Peering means exchanging traffic directly between two networks instead of paying a carrier to carry it. But a peering is an agreement, not a product bought at a counter: it is requested, negotiated, and sometimes refused. Here is how it works, with the real cases and the reasons behind a refusal.
Peering exchanges traffic directly between two networks instead of routing it through a paid carrier. The goal is threefold: cut the transit bill, lower latency by shortening the path, and gain control and resilience.
Transit means buying access to the whole internet from a carrier. Settlement-free peering exchanges traffic between two networks with no invoice. Paid peering has one of the two networks pay the other for the interconnection. Which mode applies depends on volume, traffic balance and bargaining power.
There are two routes. An internet exchange (IXP) is shared infrastructure where dozens of networks are present: you connect once, then open a BGP session with each network that agrees. A direct link (PNI, private network interconnect) is a dedicated cable between two networks, reserved for large volumes. You often start on an exchange, then move to a PNI as traffic grows.
Every network publishes a peering policy that sets its conditions. The usual criteria: a minimum volume of traffic exchanged, presence at several common interconnection points (including home markets), consistent route announcements, dual-stack IPv4 and IPv6, an up-to-date entry in the network directory, and a responsive technical contact. Meeting these does not guarantee an agreement, but failing them is enough to be refused.
The most sensitive point is traffic balance. A content network (streaming, CDN) sends far more than it receives; an access network (ISP) receives far more than it sends. That imbalance shifts the cost of carriage onto the access network, which may then refuse free peering and ask for paid peering. The ratio, often set around two to one, is the criterion that tips a negotiation.
A refusal rarely comes down to a single reason. Traffic imbalance is the first. Then comes commercial interest: keeping a competitor as a paying transit customer rather than peering with it for free. Add the cost of ports and links, and sometimes deliberately high conditions used as a polite refusal. Policies range from open, peering with almost anyone, to selective, to restrictive, peering only with networks of comparable size.
When a content network wants to reach an access network's subscribers and the latter demands payment, two outcomes appear. Pay, often as paid peering or cache servers installed inside the access network. Or fall back on transit, at the risk of congestion at peak hours. Well-known disputes have ended in severed interconnections, degrading access to a service for the duration of the standoff. It is an economic balance of power as much as a technical one.
Cutting an existing peering is a pressure lever. The network that cuts forces the other to reroute its traffic through pricier or slower transit until a deal is reached. For the end user, this shows up as sudden slowness toward certain services with no visible outage: traffic still flows, but over a degraded path.
In a data center, the richness of the interconnections present, exchanges and directly reachable networks, determines the latency and resilience of your services. Choosing a well-peered site, and knowing exactly who is present there, matters as much as the price of the rack. That is precisely what an interconnection comparator lets you check before committing.
The traffic exchange can be, which is called settlement-free peering. But costs remain: the port on the exchange, the cross-connect, membership. And if traffic is imbalanced, the other network may ask for paid peering.
Usually because traffic would be too imbalanced, because it prefers to keep you as a transit customer, or because you do not meet its published conditions (volume, multi-site presence, up-to-date directory). Some conditions also serve as a polite refusal.
On an internet exchange (IXP) you reach many networks through shared infrastructure. A PNI is a dedicated direct link between two networks, chosen when volume justifies a cable of their own.
No. Peering only gives access to the networks you establish it with. Transit is still needed to reach the rest of the internet. You combine the two.
Written on 1 September 2026.
Estimate my cost → Compare data centers