The deal at the boundary
Every packet leaving your house eventually crosses from your ISP's network onto someone else's. That handoff happens under one of two arrangements, and the difference between them is measured in dollars that travel back through the supply chain to whatever you pay every month.
Peering is the simpler deal: two networks agree to exchange each other's traffic at no charge, on the understanding that the flows balance roughly. The arrangement is almost always informal — a handshake agreement called a settlement-free peering agreement — and it holds as long as neither party thinks the other is profiting disproportionately. Transit is what happens when the traffic doesn't balance, or when one network simply isn't large enough to negotiate an equal exchange. The smaller or less well-connected network pays the larger one for the right to reach destinations it couldn't otherwise reach on its own. The paying party is the transit customer; the receiving money is the transit provider.
| Peering | two networks exchange traffic without payment, on the assumption that flows roughly balance; formalized as a settlement-free peering agreement |
| Transit | the smaller or less-connected network pays the larger for access to destinations it cannot reach on its own |
| Tier 1 carrier | a backbone network large enough to peer globally without buying transit from anyone; AT&T, Lumen, Telia and NTT are examples |
| Exchange point | a physical colocation facility where networks connect to a common switching fabric to exchange routes |
| CDN placement | a content provider installs caching hardware inside an ISP's facility so traffic never crosses a transit link |
The practical consequence is a hierarchy. Tier 1 carriers — networks whose global footprints are large enough that they need not buy transit from anyone — sit at the top. AT&T's backbone, Lumen (formerly CenturyLink), Telia, NTT and a handful of others peer with each other across a web of settlement-free agreements that cover the whole internet without anyone cutting a check. Tier 2 carriers peer where they can and pay transit where they must. Regional ISPs, including most of the cable operators serving American suburbs, are Tier 2 at best. Smaller rural providers are often Tier 3, buying transit from a Tier 2 who bought some of theirs from a Tier 1.
Where the money flows
The cost of transit has dropped sharply over the decades — bandwidth that cost hundreds of dollars per megabit per second in the late 1990s now trades at fractions of a cent in dense markets — but it has not reached zero, and geography matters enormously. An ISP in Loudoun County, Virginia, sits twenty miles from Ashburn, one of the densest concentrations of internet exchange infrastructure on earth. Buying transit there, or connecting to one of the exchange fabrics at that campus, is cheaper per bit than almost anywhere in the country. An ISP serving rural Montana buys the same transit from farther away, over a longer backhaul circuit it also has to pay for. The per-bit cost can be an order of magnitude higher.
Internet exchange points are where the peering actually happens physically: a colocation facility where networks plug into a common switching fabric and swap routes. The largest exchanges in the United States — Equinix Ashburn, CoreSite in Los Angeles, the carrier-neutral facilities in the New York area — handle enormous aggregate volumes precisely because their density makes it economical for networks to show up and peer. A regional ISP that pays for a cross-connect at one of these facilities can peer with dozens of content networks directly, reducing the transit it needs to buy. That is the economic logic behind exchange points, and it is why their addresses matter to the cost structure of every network that is too far from one to participate cheaply.
The content giants further complicate the picture. Google, Meta, Netflix and Amazon have built private backbone networks large enough to peer directly with major ISPs — sometimes installing their own caching hardware inside an ISP's facility, a practice called content delivery network peering or simply CDN placement. When Netflix traffic is served from a box in an ISP's own headend, that traffic never crosses a transit link at all. The ISP saves on transit costs; Netflix saves on egress fees; the subscriber gets lower latency. The arrangement sounds frictionless, but it rests on a negotiated agreement that can collapse — as it did publicly between Netflix and several large ISPs during 2013 and 2014, when congestion at interconnect ports became visible in subscriber throughput data.
Understanding where a given ISP sits in this hierarchy — whether it peers, who it buys transit from, and how far its traffic travels before reaching an exchange — is as important to actual network performance as anything happening in the last mile. A well-provisioned fiber drop still delivers a degraded experience if the path beyond the headend is constrained by a transit bill someone decided to minimize.