Summary

  • DE-CIX prints its group network figure three ways — "4600+" on the homepage, "more than 4,300" on the annual-report page, "more than 4,600" in the 14 October 2026 release — and not one of the twelve documents read defines "connected network" or states whether a network present at several locations counts once or once per location.
  • The company's own connected-networks lookup excludes eleven named places and the exchanges in India; no page read prints a currency port price, only pricing "according to location and bandwidth"; and the group peak, unlike the count, is printed as one measurement taken at 21:20 Central European Summer Time.

A number with three printings

Start where most readers start. The DE-CIX homepage carries a stat block reading "4600+ connected networks globally in 100+ countries", beside "60+ locations across Europe, Africa, the Americas and Asia" and "30+ years of expertise in interconnecting networks". The numeral wears its own plus sign. Nothing in that block says when it was true, what period it covers, or what one unit of the thing counted is.

The annual-report page prints a different figure in a different style: "connected networks rising to more than 4,300 – an increase of 25 percent year-on-year", followed by "Total connected customer capacity reached 220 terabits, representing growth of 40 percent". Further down the same page, the prior year appears as "over 4,000 connected networks, a 10% increase year-on-year" with 170 Tbit/s.

The release of 14 October 2026 — the only one of the three printings that carries a publication date — puts the group at "more than 4,600 networks from over 100 countries", and adds that in Frankfurt "almost 1,100 networks are connected" where "nearly 48 exabytes of data were exchanged" in 2025.

Three printings, one direction of travel, and not one sentence anywhere that says what is being added up. The figure is a floor at every appearance, which is the honest way to print a number you are not sure of, and the plus sign travels with it.

One page, two floors, one percentage that does not close

The annual-report page does something the other two do not: it puts a rate of growth next to two counts. "More than 4,300" this year, "over 4,000" last year, and "an increase of 25 percent year-on-year" in between.

Here is my arithmetic, and it is mine, not DE-CIX's. Divide 4,300 by 4,000 and you get 1.075 — roughly 7.5 per cent. A floor divided by a floor cannot establish a growth rate at all, and the page never says which pair of numbers its 25 per cent was computed from. Maybe the underlying exact counts close the gap; maybe the perimeter changed between the two years; maybe the denominator is a different quantity wearing the same noun. The page does not tell us, so this article prints both numbers and adopts neither, and it does not call either one an error.

What it can say is narrower and more useful: a reader who wants to check the most quoted growth claim in the document has two floors and a percentage that she cannot reconcile with them, and no third number to break the tie.

The noun is not defined in any document read

If the count is the headline, the counted object should be the first thing the company explains. It is nowhere.

DE-CIX maintains a page whose entire purpose is to tell you which networks connect at its exchanges. It opens: "Thousands of networks connect at DE-CIX's Internet Exchanges either to peer with all connected networks, to connect directly with selected networks, to access cloud service providers, or to use other interconnection services." And later: "You can also check out each exchange's list of connected networks directly under each location."

That is the closest thing to a definition on the site, and it is not a definition. It does not say what qualifies a network as connected — one port, one session, one contract, one physical presence, a signed order still in provisioning. It does not say whether a network that peers at Frankfurt, Munich and New York is one connected network or three. The ambiguity is not trivial bookkeeping: it is the difference between 4,600 customers and 4,600 customer-location pairs, and only one of those two readings supports the word "globally" beside "100+ countries".

I read twelve DE-CIX-controlled documents for this piece. Not one of them states the rule.

The lookup excludes part of the footprint

The page that declines to define the noun does disclose something else, in its own words, and the sentence deserves to be quoted in full: "Please note that the search does not cover the DE-CIX as a service partner exchanges in Athens, Dubai, Ruhr region, and Lagos or the partner exchanges in Berlin, Bucharest, Prague, Moscow, Sofia, St. Petersburg, Warsaw and the exchanges in India."

Count them: Athens, Dubai, the Ruhr region, Lagos, Berlin, Bucharest, Prague, Moscow, Sofia, St. Petersburg, Warsaw. Eleven named places, plus a whole national set in India. The exclusion applies to the network lookup, not to the group figure — the page offers no count for any of the excluded exchanges, and nothing here says those networks are missing from the 4,600.

But it does mean the company's own verification instrument stops at the edge of its own footprint. A reader who wants to know whether a particular network reaches DE-CIX in Lagos cannot use the directory DE-CIX offers beside the group figure to find out. Eleven cities and a country's worth of exchanges sit outside the lookup, with no per-location count printed anywhere in the documents read to substitute for it.

What the locations print, and what can be added up

If the group number cannot be decomposed by definition, perhaps it can be decomposed by arithmetic. The per-location pages are the place to try.

The Frankfurt page is the flagship, and it prints: "with more than 18 Terabit per second peak traffic, DE-CIX Frankfurt is the world's leading Internet Exchange. Right in the heart of Europe, you can reach more than 1,000 local, regional, and global networks." It also prints "About 80 percent of the connected networks are available at the DE-CIX route servers." Neither figure carries a date.

The 14 October 2026 release gives the same exchange as "almost 1,100 networks" and 21.47 Tbit/s. Two printings by one publisher, one undated, one dated; a reader who wants the newer number has to leave the location page to find it. That is not a correction and this article does not call it one — an undated page cannot be out of date in any checkable sense. It is simply two printings, and they do not agree.

Elsewhere the pattern is thinner still. Munich prints no Munich count at all; it offers "more than 1000 networks in Frankfurt or more than 200 networks in New York, just via a VLAN (GlobePEER Remote service)". New York prints its rank — "the 5th largest Internet Exchange in the US" — and no count of its own in the text retained. Singapore prints "140+ Networks live and in provisioning accessible through the DE-CIX ecosystem", which is three qualifiers stacked on one numeral: a floor, a status that includes provisioning, and a reach that is vaguely wider than Singapore. And the locations index lists exchange after exchange in prose where the only quantified entry in the retained text is Frankfurt, repeating its own two undated numbers.

So try to add it up. Two locations print a count of their own. One is a floor on an undated page; one is a floor with a provisioning qualifier and an undefined scope. The rest print nothing, or print somebody else's number. Nowhere in the documents read does any page state whether a Frankfurt network also appears in the group figure, or whether the Frankfurt and New York numbers overlap. The sum cannot be attempted, and that is a property of the printings rather than a suspicion about the count.

The peak is the honest instrument

Now the contrast that makes the count's shape visible.

The 14 October 2026 release reports two records in one paragraph: "On Tuesday evening, 21.47 terabits per second (Tbit/s) were transmitted via DE-CIX Frankfurt for the first time. At the same time, traffic across DE-CIX's global Internet Exchanges reached a new peak of 32.10 Tbit/s. Both records were measured at 21:20 Central European Summer Time, just 20 minutes after kick-off in numerous UEFA Champions League matches."

Read that last sentence again. Both records, one timestamp. The global peak is therefore one measured quantity at one instant — not an addition of per-location maxima recorded at different times, which would be a meaningless number dressed as a total. Maxima reached at different moments cannot be summed into a peak of anything. DE-CIX does not print that instrument, and this article has no complaint about this paragraph: the peak is the best-built number in the entire set, because the company told the reader exactly how it was taken.

Which is what makes the count's looseness so conspicuous. One quantity in these documents carries its measurement conditions. The other carries a plus sign.

The port is priced by form

The most surprising absence is not about counting at all.

Exchanges normally publish a port ladder: 10G at this, 100G at that, per month, plus a setup fee if the splice cabinet is involved. DE-CIX does not. Its services page says: "Our services are priced according to location and bandwidth – for information on pricing, please fill in our form and we'll get back to you with the details." That is the entire public pricing statement. No currency amount, no tariff table, no order form, no rate card in any of the twelve documents.

What a reader can learn is the shape of the object being sold. The access documentation defines it precisely: "An access is the physical port provided by DE-CIX, from which you enter our network." Capacities run 1GE, 10GE, 100GE, 400GE. The same article distinguishes "premium enabled sites where DE-CIX operates its proprietary hardware" from "partner sites where we collaborate with partners to establish your connection", and adds: "Upon establishing a connection, you can then order a multitude of services as needed."

So a buyer knows exactly which physical object she is buying, in which four speeds, at two classes of site — and cannot learn its price from anything public. My own reading, and it is analysis rather than a finding: a quote-only price on a standardised, internationally repeated product is a deliberate position, not an oversight. It costs nothing to publish a number, and publishing one hands every competitor and every large buyer a benchmark. Withholding it keeps that benchmark inside the sales conversation.

Companies are entitled to price that way; the consequence for the reader is simpler than any motive — you can price the port only by asking, and you can never check the answer against a printed list.

Remote peering is the same port, further away

One more layer sits under the count, and the documents do print it.

The GlobePEER Remote page describes remote peering as: "Peer remotely at additional DE-CIX locations. Use your existing access in one of our Internet Exchanges to peer remotely at other interconnected locations." The mechanics follow: "You peer over a VLAN on your existing access, and we will take care of all interconnects needed and the access at the additional location. The service can be delivered within one day, and there are no additional cross connect costs."

That last clause matters for the counting question, and not in an accusatory way. DE-CIX supplies "the access at the additional location". A network can therefore be present at an exchange — reachable by the other networks there, listed among those you can peer with — while owning no equipment, leasing no cross-connect and holding no port of its own in that city. It holds a VLAN on a port somewhere else. Whether such a network is one connected network, or a connected network at two locations, is precisely the question the connected-networks page never answers.

That is not a loophole and nobody is sneaking anything. It is the product working as designed, sold on a page that says so. But it does mean the phrase "connected networks globally in 100+ countries" describes reach that includes networks with no local presence at all, and the reader is entitled to know which reading the figure uses.

What a reader can actually do

Put the pieces side by side and the asymmetry is the story.

A buyer can determine the physical object being sold, its four capacities, and the two classes of site it can sit in. She can order remote peering over that access and be told honestly that there are no additional cross-connect costs. She can read a traffic record with its measurement time attached.

She cannot learn what the port costs without a conversation, cannot learn from any published page how many distinct networks sit behind a location's figure, cannot tell whether a network at three locations appears once or three times in the group total, and cannot use the company's own directory to check eleven named places and the exchanges in India.

None of this makes the numbers wrong. A floor is a truthful way to print a number you know is at least that large, and a quote-only price is a normal commercial choice. The finding is narrower and more durable than an accusation: DE-CIX publishes a count without a counted object, a growth rate that does not reconcile with the two floors printed beside it, and a price without a number — while publishing, in the same estate, a traffic peak whose measurement conditions are stated exactly.

The instrument that is built carefully tells you how it was built. The one that is not, doesn't.

Boundaries

  1. The reading is bounded to twelve DE-CIX-controlled pages retrieved on 20 November 2026 between 20:42Z and 20:54Z, each retained as a source snapshot.
  2. Four of the twelve were read as page-text excerpts returned by retrieval rather than through a browser, so a dynamically loaded element, live counter or footnote may exist outside the retained text; where that bears on a claim, the claim is narrowed to what the excerpt shows.
  3. The connected-networks search widget was not operated. Nothing here describes what the lookup returns; it describes what the page says about the lookup.
  4. No PDF annual report was read page by page; group figures past the quoted pages may exist elsewhere and are neither asserted nor denied.
  5. No page read defines "connected network", "network connection" or "location", and no page read states whether a network present at several locations is counted once or once per location.
  6. The 25 per cent figure is reported as printed and never reproduced; the roughly 7.5 per cent implied by the page's own two floors is my arithmetic, labelled as mine, and neither number is called an error.
  7. The Frankfurt pair — 1,000+ networks and 18 Tbit/s on one undated page, almost 1,100 networks and 21.47 Tbit/s in the 14 October 2026 release — is reported as two printings of one exchange. The undated page is not called stale and the release is not called its correction.
  8. The group peak is reported as a single measured quantity because the release states both records were measured at 21:20 CEST. That is a statement about the instrument, not about the count.
  9. No currency amount, tariff, price list or order form is quoted, because none appears in the twelve documents; the finding is that pricing is described only as being by location and bandwidth, through a form.
  10. The eleven named places and the exchanges in India are reported as a set excluded from the network lookup, in the page's own words, with no count and no motive attached.
  11. The route-server share is reported as printed; this article does not multiply it by any floor to produce a population.
  12. Nothing here concerns the conduct of any named party, the correctness of any count, the standing of any participant or the soundness of any price. The subject is which documents print which numbers, and what a reader can recompute from them.
  13. No BGP session, looking glass, routing table, PeeringDB record or traffic measurement was queried. Every number in the article is a number some DE-CIX page or release prints.