Summary

  • The operator's own homepage presents "Aggregate Switching Statistics" and "Number of Route Prefixes Announced" as two distinct headings, and sorts participants into four classes: "R&E Participants", "FTNS Providers", "Resellers", and "Satellite and Reseller Sites". No figure, count, port speed, site name, method statement or measurement window accompanied those labels in the retrieved text.
  • Five document retrieval attempts on 17 October 2026 — twice on the traffic-statistics page, twice on the participant list, once on a third-party exchange database record — returned zero readable bodies. That is a dated evidence boundary, not proof the data is absent from the public web.
  • A single "HKIX throughput" or "HKIX participant count" is therefore a category claim rather than a measurement: switching volume, announced prefixes, member relationships, connected networks and physical ports are different quantities, and the operator's own taxonomy is the clearest evidence they should not be added together.

What the operator actually publishes

The retrieved text from the exchange's homepage is a navigation surface, and it is informative precisely because of how little it quantifies. It presents two measurement headings side by side — "Aggregate Switching Statistics" and "Number of Route Prefixes Announced" — and it presents participants under four separate labels: "R&E Participants", "FTNS Providers", "Resellers", and "Satellite and Reseller Sites" (operator homepage).

That is the whole of what could be read. No throughput value, no prefix count, no port speed, no facility name, no methodology note and no measurement window appeared in the retrieved text. The labels are real and they are the operator's own words; the quantities they name were not obtained.

Why two headings are not one number

Switching volume and announced prefix count answer different questions and can move in opposite directions. Aggregate switching statistics describe how much traffic traverses the fabric. Announced prefixes describe how much of the global routing table is visible at the exchange — a reach measure, not a volume measure. A fabric can carry growing volume from a shrinking set of participants, or host a growing participant list carrying little traffic.

The standard public comparison uses a third metric again. The list of internet exchange networks by size is compiled "measured by peak data rate (throughput), with additional data on location, establishment and average throughput" (comparative list). Peak and average are separated there, and such lists are non-exhaustive because they include only exchanges that publish a figure at all.

So a reader comparing Hong Kong against Amsterdam, Frankfurt or Singapore is comparing peak against peak. If no peak figure is obtainable for one exchange, that exchange does not become small — it becomes absent from the comparison. Absence from a ranking is not a measurement of low traffic, and a directory listing is not a measurement of high traffic.

Why four classes are not one participant count

The four participant labels do more analytical work than the missing numbers, and they are the usable part of the published surface.

A member is a contractual relationship with the exchange. A connected network is an operational presence: a network that has configured a session and is exchanging routes. A port is a physical, capacity-bearing unit — optics and a switch port at a stated speed. These are three different things, and the operator's own taxonomy implies that presence is not uniform. Research and education networks, fixed telecom network service providers, resellers, and satellite and reseller sites occupy different roles in the interconnection market and bring different traffic profiles, contract terms and failure modes.

A single participant count collapses all of that. It cannot tell a prospective member whether the networks it needs to reach are present, whether they hold meaningful capacity, or whether the relationship is direct or intermediated through a reseller. The classification is the operator's own admission that these are separate categories; per-class counts are what would turn the admission into an answer.

What the registry record does and does not establish

The directory record is an APNIC public member directory footprint: it lists The Hong Kong Internet Exchange Limited as a medium member with country code HK, last updated on 18 June 2026. That is a statement about registry membership and resource holding, not about what the exchange operates.

The record is explicit about its limits. Its stated data gaps call for the legal registration record, public website and leadership contacts to be verified from current official or company-registry sources, and for ASN and IP resources to be mapped from RDAP, WHOIS, IRR, BGP or RIR delegation data "before making any routing-footprint claim beyond APNIC membership". It also warns against classifying the entity as an ISP, IP transit, cloud, registry or managed-network provider unless a current public source explicitly supports that service.

On legal form, the record's own field asserts a private company registered in Hong Kong. No incorporation document, certificate or company-register extract was retained in this research window, so that is reported as the directory's characterisation, not as an established fact. The distinction matters: who holds the legal duty to operate the fabric is a different question from who holds a number-resource membership, and only one of them is documented here.

No ASN or IP resource map was obtained. Routing footprint beyond APNIC membership is unestablished.

What the exchange does not operate

An exchange point sells a meeting place, not a route. Its measurable surface is ports and fabric throughput. Its value to a carrier or content network depends on two things it does not fully control: which other networks choose to be present, and the cross-border capacity those networks hold outside the fabric.

Hong Kong's regional role has always rested on the second. A local exchange can publish rising switching volume while the international links that make the location useful are provisioned, priced and constrained elsewhere — by submarine cable operators, by transit providers, by the networks that decide where to land traffic. When that capacity tightens, local fabric statistics need not fall; traffic can re-route within the city while the city's reach narrows.

That is why the missing numbers matter less than the missing decomposition. Even a published aggregate would not tell a reader whether the location's relevance is growing. The questions that would are specific: how many distinct networks are present in each class, what port speeds they hold, how many announced prefixes are visible, and what share of local traffic stays local rather than transiting abroad.

The evidence boundary

On 17 October 2026 this research made five attempts to read the underlying documents: twice on the traffic-statistics page, twice on the participant list, and once on a third-party exchange database record (traffic statistics, participant list, exchange database record). All five returned zero readable bodies. No throughput figure for any period, no participant count, no site or facility list and no fee schedule was obtained.

That is a statement about one research window, not about the public web. The pages may be readable from another network, another jurisdiction or another day. What can be said with confidence is narrower and still useful: on the date of record the labels were visible and the values were not.

The consequence is a discipline, not a conclusion. Do not attribute a throughput figure to this exchange that was not read from a dated source. Do not convert a member count, a connected-network count or a port count into another. And treat any single number quoted for Hong Kong interconnection — including one quoted approvingly — as a category claim until the category is named.