Peering and transit, the economics of moving bits

BGP determines how packets find paths. Whether two networks have a path between them is a business decision, not a technical one. The two main types of business relationship are peering and transit. Understanding them explains why some routes exist, why others don't, and why your traffic to a specific service might be slow.

The simplest model

Network A wants its users to reach network B. Two basic ways to make this happen:

Transit. A pays B (or some upstream) to carry A's traffic to the rest of the internet, including B's customers. Money flows from A.

Peering. A and B agree to exchange traffic directly between their respective customers, without anyone paying. Money does not flow.

This is the entire framework. Everything else is variation on these two patterns.

Transit

When a network wants reachability to "everything else on the internet" without negotiating individually with thousands of other networks, it buys transit.

A transit provider is a network that:

• Announces routes to most of the internet (often via its own upstreams or extensive peering).
• Accepts traffic from its customers destined anywhere.
• Charges per amount of traffic (usually $/Mbps committed plus burst, sometimes per-GB).

Common transit prices in 2026:

• Premium transit (low latency, well-peered): $0.30 to $0.80/Mbps committed.
• Standard transit: $0.10 to $0.30/Mbps.
• Discount transit: $0.05 to $0.15/Mbps.

For a hosting company pushing 1 Gbps committed, premium transit costs $300 to $800 per month. Standard transit, half that. Cheap transit, less.

Transit provides the "everything" reachability. Without it, you'd need a relationship with every other network you wanted to reach. With it, you connect once and reach all.

Peering

When two networks send each other significant traffic (and the traffic is roughly balanced), they can save money by exchanging directly. No transit provider in the middle.

Peering can be:

Settlement-free. Most common. No money changes hands. Both sides benefit.

Paid peering. Less common. One side pays the other, usually because the relationship is asymmetric (one side gets more value).

Peering reduces latency (one fewer hop, often physically shorter path) and reduces transit costs (you don't pay anyone for that traffic).

Why networks peer

The math:

A small ISP buys transit at $0.50/Mbps to push 500 Mbps of traffic to a major streaming service. That's $250/month in transit.

If they can peer directly with the streaming service at an IXP, they pay only the IXP membership ($500/month flat) and the cross-connect ($100/month). For their 500 Mbps to the streaming service, peering is breakeven at first but scales: the next 500 Mbps of streaming traffic is free, not $250 more.

For networks moving substantial traffic to specific destinations, peering wins quickly.

Settlement-free peering criteria

Networks have policies about who they'll peer with for free. Typical large-network requirements:

• Geographic presence: "Must have presence in 4+ of our regions."
• Traffic ratio: "Outbound to inbound ratio between 1:2 and 2:1." (You can't just be a content sink.)
• Network size: "Must announce 1000+ prefixes." (Small networks don't qualify.)
• Peering capacity: "Must have 10+ Gbps available at each location."
• Operational standards: "Must have a 24/7 NOC, MANRS compliance."

If you meet these, peering is a polite negotiation. If you don't, you can sometimes still peer (paid peering, asymmetric agreements) or you can grow into qualification over time.

Some networks publish their peering policies; some keep them private. PeeringDB has standardized info on many networks.

Tier 1, 2, 3 again

In the ASN article, we mentioned the tiers. Now they make business sense:

Tier 1: Has settlement-free peering with all other Tier 1s. Buys no transit. Reaches the entire internet via peering. About a dozen worldwide.

Tier 2: Has some peering. Buys some transit from Tier 1s (or other large networks). Most large ISPs.

Tier 3: Buys transit. Doesn't peer broadly. Smaller networks, regional ISPs.

The Tier 1 status is impressive but not always commercially better. A Tier 2 with good peering arrangements can have better real-world performance than a Tier 1 with sparse peering in your region.

IXPs

An Internet Exchange Point (IXP) is a physical facility where many networks connect to a shared switching infrastructure. Members can peer with each other over the IXP, with one cross-connect to the IXP fabric instead of many bilateral cross-connects.

Major IXPs by traffic (2026):

• DE-CIX (Frankfurt). Largest in the world by traffic.
• AMS-IX (Amsterdam).
• LINX (London).
• DEC-IX New York.
• HKIX (Hong Kong).
• PTT.br SP (São Paulo).
• Equinix Internet Exchange (multiple locations).
• Megaport (multi-location cloud-friendly IXP).

Joining an IXP costs:

• Membership fee ($500 to $5000/month depending on size and location).
• Cross-connect fee to the IXP's switch fabric ($50 to $500/month).
• Your own router and transit to the facility.

For a network with significant traffic at the IXP's region, this is a strong value proposition. The hosting company that pays $250/month for transit could probably reduce that by 50 percent or more by joining the local IXP.

What "good peering" means for users

If you're a customer of a host, the host's peering directly affects your experience:

Latency. Better peering means fewer hops to common destinations. Cloudflare, Google, Netflix, Steam, etc. all peer extensively; if your host peers with them, your traffic to those services is short.

Reliability. Peering provides path diversity. If one upstream has an outage, peered paths still work.

Cost (indirect). Hosts with good peering have lower transit bills and can offer competitive prices. Hosts with poor peering pass through higher costs.

When evaluating a host, you can look at their PeeringDB entry. It shows where they have presence and who they peer with. A host with extensive peering in your region is usually a better choice.

The big content providers

Companies like Google, Cloudflare, Facebook, Amazon, Microsoft, Netflix, Apple operate global networks at scale comparable to (or larger than) traditional Tier 1s.

These networks:

• Have hundreds to thousands of peering locations.
• Connect to most IXPs worldwide.
• Often offer free peering to anyone who can physically connect.
• Sometimes deploy "embedded caches" inside ISPs (so the ISP doesn't even need internet transit to reach the content).

Netflix's Open Connect, Google's Global Cache, and similar programs put physical hardware inside ISPs. For the ISP, this means Netflix traffic doesn't traverse their transit. For the content provider, this means cheap, fast delivery to that ISP's customers.

This is one reason your streaming service feels fast: the content is often physically inside your ISP's network.

When peering goes wrong

Disputes happen. A few patterns:

De-peering. Network A and Network B decide to stop peering. Suddenly their direct path goes away; traffic between them must traverse transit. Latency increases for both sides' users. Famous cases include various Verizon vs Cogent and Comcast vs Cogent peering disputes.

Asymmetric peering negotiations. One side wants payment; the other doesn't want to pay. Traffic gets congested at peering points while they negotiate.

Outages at IXPs. If a major IXP has a problem, all peering through it stops. Networks fall back to transit, often with higher latency.

These are usually invisible to users until they aren't. A specific service feeling slow for weeks might trace to a peering dispute somewhere.

Cross-border and submarine cables

Peering and transit need physical paths. International traffic crosses on submarine cables, owned by consortia of telcos and increasingly by hyperscalers (Google, Meta, Amazon).

Major cable systems:

• Transatlantic: TAT-14, MAREA, Grace Hopper, Dunant. Connect Europe and the Americas.
• Transpacific: New Cross Pacific, FASTER, Topaz. Connect Asia, Oceania, and the Americas.
• Mediterranean: SEA-ME-WE 5, AAE-1. Connect Europe, Middle East, Asia.
• Atlantic-Pacific: Various Africa-Europe-South America-Asia routes.

When a cable cuts (anchors, earthquakes, fishing trawlers), latency to entire regions can change. This happened to Northern Africa during a 2024 cable cut.

Bandwidth on these cables is allocated to operators by contract or by ownership. Your latency to a far destination depends on which cable systems your provider has capacity on.

Real numbers from a hosting company

Without naming specifics, a small to mid-size hosting company might have:

• Two transit providers (paying $0.20/Mbps each).
• Three to five IXP memberships.
• Direct settlement-free peering with major content providers via the IXPs.
• Total bandwidth bill: 30 to 50 percent of what it would cost with pure transit.

The fixed costs of peering (memberships, cross-connects) are higher, but the per-bit costs are much lower. For a network of any scale, peering wins.

What hosting customers should look for

When evaluating a host:

• Number of upstreams. More is better (redundancy, peering diversity).
• IXP memberships. Visible in PeeringDB. The more, the better in your region.
• Latency to common services. Test by ping to services like Google DNS (8.8.8.8), Cloudflare (1.1.1.1) from a server.
• Geographic position. A host near your users beats a host far from them.

Hosts that emphasize "our network" in marketing usually back it up with this. Hosts that don't mention it usually don't have much to show.

Conclusion

Peering and transit are the business relationships under BGP. They determine which paths exist. Transit is the "buy reachability to everyone" option. Peering is the "exchange directly with specific partners" option. Most networks use both.

For users, the experience of "this site feels fast" or "the connection feels reliable" often traces back to peering. Networks with good peering deliver better experiences. Networks with sparse peering deliver worse ones.

Hosting choice reflects this. Pick a host that has invested in their network, and your experience will reflect that investment.

Coming up

We've covered the business side of internet connectivity. The next article digs into the physical / logistical layer: Internet Exchange Points, where peering happens in the real world.