FLAP without the P

At the start of the summer, Gaël Hernández published a study that is well worth a careful read by anyone working on internet peering in France. By aggregating, using data from PeeringDB, the port capacity of ten of the most prominent CDN, cloud and content networks at IXPs – Akamai, Meta, Amazon, Hurricane Electric, Cloudflare, Fastly, Microsoft, Google, Netflix and ByteDance, he has ranked global internet exchange points according to the capacity these networks have actually deployed there. The metric is simple, yet telling: it shows where the networks generating the bulk of consumer traffic have chosen to locate their public peering edge.

The figures are staggering. Akamai alone maintains 79 Tbit/s of public peering capacity spread across 248 internet exchange points. IX.br São Paulo tops the global rankings with 22.8 Tbit/s of aggregate CDN capacity, more than double that of the runner-up, DECIX Frankfurt. Five internet exchange points (IX.br São Paulo, DE-CIX Frankfurt, Equinix Singapore, BBIX Tokyo and IX.br Rio de Janeiro) host all ten of the networks analysed.

But for those of us working on interconnection in France, the most striking finding is not who features in the rankings, but who does not.

The missing P

For the past twenty years, the story of European interconnection has been defined by the acronym FLAP: Frankfurt, London, Amsterdam, Paris. According to Gaël’s data, Europe has five exchange points in the global top 30: Frankfurt, London, two in Amsterdam, and Milan. Paris is absent. The fourth position in Europe is not held by a Parisian exchange point but by NL-IX, and the fifth by MIX in Milan.

Let’s be cautious about this metric: it refers to the port capacity deployed by ten specific networks – not traffic, not the number of participants, and not economic relevance.

France-IX remains a thriving internet exchange point with a strong community.

But it is in terms of peering capacity that CDNs vote with their money, and the vote is unambiguous: the ten largest sources of content on the internet have collectively concluded that Paris does not require the same level of public peering as Frankfurt,

Amsterdam or London, nor São Paulo, Johannesburg or Mumbai.

It’s no accident. It’s a design.

ARCEP’s Interconnection Barometer has been documenting this French anomaly for years. Incoming traffic on the networks of the main French ISPs reached 50.8 Tbit/s at the end of 2024, up 9.2 % year-on-year. But let’s look at how this traffic arrives: in the 2024 edition of the barometer (H2 2023 data), around 54 % came via transit, around 45 % via private peering, and barely 1.2 % via public peering at internet exchange points. The volume of public peering had even fallen by a third, returning to the level it had been at five years earlier.

The structural explanation is well known. A significant proportion of the «transit» traffic is internal to the Orange Group: traffic flowing from Open Transit International, Orange’s Tier 1 provider, to RBCI, its domestic backhaul network. Add to this an ISP market centred on four national players, a strong culture of private interconnection and paid peering, and the widespread deployment of on-net CDN caches within the ISPs’ networks: the result is a country where content reaches users very efficiently, but almost entirely outside the public, neutral and multilateral framework provided by IXPs.

Interestingly, Gaël’s study identifies the same pattern elsewhere: the absence of Australian exchange points from the top 30 is attributed to a concentrated ISP market, served by deep caches. France is not alone; it is simply the largest European economy to have organised its interconnection in this way.

Why the rest of the data should encourage regional IXPs

If the study merely confirmed Paris’s unique position, it would remain little more than a curiosity. What makes it strategically interesting for the FRNIX community are three lessons drawn from other regions.

The distributed model works. NL-IX, ranked fourth in Europe in terms of deployed CDN capacity, is not a single-metropolis giant. It is a distributed network spanning more than 83 sites across a dozen cities and more than eight countries. Gaël’s conclusion is worth noting: distributed exchange architectures could increasingly complement, or even partially replace, the traditional single-metropolis exchange model. This is precisely the premise underpinning a federation of regional exchanges. Interconnection value is not meant to be concentrated in a single building in a single capital city.

Community exchanges can reach the highest level. Three of the leading North American platforms in the ranking – SIX Seattle, CIX Atlanta and FL-IX Miami – are non-profit, member-run internet exchange points operating in the world’s most commercialised internet market. Scale and community governance are not mutually exclusive.

A coordinated national strategy pays off spectacularly. Brazil has three peering points ranked amongst the world’s top six. The IX.br model – a coordinated national network with a strong local presence, stretching from São Paulo to Fortaleza – has made public peering the country’s primary content delivery mechanism: Meta alone deploys 6.4 Tbit/s in São Paulo. A country decided that traffic should be exchanged locally, built the infrastructure to support this, and CDNs followed suit.

What this means for us

France has an excellent metropolitan fibre network, a dense ecosystem of data centres, submarine cables landing in Marseille, and one of the highest IPv6 adoption rates in the world. What it lacks is not infrastructure: it is a culture of interconnection that prioritises the public and neutral exchange of traffic as close as possible to users, in Toulouse, Lyon, Marseille, Bordeaux, Lille, Nantes and elsewhere.

The data show that there is a genuine opportunity here. CDNs are demonstrably investing in public peering where local ecosystems warrant it – in Brazil, Johannesburg, Mumbai and Seattle. The reason they have not done so in France is that demand there has centred on transit and private interconnection, not because France is unattractive.

As regards regional exchange points, the roadmap that emerges from the data is clear: to bring together rather than fragment, so that a distributed French network can offer a coherent value proposition; to embed each exchange within its local community of operators, hosting providers, universities and public networks, because, as Gaël himself wrote a few years ago, there is no internet exchange without a community; and measure and then publish the results, as visibility in PeeringDB and public statistics is what enables an exchange to appear on the map that CDN capacity planning teams actually consult.

The fact that Paris is not in the world’s top 30 is not a judgement on France. It is a snapshot of the choices made over the past twenty years, and an invitation to make new ones for the next twenty years.

Sources: Gaël Hernández, «Peering Capacity at Public Internet Exchanges: What the Data Reveals», RIPE Labs, 4 June 2026 (also featured on the APNIC Blog and ISOC Pulse); ARCEP, Barometer of Data Interconnection in France, 2024 and 2026 editions.