Summary

  • The RIPEstat AS-overview record for AS24508, captured on 18 October 2026, returns the holder as "DE-CIX-AS-AP - DE-CIX Kolkata Internet Exchange", a block of 23552-24575 described as "Assigned by APNIC", and an announced flag of false.
  • The only quantitative material retained about the operator group is a May 2026 release reporting +25% network expansion, +40% connected customer capacity, an 800 GE port and a 120% increase in 400 GE ports, with no Kolkata breakdown and no traffic figure at any site.
  • Installed capacity is a stock; carried traffic is a flow. The public record for this site supplies the first kind of quantity and not the second, so a claim that the fabric carries material traffic is, at present, an assumption rather than a published measurement.

What the registry entry establishes, and what it does not

The record is short enough to quote in full. For AS24508 it returns a type of "as", a resource of 24508, a block spanning 23552-24575 whose description reads "Assigned by APNIC" and whose registry is named as the IANA 16-bit Autonomous System Numbers Registry, a holder of "DE-CIX-AS-AP - DE-CIX Kolkata Internet Exchange", and "announced": false, with a query window of the single day 18 October 2026 (RIPEstat AS-overview, AS24508).

That settles a narrow identity question. Something in the APNIC-administered range is registered under a name that contains the exchange's own name. The BTW directory record for this object has until now carried one supporting public reference and an explicit warning that no operator is confirmed (directory record). A registry holder name is not a corporate registration and not a licence, but it is a stronger identity signal than a marketing page, because the registry field is a structured field rather than prose.

It settles nothing about scale. There is no participant count in the record, no port count, no port-speed mix and no traffic figure. The announced flag deserves particular care, because it is the field most likely to be misread. It reports whether the autonomous system was observed announcing routes in the routing data queried on that date. It does not report whether participants exchange traffic across a switching fabric in Kolkata, and an exchange's own ASN being unannounced is compatible with a fully working fabric whose participants use their own ASNs.

Anyone who reads false as "the site is dark" has converted a routing observation into a traffic conclusion the document does not support.

The only numbers, and the scope they come with

The retained quantitative material is a group-level release, and it is worth reading for exactly what its scope excludes. As reported, the operator described a "strong start to 2026 following rapid network expansion (+25%), increased connected customer capacity (+40%), and deployment of the world's first 800 GE IX port with a 120% increase in 400 GE ports" (TelecomTV report, May 2026).

Four details matter more than the direction of the numbers. The scope is the group across Europe, the Middle East and India, not a site. There is no per-location breakdown that would let a reader allocate any part of a 25 or 40 per cent figure to Kolkata. There are no base counts, so even a reader willing to assume proportional growth cannot compute an absolute quantity.

And none of the four figures is a traffic measurement: network expansion, connected customer capacity and port counts are all descriptions of what has been provisioned or contracted, and the 800 GE and 400 GE references are port speeds, which are the size of a pipe rather than the volume moving through it.

This is the crux of the distinction. A stock is what has been installed, ordered or offered: ports, speeds, sites, connected networks. A flow is what actually moves, measured over a stated interval: a peak, an average, a 95th percentile. Stock growth is usually a precondition for flow growth, and it is reported far more readily, because it is a purchasing and engineering fact rather than a measurement. A group that adds 120 per cent more 400 GE ports has more capacity available somewhere; that sentence cannot tell you whether a particular city's fabric is busy, half-idle, or not carrying the traffic the port count implies.

An author's own arithmetic would be the obvious next step and is not available. Deriving a Kolkata share would require a Kolkata figure and a group total measured on the same date. Neither exists in the retained record. Dividing the percentages, or applying them to an assumed base, would manufacture a number and present an invention as an estimate, so this article asserts no such share.

The absence, stated where it stops

The operator's own locations page is the place a reader would expect to find per-site statistics, and for this site it did not supply them as read for this run. The provider reported that DE-CIX lists Kolkata among its locations and that site-specific telemetry — connected network counts, aggregate port capacity or current traffic volume — was not accessible through the standard public locations page (DE-CIX locations).

That claim must be labelled for what it is. No verbatim text was returned from that page, so the evidence is summary-strength: the reader should treat it as a description of what an inspection surfaced, not as a quotation, and not as a statement by the operator. It is scoped to the page as read on 18 October 2026. It does not establish that the operator publishes no site statistics anywhere, that no subscriber-only or member-facing dashboard exists, or that any other page, PDF or portal lacks such figures.

A companion query aimed at participant-submitted registry data returned zero candidate sources, which is a null search result and not evidence of a missing record; it is reported here only so that the reader can see the search was attempted.

Within those limits the finding is still useful, because it is the correct shape of finding for this kind of subject. Exchanges are named in announcements and described in the language of presence: sites, ports, gigabits offered. The evidence that a named exchange actually functions as a meeting point is a traffic measurement at that site, or a participant list that lets a reader count the networks that chose to connect. Where neither is public, the presence claim is doing work that measurement would otherwise do.

Why the distinction is not academic

The practical consequence runs through two decisions that are made from exactly this kind of public material.

For a network deciding whether to peer at a site, the useful question is whether the counterparties it wants are already there and whether the fabric has capacity headroom for a new port. A registry assignment answers neither. What it does answer is that an addressing identity exists, which is a necessary condition for an exchange to run route servers or a looking-glass service under its own number. That is a real operational fact and a weak commercial one.

For an enterprise or an institution placing latency-sensitive capacity in a city, the reasoning is blunter: a named exchange in a named place is routinely treated as evidence that traffic can be exchanged there, and that a path exists without backhauling to another metro. Where the public record contains only an assignment and group-level port language, that assumption is being supplied by the announcement rather than by the operator's statistics. The commitment that follows — cross-connects, a co-location contract, a backhaul route — is signed against the assumption.

None of this means the Kolkata fabric is empty. It means the public record does not say, and that the gap is being filled by inference from capacity language. The distinction between provisioned and carried is precisely what a buyer cannot see when only one of the two is published.

What would settle it

The falsification test is concrete and cheap, which is why the current gap is worth stating rather than deploring. Any one of the following would move the site from stock to flow: a dated operator figure for peak or average traffic at Kolkata with its measurement window; a Kolkata connected-network count; a published port-speed mix for the site; or a participant list detailed enough to count networks that have chosen to connect. Two of these would be better than one, and a series across reporting periods would be better still, because a single peak is a snapshot rather than a trend.

A negative result would also be informative. If a future publication shows a Kolkata port inventory that grows while no traffic series ever appears, or shows a participant count that rises while per-participant volumes fall, the reasonable reading is a site being provisioned ahead of demand — an option held rather than a fabric in use. That reading would require the traffic or participant series to exist, which is exactly what is missing today.

Evidence boundary

Everything stated above rests on three retained public documents, captured on 18 October 2026, and on one directory record. Every figure carries its own date: the registry fields belong to a query window of 18 October 2026, and the group growth figures belong to May 2026 as reported.

The boundary should be read as a property of the record, not as a claim about the world. No operator launch announcement, membership agreement, Indian licensing or regulatory text, landing-party document, route-server configuration or peering-session record was obtained as retained text in this run. Several retrieval attempts failed at the provider layer, and one returned an explicit statement that web access was unavailable; those are failed dependencies, not documents, and the absence of their content is not evidence of anything.

Where a future record supplies site figures, the thesis here is falsified, and it should be reported as falsified rather than defended.