Summary

  • The current KIXP policy publishes one amount, a Ksh 30,000 joining fee, and hands the recurring charge to the members: the monthly subscription runs "as per level of traffic as agreed upon by members", while the port subscription charges are stated to be the same for all members.
  • Four counts of one membership and one empty field sit on public pages read on 22 November 2026: 162 members on the association's undated homepage, 182 rows counted on the exchange's own roster, 163 networks counted from its published IX-F export, 146 peers on PeeringDB, and zero ASNs on the Euro-IX record above a sub-page that lists the members.
  • The one per-port amount found with a date on it belongs to an Internet Society study of June 2020: US$450 a month for a 1 Gbps port, which divides to US$0.45 per month per megabit. Nothing reviewed prints what a port costs in 2026.

Fifteen documents were read for this briefing on 22 November 2026: the association's homepage and member pages, the exchange's portal roster and policy page, the current and a mirrored older policy text, two PeeringDB records, an Internet Society Pulse tracker page, two Euro-IX IXPDB records, two IX.NET aggregator pages, and two PDF studies. Every figure below is quoted with the document that carries it, and every number I derived is marked as mine.

The clause prints a door fee and hands the meter to the room

The KIXP policies page on TESPOK's domain carries the whole commercial contract a stranger can read, in one sentence of clause 5.6.1: "Members connecting to the KIXP will be registered as TESPOK members of a special category with a joining fee of Ksh 30,000/- and monthly subscription as per level of traffic as agreed upon by members. The port subscription charges are the same for all members."

That sentence does three different things. It fixes an amount, Ksh 30,000, at the door. It declares a property of the tariff — port subscription charges are the same for all members. And it refuses to publish the tariff itself, because the monthly figure is whatever the members agree, scaled by traffic, in a room the reader is not in. The same page prints no effective date, so the one shilling amount cannot be dated to 2026 from the page alone. An older policy text still served from TESPOK's event host repeats the identical joining-fee sentence, lists only 1 Gbit/s and 10 Gbit/s ports, and carries no date and no notice that it has been superseded.

The exchange's own portal is blunter still. Its public policy page describes Layer 2 at 1Gb, 10Gb, 40Gb and 100Gb, names the Nairobi route server as AS37704 present at Africa Data Centers NBO1, Digital Realty NBO1, PAIX and IXAfrica NBOX1, adds a Mombasa presence at Digital Realty MBA1 and MBA2 on AS37786, and prints no amount of any kind. PeeringDB, the register the exchange writes itself, records the same silence in a field: "Service Level Not Disclosed" and "Terms Not Disclosed".

So the meter exists, is uniform by policy, and is unreadable by design. A procurement officer at a Kenyan network can learn the entrance fee from a public page and cannot learn the recurring line from any page at all.

One population, four counts and a zero

Start with the association's front page. It prints "162 Members", "2.7 Tb Peak Traffic" and "222 Connected Ports" in three headline blocks, with no date anywhere near them and no definition of any of the three nouns. Members of what — TESPOK, or the exchange, or both? Connected ports counted as physical ports, as services, or as member-facing connections? Peak traffic over what window, computed by whom? The page does not say, and a homepage is under no obligation to. It is simply the largest number a reader will carry away.

Then the rosters. TESPOK publishes a combined list whose own opening sentence mixes two populations: "Below is a detailed list of TESPOK members and members peering at the KIXP, including AS Numbers (ASNs)." One table, two memberships, six columns — Name, ASN, Location, Website, Peering Policy, Route Server Peer — and, in the returned text, no rows and no total. The exchange's portal roster is better: Member, Joined, ASN, with real dates attached to real networks. AccessKenya Group joined 2018-11-08 on AS15808. ACE CDN joined 2026-10-26 on AS139341. AFR-IX Telecom, whose name field also carries its MANRS status, joined 2024-07-22 on AS60171. Afrinic, MANRS-labelled the same way, joined 2026-06-22 on AS37301. That page has no total row and no date, and it lists the exchange itself and its own route servers among the members, so a raw row count is not a count of independent networks.

The IX.NET aggregator is the only surface that puts the counts side by side. It reports, from a read stamped 2026-11-22 00:24 UTC, "146 In PeeringDB", "163 From IX-F" and "182 From members list". Those three numbers describe the same exchange on the same day and differ by 36 at the widest. The same page prints the port inventory it read from the exchange's IX-F export — 4x100M, 40x1G, 126x10G, 11x40G, 23x100G — and it prints the sentence every reader of African peering should have seen years ago: "We couldn't find KIXP - Nairobi's port fees on its website."

Fourth count, PeeringDB: "Peers 146 Connections 164 Open Peers 117 Total Capacity 3.4T % with IPv6 85", organisation TESPOK, last updated 2026-04-20T10:58:31Z. Fifth surface, Pulse, restates that same register: 3,235 Gbps, 146 member ASNs, 113 of 146 using RPKI, 33 ASNs joined and 1 left in twelve months, with "Data last updated November 2026" and "Last sync from PeeringDB: November 2026" printed beside it. Pulse says what it is doing; 3,235 Gbps is a restatement of cumulative port speeds, not a measurement of traffic, and the page states as much.

And then the zero. The Euro-IX IXPDB record for IX-F ID 49, named "KIXP IXPAFRICA NAIROBI (Kenya Internet Exchange Point)", prints Location Nairobi, Kenya, MANRS True, its own line "Updated Nov. 17, 2026, 4:39 p.m.", and then "Number of ASNs: 0" and "Number of locations: 0". The same database's ASN sub-page for that record lists populated rows, headed by AS42 Packet Clearing House with two addresses and a PeeringDB flag, and continues through AS2906 Netflix, AS3856 PCH, AS6939 Hurricane Electric, AS8075 Microsoft, AS16509 Amazon and AS37704 KIXP Nairobi Route Servers. A record that reports zero networks above a page that names Amazon and Microsoft is not reporting a small exchange. It is reporting an empty or failed import, and no reviewed page says which. A second record in the same database, "Kenya Internet Exchange Point - Nairobi Icolo", IX-F ID 951, was last updated 6 July 2021 and also stands at zero.

Four counts, one zero, no reconciliation. My own subtraction, offered as mine: 162 on the homepage against 146 on PeeringDB is a gap of 16, and 182 on the roster against the same 146 is a gap of 36. Neither gap is an error. Each is what happens when four surfaces count four different nouns — association members, roster rows, exported networks, register peers — and none of them prints its noun.

Two traffic quantities, and a port tally that outruns both

Now the capacity figures, where the arithmetic gets interesting. The homepage prints 2.7 Tb as peak traffic with no window. The aggregator, reading the exchange's own traffic graph, records 604 Gbps peak and 353 Gbps average for the week to 2026-11-21. My division, not anyone's printed ratio: 2,700 divided by 604 is about 4.47, so the undated homepage figure sits roughly four and a half times the only weekly peak I could date. Both could be true of different perimeters. One of them names its week.

The register's capacity is a third quantity entirely. PeeringDB prints 3.4T, Pulse prints 3,235 Gbps, and my subtraction makes those about 165 Gbps apart, roughly 5 per cent of the larger figure — consistent with rounding of one snapshot, and not a disagreement about the world. A second front-end at IX.NET restates the same register as "146 Participants / 3.4 Tbps Connected capacity / 164 Public ports / 4 Facilities", which adds a facility count but no measurement date.

Then take the port inventory the aggregator read from the exchange's export and do what no reviewed page does: add it up. My multiplication and addition, mine alone: 4 × 0.1 + 40 × 1 + 126 × 10 + 11 × 40 + 23 × 100 = 4,040.4 Gbps of provisioned port speed, across 4 + 40 + 126 + 11 + 23 = 204 ports. That sum exceeds PeeringDB's 3.4T and Pulse's 3,235 Gbps of stated capacity, and it is 204 ports against the homepage's 222 connected ports and PeeringDB's 164 connections.

A sum of port speeds is not measured traffic; it is the width of the pipes, and it should normally match or slightly exceed a register's stated capacity rather than overrun it. The gap is explainable — ports decommissioned, ports provisioned but not published, one snapshot older than the other — and nothing reviewed explains it. What can be said plainly is that no reviewed page joins a port tally to a capacity figure to a traffic figure. Three quantities, three publishers, no join.

The only port price with a year on it

The single dated per-port amount found for KIXP comes from outside Kenya's operators entirely. The Internet Society's June 2020 study of Kenya and Nigeria, "Anchoring the African Internet Ecosystem", records a 1 Gbps port at KIXP at US$450 a month — my division puts that at US$0.45 per month per megabit — against US$120 a month at Asteroid, and records the exchange at 56 peering networks and about 19–20 Gbps peak in 2020, up from 25 networks and 1 Gbps in 2012.

That is a good number in a bad position. It is six and a half years old at this dateline, it belongs to a research publication rather than to the exchange, and it describes a port speed the exchange now sells alongside three larger ones. It cannot be quoted as the current tariff, and I am not quoting it as one.

What the 2025 African Telecommunications Union report does with it is instructive. "IXP Status in Africa v5.5", dated 22 August 2025, restates a hedged version — a Kenyan IXP port "might cost US0.45 per month per Mbps for a 1Gbps port" — and attributes it onward to a 2021 data-centre source. By then the figure has been through three hands and lost its month and its publisher, and it is still the only per-port amount in circulation. That is what happens to a price nobody prints: it becomes folklore with a decimal point.

What a stranger can compose, and what they cannot

Ten absences, counted over the fifteen documents named above and read on 22 November 2026:

  1. No reviewed KIXP or TESPOK page defines "members", "peak traffic" or "connected ports", or dates the three homepage figures.
  2. No reviewed page prints a current monthly or per-port amount for a KIXP port with its period, currency, VAT basis and contracting party.
  3. No reviewed page reconciles any two of the membership counts: 162, 182, 163, 146 and 0.
  4. No reviewed page states a measurement window or a computing party for "2.7 Tb Peak Traffic".
  5. No reviewed page joins the port tallies to either traffic figure.
  6. The Euro-IX record prints 0 ASNs above a sub-page that lists populated member rows, and no reviewed page explains the difference.
  7. The member roster prints no total row and no page date.
  8. Full rendered bodies were not obtained for the TESPOK homepage, the TESPOK member table, the IXPDB traffic view and the page-level text of both PDFs, so a definition could sit in material that was not returned to me.
  9. No KIXP participation agreement, tariff schedule or invoice basis was obtained.
  10. Whether the roster's rows include departed or non-connected members is stated nowhere reviewed.

A reader with only these documents can compose the facilities, the route-server ASNs, the four offered port speeds, the joining fee, four membership counts, one zero, two capacities and one dated price from 2020. That is a real picture of a working fabric: two sites, six facilities named across them, four port sizes, a member roster dated member by member, a route server carrying 180,000 IPv4 and 75,000 IPv6 prefixes on its own PeeringDB record. What the same reader cannot compose is the cost of staying.

The exchange is not hiding. It publishes a policy, a roster with join dates, an IX-F export, four port speeds and a route-server ASN. It simply puts the recurring price into the hands of the people who pay it, and then leaves the count of those people to whoever is quoting.

That is the finding, and it belongs to the operator rather than to its critics. A joining fee is a screening device: it tells a new network exactly what it costs to arrive. A negotiated monthly charge tells an incumbent nothing it does not already know, and tells everyone outside the room nothing at all. The duty to make a shared resource legible runs to the recurring line, not only to the door.