estimate has been rewritten three times in two years without a published recovery rate, and a transfer clause with a three-year lock on allocated space and a one-year lock on whatever moves. A plan that quotes any of those without its unit, its date and its custodian is quoting a direction that another counter reverses.