Blog Post

Why do U.S. ISPs want to charge for peering? Peering makes the internet cheaper. Here’s how

As more broadband networks connect directly to each other via peering agreements, the amount of money paid for internet transit could fall, according to a report from TeleGeography. That’s good for users of the internet, but will cut into transit revenue at companies that range from Level 3 and Cogent to ISPs. Of course, TeleGeography isn’t sure if this will take place. From its report:

As Internet service providers worldwide have gradually migrated from purchasing transit to establishing mostly free peering arrangements, the share of global Internet traffic connected via transit agreements declined from 47 percent in 2010 to 41 percent in 2014. As long as this relative decline of transit continues, TeleGeography forecasts that IP transit-related revenues will fall from $4.6 billion in 2013 to $4.1 billion in 2020. If the ratios of traffic routed via transit and peering were to stabilize at current levels, IP transit revenues would increase to $5.5 billion by 2020.

By cutting out the internet’s middlemen, peering agreements lower the cost of bandwidth and the cost of IP services. The TeleGeography report lays out the case for peering gaining ground over transit, and shows how it expects peering to grow. This is good for the internet as a whole, but this anticipated shift to peering over paying for transit has led to some behavior shifts that are causing harms for consumers. It also means trouble for existing transit providers, especially those without other lines of business.

Peering, the practice of two networks exchanging traffic directly either for free or for money, has been going on for decades. Historically, networks of a certain size would sign peering agreements because it would save them money. Instead of building out an internet pipeline to every endpoint, two networks meet in the middle and exchange traffic. In 2012, an OECD report found that peering has helped lower the overall cost of providing bandwidth and that most peering agreements are unpaid.


Transit, on the other hand, is when a company like Google or Disney pays a middle mile provider such as Level 3, Tata or Cogent to bring its traffic to the ISP. ISPs like AT&T, Verizon and Comcast also offer transit services. But as peering (most of it unpaid) rises, the demand for transit goes down. In many cases networks only peer with networks that meet some sort of minimum traffic level or at internet exchange points. So the growth in networks (driven by overall demand for broadband) and the rise of internet exchange points around the globe is causing peering to become more popular by making it possible for more networks to peer.


This is good for most participants on the internet but it’s obviously not good for transit providers. Many of them have seen this coming, which means they have branched out into other businesses such as providing CDN services, data center colocation — or in the case of U.S. ISPs doubling down on charging for peering. That’s not the primary reason that Netflix, for example, is having so much drama with ISPs over peering, but it doesn’t help either.

8 Responses to “Why do U.S. ISPs want to charge for peering? Peering makes the internet cheaper. Here’s how”

  1. Mr. Cleofás Leandro

    Remote peering has become quite relevant over the last decade (or so says my data) and remote peering is frequently provided by transit providers, hence more peering does not necessarily imply a reduction of transit providers benefits.

  2. “or in the case of US ISPs, doubling down on charging for peering”
    [citation needed]
    — i.e. what are you using as the basis for your assertion that paid peering is somehow new?
    At the risk of deflating your narrative that this is all about the big transit ISPs being desperate and greedy, some background:

    Transit vs Peering is a matter of routing policy and the resulting reachability it gives a network.
    Transit service means a customer pays an ISP to deliver their bits to the right destination, whether it’s local to the ISP’s network (on-net) or not, meaning that the customer uses the ISP network to reach the ISP’s other customers, and uses the ISP’s interconnection to other ISPs (their peering) to reach destinations that aren’t direct customers of their ISP (off-net). The converse is true for inbound traffic destined for the customer’s network.
    If your network doesn’t need transit service, because it is well-connected to other networks (via IXPs and other peering/transit deals), you can use non-transit service, which means that for the most part, you only send bits to your non-transit ISP connections when they are destined for that network’s directly-connected customers, and conversely, they only send traffic to you that originated somewhere on their network or their customers’ networks, meaning that you won’t receive traffic from that ISP’s other non-transit peers. From a policy perspective, peering is a variant of non-transit service, in that it typically also includes some policy tweaks so that any local (on-net) route will be preferred in order to keep the traffic on-net as much as possible, and peer routes will be the last resort.

    Independent of the routing arrangement (transit vs non-transit), there’s the question of what the fee structure looks like for non-transit connections. A network that wants non-transit connections to an ISP can try to qualify for settlement-free peering, which typically comes with those minimum requirements around the amount and nature of the data exchange, locations, etc. with a focus on ensuring that it’s an equitable exchange between both parties since they are essentially agreeing to share the cost of interconnect. If a network doesn’t meet those peering qualifications, but still wants direct connectivity to a given ISP’s customer base (instead of paying another ISP transit to reach it), they pay for non-transit service. Non-transit has also come to be known as paid peering, probably because it implies a relationship that is more like a peer than a customer. I’m not going to defend large ISPs’ peering requirements, because while it does make sense that an ISP is going to try to sell connectivity instead of giving it away wherever they can, they probably need to evolve to reflect the way that interconnection is changing. Either way, paid peering is certainly not a new concept.

  3. my2cents

    Cheaper? Its been getting cheaper every year to move data – but do we see reductions in cost to the consumer? Nope. Nevermind the unscrupulous practice by ISPs allowing the ports to saturate at their own customers expense in order to force providers to start paying the ISP directly to simply allow proper routing (which had been working fine until the greedy exec found how to double dip). If they really wanted to reduce costs, they could adopt P2P – but instead they have been discouraging that. Perhaps I’m reading a paid advertisement?

    • Alex: this is incorrect.

      Firstly, when you say “pay someone” – ultimately those “someones” will need to peer with each other in order to deliver your data to each others customer bases – so at some level there has to be peering.

      Secondly, common sense dictates that it is cheaper to move data a small distance than it would be a long distance. Since your transit provider is delivering traffic local, nationally and globally logically their costs must be a mixture of these and hence their price to you based on some sort of average. Providing you have enough traffic and can reach a nearby peering hub there must be a point at which your cost to peer is lower than transit. At present that level is in the gigabit-plus range (altho it varies between regions).

      Companies like ours (IX Reach) specialise in providing tailored peering access – that is we’ll provide the transport to the peering hubs, and the more traffic you have and bigger budget you hence have the further afield you can reach at less than transit rates. For the largest customers the numbers we see now are that we are providing peering for about 1/3 the price of even the cheapest transit providers.

      As the cost of long haul capacity falls, transit prices fall but so do peering costs. However if you don’t have much traffic then there is always a point at which other costs (colocation, cross connects, router ports) become more expensive – the trick is to wait until you have enough traffic that those are not the major costs.

      – Steve

      • Steve,

        What would be the current blended cost of getting 500Gbit/sec to any destination without having using “peering”?

        Because today it can be done for about $1.02 on a 1 year term including equipment via transit.