Summary
- Safaricom Limited's AFRINIC organisation record lists ten Kenyan IPv4 blocks that sum to a derived 3,448,832 addresses. Two of them are /12s, and together they are 2,097,152 addresses — exactly the one /11 of non-reserved space at which AFRINIC's soft-landing policy says Phase 2 begins (my arithmetic on the published block sizes).
- RIPE NCC's RIS view of AS33771 shows 1,773,568 announced addresses across 107 prefixes, which is 51.4 per cent of the registered estate by my division. A second tracker prints 1,752,576 across 114 prefixes. Two documents, two dates, no reconciliation.
- The IPv6 answer is split inside one organisation handle: about 42.4 per cent of sampled traffic reached AS33771 over IPv6 on 2026-10-14, and 0.1 per cent reached AS37061, the second AS number on the same AFRINIC record.
- AFRINIC has capped a request at a /22 with a 90 per cent utilisation test since 13 January 2020. Eight maximum requests a year come to 8,192 addresses. Safaricom's registered estate is exactly 421 such years.
The register keeps a photograph, not a balance sheet
The AFRINIC RDAP object for AS33771 is the place to start, and it is deliberately unhelpful about size. It names Safaricom Limited, organisation handle ORG-SL70-AFRINIC, country KE, status active, with a registration event stamped 11 April 2005 and a last-changed stamp of 11 November 2025. Its remarks carry routing policy and nothing else: import and export terms with AS12491 and AS12455. Not one prefix is enumerated (AFRINIC RDAP aut-num 33771).
That is not a defect in the page. An aut-num object registers a handle; it does not inventory the address space behind the handle. Twenty years after the object was created, the only AFRINIC-signed fact it gives a reader about Safaricom's estate is that the estate belongs to a company that has existed in the registry since 2005.
The inventory has to come from elsewhere, and the honest description of that elsewhere is a mirror. The organisation index for ORG-SL70-AFRINIC, a third-party reproduction of AFRINIC output, lists two AS numbers and thirteen networks: ten Kenyan IPv4 blocks marked ALLOCATED PA, one Ethiopian block marked SUB-ALLOCATED PA, and two IPv6 /32s (organisation index for ORG-SL70-AFRINIC). The ten Kenyan blocks are 41.80.0.0/15, 41.90.0.0/16, 41.139.128.0/17, 41.203.208.0/20, 105.48.0.0/12, 105.160.0.0/13, 196.96.0.0/12, 196.201.208.0/20, 197.176.0.0/13 and 197.248.0.0/16.
Add them and you get 3,448,832 addresses. That total is mine, not the registry's, and it deserves a warning attached to it: the page does aggregate twelve-plus-one networks, but the sum is arithmetic I performed on published prefix sizes, not a figure AFRINIC, Safaricom or any routing database prints. Where a mirror is the only witness to a block, the block is real enough for a reader to reason about and never strong enough to settle a dispute.
One detail in the mirror is worth keeping, because it shows the page is reproducing something with history in it. The whois range record for 196.201.208.0 - 196.201.223.255 carries netname KE-SAFARI-20041207, maintainer MNT-SAF2004, and a remark that its data was transferred from the RIPE Whois Database on 21 February 2005 (whois record 196.201.208.0/20). The allocation predates the transfer; the record arrived with the paperwork.
Two /12s, one /11, one date
Two blocks in that list are not like the others. 105.48.0.0/12 and 196.96.0.0/12 are 1,048,576 addresses each, so 2,097,152 together, which is one /11 (my multiplication and addition).
Now read the policy. AFRINIC's exhaustion page states that Phase 2 begins when the registry has no more than one /11 of non-reserved IPv4 space available in the final /8, and that Phase 2 has been the current phase since 13 January 2020. Phase 1 began on 31 March 2017. In Phase 2 the minimum allocation or assignment is a /24 and the maximum is a /22 per allocation or assignment (AFRINIC IPv4 exhaustion page).
Safaricom's two /12s are not a metaphor for the pool's last /11. They are the same quantity of addresses, exactly, and the holding predates the ration. In Phase 1 the ceiling had already fallen: AFRINIC's own announcement of 3 April 2017 records that the maximum allocation or assignment became a /13, that the LIR minimum became a /22 and the end-user minimum a /24, that the planning window shrank from twelve months to eight, and that additional blocks required 90 per cent efficient use of everything already delegated (AFRINIC Phase 1 announcement). A single /12 is 128 times the Phase 1 ceiling.
Then came the harder switch. On 14 January 2020 AFRINIC told its announce list that Phase 2 had officially begun, that it would change its systems to limit the size of IPv4 requests that can be submitted, and that every open IPv4 request larger than a /22 would be scaled down to a /22 (AFRINIC Phase 2 announcement). A member arriving after that date cannot ask for a /12 at all. A member who holds one does not have to.
That asymmetry is the whole control surface, and it is worth stating precisely, because it is easy to inflate. Safaricom did nothing improper. Blocks allocated in 2005 and in the years around it were allocated under the rules of their time. The point is arithmetic, not conduct: the same address space that is a routine line in an incumbent's registry record is, for a new entrant, 2,048 separate maximum-sized requests — and each of those requests would need a demonstrated 90 per cent utilisation of everything the entrant already holds before the next one is considered.
The exhaustion page carries no publication date, and it contradicts itself in a way that matters for anyone quoting it: its phase comparison table still reads that Phase 2 has not been reached, while its header and timeline say the opposite. The dates in this article come from the two announcement messages, not from the table.
What the routers see
Registered space and announced space are two different ledgers, and the second one is where a reader learns whether an estate is being used. RIPE NCC's RIPEstat characterises AS33771 as Safaricom Limited, AFRINIC, status ALLOCATED, registration date 11 April 2005, country KE, first seen in RIS on 21 April 2005 announcing 196.201.208.0/24. At a snapshot shown as 2026-11-05 it records 107 announced IPv4 prefixes carrying 1,773,568 addresses, 20 announced IPv6 prefixes equivalent to 65,537 /48s, visible to 100 per cent of the RIS full peers (RIPEstat AS33771).
Divide one by the other and 1,773,568 is 51.4 per cent of 3,448,832, leaving a residue of 1,675,264 addresses (my division and my subtraction). That sentence has to be handled carefully, because it is exactly the kind of comparison that turns into a false accusation if a writer is lazy. The two numbers come from different documents with different observation dates and different purposes. They are not a reconciliation of one record. What they do establish is a direction: roughly half of the registered estate is visible in BGP, and roughly half is not.
Not visible is not the same as not used. Registry space can be held for future deployment, subdivided internally, announced by a downstream customer's AS rather than the holder's, or filtered out of a particular collector's view. The routing record is an observation of what one network of collectors saw on one date. It is not a statement about whether addresses carry traffic, and no document read for this article supports a claim that any Safaricom block is idle, dark, withheld or wasted.
RIPEstat's own top-prefix table shows how concentrated the visible estate is: 59.12 per cent of the announced space sits in 196.96.0.0/12, 29.56 per cent in 105.160.0.0/13, 7.39 per cent in 41.80.0.0/15, 3.7 per cent in 41.90.0.0/16 and 0.23 per cent in 196.201.208.0/20 (RIPEstat AS33771 prefix detail). Two blocks carry nearly nine tenths of everything announced, and the largest of the two is also the block that contains the Ethiopian sub-allocation discussed below.
A second tracker prints a different pair of numbers for the same AS number: 1,752,576 addresses across 114 prefixes and 20 IPv6 prefixes, with 100 per cent of announced routes RPKI-valid and every announced IPv4 prefix geolocated to Kenya. Its principal observed IPv6 block is 2c0f:fe38::/32, subdivided into /48 and /43 more-specifics, alongside 2001:43d0:8::/48 (internetdata.io AS33771).
Two aggregators, two counts: 107 prefixes against 114, 1,773,568 addresses against 1,752,576. Neither is the correction of the other. Both are printed here with their own publisher attached, because a reader who is shown only one of them is being handed a false precision that the underlying measurement does not support. What survives both renderings is the rate: about half.
The Ethiopian block that is exactly one request
The thirteenth network on the organisation index is 196.97.16.0 - 196.97.19.255, status SUB-ALLOCATED PA, netname SAFARICOM-ETHIOPIA-STEP-SUBALLOC, and by the page's own range arithmetic it sits inside the Kenyan 196.96.0.0/12 (organisation index for ORG-SL70-AFRINIC).
That range is a /22, which is 1,024 addresses (my arithmetic). It is a tidy detail for this story rather than a curiosity: a /22 is exactly the largest single allocation or assignment AFRINIC has permitted since 13 January 2020. The Ethiopian step sub-allocation is, in size, one Phase 2 maximum. The Kenyan parent that contains it is a thousand and twenty-four of them.
Read that the other way and the mechanism becomes clear without any claim about intent. A cross-border footprint can be carved out of an inherited holding in a single internal act, and the registry records the carve-out as a sub-allocation. For an operator that arrived after the pool closed, the same footprint would require a request, a utilisation showing, and a queue — and it would arrive at the same size, because that is all the policy now allows at once.
Two AS numbers, two IPv6 answers
IPv6 is the part of this estate where the registry record is generous and the operational record is not. On paper Safaricom holds two /32s, which in registry terms is 2 x 2^96 addresses — a quantity that makes the IPv4 argument look like a rounding difference.
2c0f:fe38::/32 carries netname Safaricom-Kenya, descr Safaricom (KE), country KE, status ALLOCATED-BY-RIR, and it is announced by AS33771 with RPKI-valid status (2c0f:fe38::/32 whois). 2001:43d0::/32 carries netname Safaricom, descr Safaricom Limited, status ALLOCATED-BY-RIR, parent 2001:4200::/23 — address space parentage, not corporate parentage — and the route6 object on the same page lists origin AS37061, not AS33771 (2001:43d0::/32 whois). Both AS numbers sit under the same AFRINIC organisation handle, ORG-SL70-AFRINIC, so the two announcements belong to one registered holder and two routing identities.
The measurement mirror makes that split concrete. It reports Kenya at about 16.7 per cent IPv6-capable and 16.5 per cent preferring IPv6; AS33771 at about 42.4 per cent capable and 42.0 per cent preferring across about 1,099,331 samples; and both AS37061 and Airtel Networks Kenya (AS36926) at about 0.1 per cent (IPv6 capability mirror for Kenya). The as-of date on the mirror is 2026-10-14, and the figures are population estimates derived from a sample, not operator-reported deployment numbers. They are not a Safaricom disclosure and must not be read as one.
What they do show is an asymmetry inside a single organisation handle. One AS number reaches more than four in ten measured samples over IPv6; the sibling AS number reaches roughly one in a thousand. Nothing in the documents read explains the difference, and this article will not invent one. An operator can run IPv6 on one routing identity and not another for reasons as mundane as which platform terminates which traffic; the observation is the fact, the reason is not in the record.
The 2012 promise and the 2026 mirror
Fourteen years separate the two halves of this comparison, and both halves are real.
In May 2012 Safaricom opened its network for IPv6 testing through a partnership with iHub Nairobi, described in trade press as the first mobile operator in Africa to do so, after running IPv6 internally for about a year. The report attributes to the company's then chief technical officer a business-continuity rationale tied to IPv4 scarcity and multi-device households, and states that full IPv6 deployment would remove the need to invest in carrier-grade NAT (TahawulTech, 29 May 2012).
That last clause is the one to handle with care. It is a forward-looking statement by an executive in 2012, reported by trade press. It does not establish whether carrier-grade NAT was later deployed, and no document read for this article says either way. What it does establish is that the strategic logic of IPv6 was understood inside the company before the AFRINIC pool began its final slide — three years before Phase 1, eight years before Phase 2.
Set 2012 against 42.4 per cent on AS33771 and the record reads as a long, real, partial conversion. Set it against 0.1 per cent on AS37061 and it reads as a capability that lives in one part of the estate. Both readings are the same company on the same date.
What a member can still get
The rules are not mysterious, and that is what makes them a control surface rather than a rumour. In Phase 2 a member can request at most a /22 per allocation or assignment, and a member asking for additional space must show at least 90 per cent efficient use of all the space AFRINIC has already delegated to it (AFRINIC IPv4 exhaustion page). AFRINIC also states that if the /12 it reserved out of 102/8 remains unused when the remaining available space has been allocated, that /12 returns to the pool for distribution under Phase 2 conditions, per CPM 5.4.7.2 (AFRINIC IPv4 exhaustion page).
The registry's own staff arithmetic, quoted in a ratified policy proposal, puts the remaining picture at 783,872 IPv4 addresses available at drafting time — roughly two years to deplete at the current delegation rate — plus about three further years for the policy-reserved /12 of around one million addresses. The same document records a hostmaster observation that several Resource Members received more than 8 x /22 within a calendar year by submitting multiple requests up to the maximum, that AFRINIC has recovered more than three million IPv4 addresses since entering Phase 2, and that the proposal would make the soft-landing phase final and irreversible with a twelve-month quarantine on recovered space, presented at AF37 on 24 June 2026 and recorded as ratified by the Board on 25 September 2026 (AFPUB-2026-IPv4-001-DRAFT02).
The 783,872 figure is attributed to staff at drafting time. It is not a dated statistics snapshot and should not be quoted as one.
Put the numbers side by side and the estate's meaning changes character. Eight maximum-sized requests a year come to 8,192 addresses (my multiplication of 8 x 1,024). At that cadence a /12 is 128 years of requests, and Safaricom's ten Kenyan blocks — 3,448,832 addresses — are exactly 421 such years (my division; the product lands on the figure precisely). A company cannot request its way to this estate any more. It can only inherit it, be sold it, or rent it, and no transfer listing or transaction record was read for this article.
The conversion test
Here is the test this article proposes, and it is deliberately dull: measure the estate by what the network sees, not by what the register says.
On that test Safaricom is roughly half-converted in IPv4 and split in IPv6. About 1.77 million addresses are announced; about 1.68 million are held in the register and were not visible in the collector's view on the snapshot date. More than four in ten measured samples reach AS33771 over IPv6; roughly one in a thousand reaches AS37061. Those are the two numbers a reader should carry away, and neither of them is a criticism. They are what a mature incumbent's address estate looks like from outside, in the first decade of a ration that will not end.
The register's half of the story is finished. Safaricom's holdings were settled before the ceiling dropped, and nothing read here suggests they will change by administrative act. The routing half is live: it moves as networks are built, as sub-allocations are announced and as IPv6 is switched on platform by platform. That is why the interesting question about Safaricom's address estate is no longer how large it is. It is how much of it the internet can reach, and how much of the traffic that reaches it can arrive over a protocol the registry stopped counting decades ago (see Safaricom Limited in the directory).
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.






