Summary

  • DE-CIX's own documents print the connected-network figure at four scales — "more than 2,000" in the FAQ, "more than 3,000" on the About page, "more than 4,300" for 2025 in the annual report and "4600+" on the homepage — and the page dedicated to that count states no rule for what is counted.
  • The peak-traffic record carries a clock time but no sampling interval anywhere in the eleven documents read, and the 220-terabit capacity total is never defined as a sum of contracted port speeds; by my own division, the November global peak equals 15.0 percent of that total.

A rate with a clock and no window

At 20:20 Central European Time on Wednesday 18 November 2026, data throughput across DE-CIX's Internet Exchanges worldwide reached "a new record-breaking high of 33.03 terabits per second" (DE-CIX, 19 November 2026), and DE-CIX Frankfurt carried 22.15 Tbit/s at the same instant. The release also names 22.30 Tbit/s at Frankfurt on 10 November, so the record announced on the 19th was not the highest value that same document prints for the month. Five weeks earlier the operator reported 21.47 Tbit/s at Frankfurt and 32.10 Tbit/s globally, "measured at 21:20 Central European Summer Time, just 20 minutes after kick-off in numerous UEFA Champions League matches" (DE-CIX, 14 October 2026). DE-CIX Management GmbH, the Frankfurt operating company whose pages carry all of this, is the identity behind the figures in our directory entry.

Two releases, two clock times, one unit. What neither release prints is the interval over which the maximum was taken — whether 33.03 Tbit/s is an instantaneous sample, a one-minute value, or a five-minute average. Move from releases to reports and the window changes shape rather than gaining precision. The 2025 annual report labels each location's maximum "Peak traffic (Dec 2025)" and puts Frankfurt at 18.7 Tbit/s (Global Annual Report 2025). The 2024 report says "Global peak traffic from peering finished the year just shy of 25 terabits per second", alongside 68 exabytes "shared across the interconnected platforms throughout 2024" (Global Annual Report 2024). And the corporate About page still prints "more than 18 Terabits per second peak traffic" for Frankfurt, with no date attached (About DE-CIX).

Eleven documents were read for this piece: ten published by DE-CIX and one industry report. In none of them does a sentence fix the sampling interval, averaging period or percentile behind the public peak record. The only granularity statement in that corpus is about a customer dashboard, not the headline: "This dashboard shows the average bits/s of a 5-minute bucket of data of the chosen service" (DE-CIX documentation). Euro-IX explains why the omission matters without filling it: "Not all IXPs measure their peak traffic using the same periodic average or use the same method. Most IXPs choose to take samples every five minutes" (Euro-IX IXP Report 2021). That is an industry statement. It is not this operator's method, and no page read here claims it is.

The count that has no rule

The network total travels further than the rate, and it changes size on the way. The homepage stat block reads "4600+ connected networks globally in 100+ countries", with no as-of date (de-cix.net). The annual-report landing page prints "connected networks rising to more than 4,300 – an increase of 25 percent year-on-year", and restates 2024 as "over 4,000 connected networks" (Annual Report landing page). The 2025 report renders the same growth as "Network connections across DE-CIX's platforms grew by 25%, reaching more than 4,300 globally", using "network connections" and "connected networks" for one figure. The October 2026 release says "more than 4,600 networks from over 100 countries". The About page says "more than 3,000 networks worldwide trust our experience and service quality". The FAQ says "Although more than 2,000 networks peer at DE-CIX's exchanges, not all ASNs are available" (DE-CIX FAQs). Four scales: 2,000, 3,000, 4,300, 4,600.

Dividing the largest by the smallest gives 2.3. That is my arithmetic, printed on no page, and it measures the spread between documents rather than any single change in the network.

Some of the spread is age. An undated About page can lag, and the 18 Tbit/s peak on it lags the 2025 report by a year. But the page that exists to explain the count does not resolve it. DE-CIX's connected-networks page describes what networks do at its exchanges and points readers to per-location lists; it contains no counting rule. Nothing on it says whether a connected network is a distinct autonomous system, a customer, a contract or a port, and nothing says whether a network present in Frankfurt and Singapore is counted once or twice (Connected networks).

The nearest thing to a definition in the whole corpus is a peering requirement rather than a counting rule: "Any network that wants to peer at a public Internet Exchange must have a public ASN." That tells a participant what it needs. It does not tell a reader what the total contains. The 2024 report shows that scope qualifiers exist and get printed when someone chooses to print them: its figure carries the tag "connected networks (including India)".

The same connected-networks page also carries a scope statement the headline figures never repeat. Its search, the page says, does not cover the DE-CIX-as-a-Service partner exchanges in Athens, Dubai, the Ruhr region and Lagos, the partner exchanges in Berlin, Bucharest, Prague, Moscow, Sofia, St. Petersburg and Warsaw, or the exchanges in India. Whether the 4,600+ total includes those populations or excludes them the way the search does is not stated on any page read here.

The third number never sat on a meter

Capacity is a separate object again. "Total connected customer capacity reached 220 terabits, representing growth of 40 percent" for 2025, against 170 terabits a year earlier and growth of 20 percent in the 2024 report. The 2025 report prints Frankfurt at 105 Terabits under the label "Connected Customer Capacity". No sentence in the corpus says whether that total is a sum of contracted port speeds, installed switching capacity, or some other bookkeeping quantity.

The 2025 report discusses the capacity rise beside 10 GE, 100 GE and 400 GE port growth and "an 800 GE customer port", which is consistent with a sum of ordered port speeds. Consistent with is not the same as defined, and the difference is the whole of field four.

What the three numbers do when you divide them

My arithmetic, labelled mine, held apart from every printed figure.

Forty-eight exabytes across Frankfurt in 2025 is 384 exabits over 31,536,000 seconds: an annual mean of 12.18 Tbit/s. The December peak the same report prints for Frankfurt, 18.7 Tbit/s, is 1.54 times that mean. Seventy-nine exabytes globally is 20.04 Tbit/s averaged across 2025. And 68 exabytes in 2024 is 17.25 Tbit/s, against a year-end peak "just shy of 25" — a factor of 1.45. A peak and an integral behave differently by construction; what matters here is that both are quoted as evidence of the same growth while only one of them ever crossed a switch.

Capacity divides too. Two hundred and twenty terabits over 4,300 counted networks is 51.2 Gbit/s of contracted capacity per counted network. Frankfurt's 105 terabits over the 1,100 networks its November release claims is 95.5 Gbit/s. And the global peak of 33.03 Tbit/s equals 15.0 percent of the 220-terabit total printed for 2025, one year earlier. No reviewed page prints any of those four divisions.

What the documents cannot answer

Whether a rise in traffic reflects more connected networks, larger ports, or a different measurement window cannot be separated from the published surface, because no page prints two of those three terms for one period. The composition of the 220 terabits cannot be checked, so the ratio of sold capacity to observed traffic cannot be attributed to any stated rule. And whether the 4,600+ figures include the partner and India populations that the lookup excludes is unstated.

Each gap has a document that would close it: a statistics endpoint publishing the interval, a methodology note defining the count and the capacity total, and the lookup run without its exclusions.

The stop

Those three documents were cited, implied or required and not obtained in readable form, so fields two, three and four cannot be closed here and field five stands as my arithmetic only. Both annual reports were read as search-retrieved text, truncated in places, and a definition sentence could sit in a section that was not retrieved. That is a limit of this reading, recorded as one rather than smoothed into the finding.

Why the order matters

DE-CIX measures traffic. Its switches see the bits, and the clock times it prints are the mark of an operator that knows when its maximum happened. The trouble starts one noun to the right. A network count with no counting rule is not a measurement of a network; it is a count of relationships, kept current by whoever edits the page. A capacity total built from ordered ports is a statement about what customers bought, not about what crossed a switch. Three objects, four scales, one sentence of context.

The exchange publishes the numerator, and the denominator's definition is left to whoever quotes it — which, in this industry, is everybody.