1,323 registered internet exchanges and 5,861 colocation facilities, read from the registry operators keep about themselves. Counting buildings is the wrong measure: the United States holds 23.48% of the world’s facilities and gets 9.0 networks per building, against the Netherlands’ 21.5. And 162 of the exchanges have no networks on them, or exactly one.
The largest internet exchange in the world is IX.br (PTT.br) São Paulo, with 1,860 networks on it — 1.83 times DE-CIX Frankfurt’s 1,018 and more than double AMS-IX’s 854. 3 of the ten largest are Indonesian.
The number that gets quoted is how many data centres a country has. The number that decides whether traffic stays local is how many networks are inside them. Across 5,861 facilities there are 61,713 network presences, and they are not distributed the way the buildings are.
1,376 facilities, 23.48% of the world’s, holding 20.04% of its network presences.
2.39 times the American figure. A Dutch facility is a different kind of place, not a bigger one.
203 facilities holding 4,095 presences — a seventh of America’s buildings and a third of its networks.
An internet exchange is a place networks agree to meet. Building one is a project with a ribbon; filling it is not. Of the 1,323 registered here, 93 have no networks on them and 69 have exactly one.
7.03% of every registered exchange. Not a low number of members — none.
Together with the empty ones that is 162 exchanges, 12.24% of the list. One network peering with itself is not an exchange.
Countries whose every registered exchange is empty or has a single member. A further 14 have facilities and no exchange at all.
The mean is 42.9. When a mean is three and a half times its median, the average is describing a handful of places and nothing else on the list.
43.5% of all exchanges. Small enough that what peering saves is unlikely to pay for the switch.
Share of every registered network presence held by the smaller half of all exchanges, between them.
Share held by twenty exchanges. The ten largest hold 15.04%.
The standard account of internet geography names four European cities. The registry does not agree: the largest exchange in the world is in Brazil, and 3 of the ten largest are in Indonesia.
Networks present, in São Paulo/SP. That is 1.83 times DE-CIX Frankfurt.
The largest in Europe, against AMS-IX’s 854 and LINX LON1’s 837.
Jakarta appears three times in the ten largest exchanges on earth, which is not where the usual story puts it.
An archipelago of a hundred and twenty million people, and the country this site is mostly about. It has 15 registered exchanges and 423 network presences across them.
Holding 423 network presences between them — fewer than a quarter of what sits on IX.br São Paulo alone.
Networks on the biggest one. The world’s largest has 1,860.
Across 24 facilities — higher than the United States’ 9.0, which is a reminder that this ratio measures concentration rather than capacity.
The claim that an exchange has no networks on it is the strongest one on this page, so it is the one with a second source behind it.
PeeringDB records the number of networks an exchange operator says are present, and separately imports a count from the exchange's own published member list. 395 exchanges publish one. Where both exist the median disagreement is 3 network, so the two track each other closely enough to be used together.
4 exchanges are recorded as having no networks by their operator while publishing a member list that shows members. Those are counted as registry gaps and excluded from the empty figure, which is why the number here is 93 rather than 97.
PeeringDB is maintained by network operators about themselves. A country with no rows may have no exchange, or may have nobody filling in the form, and nothing in this data separates those two. Every count here is of what is registered.
Four limits, each of which would change a conclusion above if ignored.
A count of networks at an exchange says nothing about how much passes over it. Two networks moving a terabit are one row; two hundred moving nothing are two hundred. No traffic figure is claimed anywhere on this page because the free sources do not carry one consistently.
It is a ratio of two registered counts, and a high value means the networks are gathered in few buildings rather than that the buildings are good. The Philippines scores 15.8 and the United States 9.0, which is a fact about how many facilities each has rather than about either country's capacity.
The submarine map contributes 728 cables and 1,925 landing points as a scale marker. Its landing-point records carry no country code and the routes it draws are schematic rather than surveyed, so no per-country split and no distance is computed from them.
PeeringDB records current state, not history, so nothing here shows change over time. An exchange that emptied last year and one that opened last week look identical.
Every figure on this page traces to one of these, through a CSV in
data/global-interconnect/. Each is checked against its source query on every
build.
Every registered internet exchange point and colocation facility worldwide, with city, country, continent, and the number of networks present at each — CC BY 4.0, no key. A registry operators maintain about themselves, so absence is not proof of absence: a country with no rows may have no exchange, or may simply have nobody filling in the form. Rate-limited for anonymous clients, so every response is cached on disk.
A second count of the networks at each exchange, imported from the exchange's own published IX-F member list rather than typed in by its operator — Carried on the same records as ixf_net_count. Used as the cross-check: an exchange the operator records as empty whose own member list shows members is a registry gap, not an empty exchange, and is excluded from the empty tally.
Submarine cables and their landing points, as GeoJSON — Publicly served JSON, no key. Used for landing-point counts only; the cable routes on that map are schematic rather than surveyed positions and no distance is computed from them here.
Population (SP.POP.TOTL), income group and region for each country — No key. Used to describe countries, never to compute an interconnection figure.