Summary

  • APNIC's public member directory prints Singapore Internet Exchange Limited in the row | Singapore Internet Exchange Limited | VERY SMALL | SG |, a category the registry describes as corresponding to the number of IP address resources an organisation holds.
  • The operator's own site states that "As Singapore's largest not-for-profit Internet Exchange, SGIX provides a robust, open, and neutral peering platform for all its peering members" — a claim about a fabric, not a measurement of traffic.

A row and a sentence

On 20 October 2026 the APNIC public member directory carried a single table line for this company: the name, the category VERY SMALL, and the economy code SG. On the same date, reading the operator's own site returned one sentence describing the business: "As Singapore's largest not-for-profit Internet Exchange, SGIX provides a robust, open, and neutral peering platform for all its peering members."

Set beside each other those two statements read like an argument. They are not. They are answers to different questions, filed by different institutions, about different populations. The useful work is not deciding which one to believe. It is noticing that the industry's most-quoted descriptors of an exchange — a registry category on one side, a superlative on the other — are both silent on the thing an operator, a customer or a regulator most wants to know.

What a registry category counts

APNIC describes its membership tiers as corresponding to the number of IP address resources each organization has rights over, as reported in the same retrieval that produced the row. The retrieved text does not print a numeric band beside each label, so the category is best read as an ordinal position within the registry's own scheme rather than a threshold anyone outside can compute against.

That matters because of what an exchange is. A peering exchange is a switch fabric. The addresses its participants use belong to the participants — the access networks, carriers, content networks and enterprises that connect to it. An exchange that holds few or no addresses of its own is the ordinary case, not an anomaly, and it is exactly the shape that produces a small registry category alongside a busy fabric. The tier is an administrative fact about one member's registry footprint. It is not a participant count, it is not a port count, and it is most certainly not a traffic figure.

BTW's own directory entry for the company, which is where this reading began, says as much in its own words: it records a registry member and resource-holder footprint and states that this is not proof the entity sells ISP, IP transit, cloud, registry or managed-network services. The entry can be read in full at the BTW directory.

What a fabric claim counts

"Largest" is a relational word, and on the retained page it arrives without a comparison set. It could mean the most participants, the most connected capacity, the widest footprint across buildings, or the greatest volume of traffic exchanged; the sentence does not choose. "Not-for-profit" speaks to corporate purpose. "Neutral" speaks to how the switch is governed — who may join, on what terms, and whether any member's traffic is treated differently from another's. All three are structural and relational descriptors. None of them is a rate.

This is not a criticism of the sentence. A front page is a positioning surface, and an exchange that described itself in gigabits per second would be making a claim it could not keep current. The point is narrower and more useful: the two published labels that a reader is most likely to encounter about this company cannot be used as proxies for each other, in either direction.

A difference of strings, not of status

The operator's site footer prints "Copyright © 2026 Singapore Internet Exchange Ltd. All rights reserved". The directory row prints "Limited". That is a difference between two published strings, and on its own it establishes nothing about corporate form, incorporation, shareholding or control. A company-registry extract would settle those questions; the two strings will not, and neither will the directory card's own summary. Readers who need the legal position should treat it as open rather than as settled by the shorter spelling.

Where a traffic number would sit

The retrieval of the operator's site reports that its services navigation carries sections for Traffic Statistics, Port Price, Participants, Infrastructure and Service Level. That is a summary of structure rather than quoted page text, and the contents behind those sections were not returned. The distinction is worth holding precisely, because it locates the missing quantity instead of merely regretting it.

A port price list answers what connectivity costs at a given speed. A participants page answers who is present. An infrastructure page answers which buildings the fabric reaches. A traffic statistics page is the only one of the four that can answer how much is actually exchanged — and only if it states its method. A peak figure in bits per second at a named percentile, a daily maximum, and a total of contracted port capacity are three different quantities that a casual reader will collapse into one number. Two of them describe what the fabric could carry. Only the first describes what it did.

The boundary of this reading

What follows is scoped to documents rather than to the world. The reading retained for 20 October 2026 comprises the APNIC member directory row, two retrievals of the operator's front page, and one retrieval of the IMDA portal's front material. It did not obtain the operator's Traffic Statistics, Port Price or Participants content, any company-registry extract, or any gazetted data-centre capacity instrument. That is a statement about four documents reached on one date. It is not a finding that such material is unpublished, and it is not a finding that traffic is absent, flat or declining.

Why a capacity constraint cannot do the explaining

A version of this question that circulates widely runs like this: Singapore restricted data-centre capacity, land and power in the city-state became scarce, alternatives grew across the water in Johor and Batam, and therefore the shape of domestic interconnection changed. Each step of that argument is plausible. The first step, however, did not survive this reading. A retrieval of the IMDA portal surfaced digital-economy and regulatory material but no instrument text imposing or describing a constraint on data-centre capacity, and without the instrument there is no effective date, no defined constrained resource, and no period against which any traffic series could be aligned.

Even with the instrument in hand, the inference would still not run. A restriction on building bounds expansion; floor area and power measure physical stock; port speed and contracted ports measure carrying capacity; throughput over a stated interval measures realised exchange. These are four layers, and only the last one answers a question about deepening. A constraint on the first layer has an ambiguous effect on the fourth: it can push traffic onto fabric that already exists, which would raise measured exchange, or it can suppress new deployment, which would lower it.

The same policy can produce either result depending on demand, routing and who happens to be in the building. Direction is not derivable from the rule.

The test that would settle it

The question of whether domestic peering deepened is answerable, and it is worth stating what would answer it in public rather than leaving the standard implicit.

A peak or aggregate exchanged-traffic figure published by the exchange for a named period, with its method stated — bits per second at a named percentile, or a daily maximum — settles volume for that period. A participant count published at two dates against a stable definition settles direction without settling volume, and must be reported as direction only. Offered and active port sizes with counts at comparable dates settle capacity, which is not exchange. A member-level port-utilisation disclosure, or a peering statistic collected by a regulator, could address domestic scope and realised load together, which none of the others can.

On the other side, the items that would not answer it are as important. A registry tier ranks address holdings. A superlative ranks nothing measurable. A price list states what capacity costs. A single current utilisation snapshot, without a comparable baseline, describes a moment rather than a trend. And a company-registry extract would settle the corporate question without touching the traffic one, because they are unrelated.

What is actually established

Two institutions published two true-sounding statements about one Singapore company, and the statements measure different things. That is the finding, and it is narrower and more durable than a verdict about the market. It also produces a specific and falsifiable consequence: if any dated document appears that states exchanged traffic for this fabric with its method, this account is superseded for that period and should be replaced rather than defended. If a document appears that states member and port counts at two dates, the account gains direction and still lacks volume.

If a gazetted instrument appears with effective dates, the constraint premise becomes establishable and the causal question becomes askable rather than assumable.

Until one of those documents is in hand, the honest answer to the question that prompted this reading — did domestic peering deepen in Singapore while capacity policy held the floor — is that the retained record does not determine it. That is a different answer from "it did not", and the difference is the whole of the discipline. The next reader who wants the real number knows exactly which page to ask for, and which sentence on that page would make the number usable.