The Economics of Internet Peering

The internet is often described as a "network of networks," but that framing glosses over a critical dimension: money. Every connection between autonomous systems is the product of a business decision. Networks exchange traffic because doing so is economically rational, and the terms of those exchanges — who pays whom, how much, and under what conditions — determine the physical topology of the internet. Understanding peering economics is essential to understanding why traffic takes the paths it does, why some routes are shorter than others, and why outages sometimes cascade across continents.

Two Fundamental Relationships: Peering and Transit

There are exactly two commercial models for interconnection on the internet. Every link between two autonomous systems is governed by one of them.

Settlement-free peering is a bilateral agreement in which two networks exchange traffic destined for each other's customers at no charge. Neither side pays the other. Each network announces only its own prefixes and its customers' prefixes to the peer — not routes learned from other peers or transit providers. The economic logic is simple: if two networks send roughly equal amounts of traffic to each other's customers, it would be inefficient for money to change hands in both directions. They cancel out and agree to carry each other's traffic for free.

Paid transit is a customer-provider relationship. The customer pays the transit provider to carry its traffic to and from the rest of the internet. In exchange, the transit provider announces the customer's prefixes to all of its peers, customers, and upstream providers — giving the customer global reachability. The transit provider also provides a default route, meaning the customer can reach any destination on the internet through the provider.

These two models produce radically different AS paths. A peering link between two networks creates a two-hop path: your network and theirs. A transit chain adds intermediate hops. When you look up an IP in the looking glass and see a short AS path, it often means the origin network has extensive peering. When you see a longer path threading through networks like Lumen (AS3356) or NTT (AS2914), you are seeing transit in action.

The Tier Hierarchy: Who Pays Whom

The internet's interconnection structure forms an economic hierarchy. The terminology — Tier 1, Tier 2, Tier 3 — is informal but widely understood, and it maps directly to who writes checks to whom.

Tier 1 Networks

A Tier 1 network can reach every destination on the internet without purchasing transit from anyone. It achieves this by maintaining settlement-free peering agreements with every other Tier 1 network. Because Tier 1 networks peer with each other at no cost and collectively cover the entire routing table, none of them needs to pay an upstream provider. Notable Tier 1 networks include Lumen/Level 3 (AS3356), Arelion/Telia (AS1299), NTT (AS2914), Telecom Italia Sparkle (AS6762), Cogent (AS174), and Zayo (AS6461).

The definition is precise: if you need to buy transit from even one other network to reach any destination, you are not Tier 1. There are roughly a dozen Tier 1 networks globally, and the list changes slowly. When a new entrant reaches Tier 1 status, it means they have successfully negotiated settlement-free peering with every existing Tier 1 — a process that can take years and considerable leverage.

Tier 2 Networks

A Tier 2 network purchases transit from one or more Tier 1 providers to reach the full internet, but also maintains settlement-free peering with other networks to reduce its transit costs. Most large regional ISPs, hosting companies, and content providers operate at this tier. They peer aggressively at IXPs and via private interconnects to shift traffic off expensive transit links.

For a Tier 2 network, the economic calculus is straightforward: every megabit per second that can be moved from a paid transit port to a settlement-free peering port is money saved. This is why peering coordinators at Tier 2 networks spend their days negotiating new peering arrangements — the savings are real and significant.

Tier 3 Networks

Tier 3 networks are typically small ISPs, enterprises, and access networks that buy all of their connectivity from upstream providers and do little or no peering. They are transit customers exclusively. A small regional ISP that purchases a 10 Gbps port from a Tier 2 provider and resells connectivity to end users is a Tier 3 network. They appear at the leaf edges of the AS graph.

Peering Requirements and Ratio Policies

Settlement-free peering is not automatic. Networks evaluate potential peers against a set of requirements, and these requirements are a direct expression of economic self-interest.

Common Peering Requirements

A typical large network's peering policy might require potential peers to meet criteria such as:

• Minimum traffic volume — Often stated as a minimum committed traffic rate, such as "at least 500 Mbps of peak traffic exchange." Small networks that would send only a trickle of traffic are not worth the operational overhead of maintaining a BGP session.
• Geographic diversity — Many policies require peers to interconnect at multiple locations (e.g., "at least 3 points of presence in 2 or more regions"). This ensures that traffic can be exchanged close to where it originates, rather than being backhauled across a continent.
• Operational competence — 24x7 NOC (Network Operations Center) staffing, up-to-date PeeringDB records, functional abuse contacts, and an IRR-registered routing policy. Nobody wants to peer with a network that cannot respond to incidents.
• Mutual benefit — The implicit requirement underpinning all others. If one side sends vastly more traffic than it receives, the relationship is asymmetric and the receiving side is effectively providing free transit.

Traffic Ratio Policies

The most contentious peering requirement is the traffic ratio. Many networks stipulate that the ratio of inbound to outbound traffic must remain within a certain range — commonly between 1:1 and 2:1. If the ratio exceeds this threshold, the network sending more traffic may be asked to either pay for the excess or have the peering agreement terminated.

This policy emerged because peering was originally conceived for networks with roughly symmetric traffic patterns. When two ISPs exchange traffic, their customers communicate in both directions. But the rise of content providers fundamentally disrupted this symmetry: a network like Netflix sends enormous amounts of data outbound but receives very little inbound. A traditional ISP peering with Netflix would see a ratio of perhaps 100:1 — almost all traffic flowing from Netflix into the ISP's network.

Whether this asymmetry justifies payment is the central theological debate of internet peering economics. Content providers argue they are making ISPs' networks more valuable to end users. ISPs argue they bear the cost of delivering that content across their last-mile infrastructure. Both sides have a point, and the dispute has driven some of the most consequential depeering events in internet history.

Peering Policies: Open, Selective, and Restrictive

Networks publish their peering policies on PeeringDB, the de facto registry for interconnection information. These policies fall into three broad categories:

• Open peering — The network will peer with anyone who meets basic technical requirements. Hurricane Electric (AS6939) is the canonical example of an open peering policy. They peer with almost anyone, at almost any location, with minimal requirements. This strategy maximizes route diversity and minimizes transit costs.
• Selective peering — The network evaluates each peering request individually against a set of published criteria. Most large ISPs and content providers operate under selective policies. They peer where it makes economic sense and decline where it does not.
• Restrictive peering — The network sets high bars for peering and declines most requests. Some Tier 1 networks historically maintained restrictive policies, peering only with other Tier 1 networks and requiring everyone else to purchase transit.

Depeering and Peering Disputes

When two networks cannot agree on peering terms, the result is depeering — the termination of a peering relationship. Depeering is the nuclear option of internet interconnection, and it has caused real disruptions.

Cogent vs. Sprint (2008)

In October 2008, Cogent (AS174) and Sprint (AS1239, now part of T-Mobile) terminated their settlement-free peering agreement. The dispute was over traffic ratios and financial terms. When the peering link went down, customers of Cogent could not reach customers of Sprint, and vice versa. The internet effectively split into two partitions for traffic between those networks and their downstream customers.

The outage was resolved within days as both networks faced pressure from customers, but it demonstrated a critical point: the internet's connectivity is not guaranteed by any authority. It is the product of thousands of bilateral agreements, and when those agreements fail, reachability fails with them.

Cogent vs. Telia (2008)

In the same year, Cogent was involved in a similar dispute with Telia (AS1299). Cogent has historically been involved in multiple depeering events because its business model — offering very low-cost transit — creates traffic asymmetries that peers find unacceptable. Cogent's customers are disproportionately content-heavy, meaning Cogent sends much more traffic than it receives from backbone peers.

Netflix vs. ISPs (2013-2014)

The most publicly visible peering dispute involved Netflix and multiple US ISPs, including Comcast, Verizon, AT&T, and Time Warner Cable. Netflix traffic was growing to represent over 30% of all downstream internet traffic in North America. ISPs argued that Netflix should pay for the infrastructure needed to handle this volume. Netflix argued that ISPs were deliberately allowing interconnection points to become congested to extract payment.

The dispute played out publicly through degraded Netflix performance for ISP customers. Netflix eventually agreed to paid interconnection arrangements with major ISPs, but the episode helped drive the US net neutrality debate and illustrated how peering economics directly affect end-user experience.

How Cloud Providers Approach Peering

The major cloud and infrastructure providers have adopted distinct peering strategies that reflect their business models.

Google

Google (AS15169) operates one of the most extensive peering networks on the planet. Google maintains edge nodes and caches in thousands of ISP networks worldwide (the Google Global Cache program), and peers aggressively at IXPs and via private interconnects. Google's peering policy is generally considered selective but generous — they will peer with networks that meet reasonable traffic thresholds, and they invest heavily in direct interconnection to reduce latency for their services.

Google's approach is driven by the fact that every millisecond of latency costs them revenue. Shorter paths mean faster page loads, which means more searches, more ad clicks, and more YouTube streams. The investment in peering and edge caching pays for itself many times over.

Amazon Web Services

Amazon (AS16509) takes a more traditional approach. AWS peers selectively and maintains paid transit relationships with Tier 1 providers. AWS charges customers for data transfer out of its cloud, so there is less incentive to minimize transit costs by peering freely — data transfer fees are a significant revenue stream. AWS does peer at major IXPs and via private interconnects, but the bar is higher than Google's.

Cloudflare

Cloudflare (AS13335) is perhaps the most aggressively peered network among content and infrastructure providers. Cloudflare peers at over 300 IXPs worldwide and maintains an open peering policy — they will peer with any network, at any location, with no traffic requirements. This strategy makes sense for Cloudflare's business model: as a CDN, DDoS mitigation provider, and DNS resolver (operating 1.1.1.1), Cloudflare's value proposition depends on being as close as possible to every user on the internet.

Cloudflare has been vocal about the concept of "bandwidth as a commodity" and has criticized ISPs in some regions (particularly in parts of Europe and Australia) for demanding payment for peering even when the traffic exchange is mutually beneficial. Their "bandwidth alliance" program attempts to reduce or eliminate data transfer costs between Cloudflare and partner cloud providers.

Meta (Facebook)

Meta (AS32934) operates a large private backbone and peers extensively. Like Google, Meta invests in peering to reduce latency for its social media and messaging platforms. Meta also operates edge caches that can be deployed inside ISP networks, similar to Google Global Cache.

Internet Exchange Points: The Peering Marketplace

Internet Exchange Points are the physical infrastructure that makes peering practical. Without IXPs, every peering relationship would require a dedicated private circuit between two networks — an arrangement that does not scale. IXPs provide a shared fabric where hundreds of networks can interconnect, each needing only a single physical connection to reach all other participants.

How IXPs Reduce Transit Costs

Consider a regional ISP paying $1 per Mbps per month for transit. If they exchange 5 Gbps of traffic with a content provider, that costs $5,000/month through their transit provider. By peering directly with the content provider at an IXP — where the monthly port fee might be $500 for a 10 GbE connection — they eliminate that transit cost entirely. The savings compound as they add more peers at the same IXP, all sharing the same physical port.

The largest IXPs handle staggering volumes. DE-CIX Frankfurt (AS6695) regularly exceeds 14 Tbps of peak traffic across more than 1,100 connected networks. AMS-IX Amsterdam (AS1200) peers a similarly large community. Each terabit exchanged at the IXP is a terabit that does not flow over paid transit links.

The Economics of Running an IXP

Most IXPs operate as non-profit associations or cooperatives, funded by membership and port fees. A typical IXP's cost structure includes:

• Switching infrastructure — High-capacity Ethernet switches forming the peering fabric. Modern IXPs use 100 GbE and 400 GbE switches.
• Data center space and power — Colocation in carrier-neutral facilities where many networks already have equipment.
• Route servers — Servers running BGP route server software (typically BIRD or OpenBGPD) that simplify peering by letting participants exchange routes without bilateral sessions.
• Operations staff — NOC engineers, peering coordinators, management.
• Redundancy and resilience — Redundant switches, power, and often multiple physical sites connected by dark fiber.

Revenue comes from port fees, which are typically structured by port speed. A 1 GbE port might cost $200-500/month, a 10 GbE port $500-2,000/month, and a 100 GbE port $2,000-8,000/month, depending on the exchange. Some IXPs also charge a one-time joining fee. The business model works because the cost of a port is shared across all the peering relationships it enables — a single 10 GbE port might support BGP sessions with 200+ networks via route servers.

Some IXPs also generate revenue through value-added services: private VLANs for dedicated peering, DDoS scrubbing, reseller programs, and remote peering services that let networks connect without physically colocating at the exchange.

How CDNs Changed the Peering Landscape

The rise of content delivery networks in the 2000s fundamentally restructured internet peering economics. Before CDNs, content was served from origin servers in a few locations, and traffic flowed through the transit hierarchy to reach end users. CDNs inverted this model by pushing content to the edge, placing servers inside or adjacent to the networks where users connect.

The Shift from Backbone to Edge

In the traditional model, a web request from a user in Sydney to a server in San Francisco would traverse the user's ISP, one or more transit providers, a trans-Pacific submarine cable, and the server's hosting provider. Every one of those transit links had a price attached. CDNs eliminated most of this chain by caching the content at an edge node in Sydney, served directly to the ISP via a local peering connection.

This shift had cascading economic effects:

• Reduced transit revenue for backbone providers. As more traffic moved to local exchanges between CDNs and ISPs, Tier 1 providers saw less long-haul traffic and less transit revenue. This is part of what has driven transit prices steadily downward over the past two decades — from $100+ per Mbps/month in the early 2000s to under $0.50 per Mbps/month on competitive routes today.
• Increased peering importance for ISPs. ISPs that peer effectively with CDNs deliver better performance to their customers. An ISP that refuses to peer with Cloudflare or Google and instead receives their content through a distant transit link will have noticeably worse latency for its users.
• Changed the traffic ratio equation. CDN traffic is inherently asymmetric — it flows almost entirely from CDN to ISP. This challenged the traditional peering model based on roughly symmetric traffic exchange and fueled disputes about whether CDNs should pay ISPs for delivering their content.

Content Provider Peering Programs

The largest content providers have gone beyond traditional peering and CDN arrangements to deploy infrastructure directly inside ISP networks. These programs are explicitly designed to reduce transit costs to zero for the heaviest traffic flows.

Netflix Open Connect

Netflix (AS2906) operates the Open Connect program, through which it deploys dedicated cache appliances inside ISP networks at no cost to the ISP. These appliances — custom hardware running FreeBSD — store the Netflix content catalog and serve streams directly to the ISP's subscribers without the traffic ever crossing an interconnection point.

The economics are compelling for both sides. Netflix eliminates transit costs entirely for participating ISPs — its content is served from inside the ISP's network. The ISP eliminates the bandwidth cost of carrying Netflix traffic across its peering and transit links, which can be substantial: Netflix has represented 15-20% of all downstream traffic in North America during peak evening hours. The only cost to the ISP is rack space and power for the appliance.

Open Connect appliances serve over 90% of Netflix traffic globally. The program's success has made the 2013-2014 peering disputes with ISPs largely moot — by embedding inside ISP networks, Netflix bypassed the interconnection bottleneck entirely.

Google Global Cache and Edge Nodes

Google (AS15169) runs two complementary programs. Google Global Cache (GGC) deploys cache servers inside ISP networks, similar to Netflix Open Connect. These servers cache YouTube content, Google Play downloads, and other static assets. Google Edge Nodes are larger deployments in carrier-neutral facilities that serve a broader range of Google services.

Combined, these programs mean that for most users worldwide, Google's content is served from within their ISP's network or from a facility directly peered with it. The traffic never touches a transit provider. Google's investment in this infrastructure is massive — they operate one of the largest private backbone networks on the planet, with submarine cables, terrestrial fiber, and edge deployments in over 200 countries.

Meta and Apple

Meta (AS32934) operates a similar edge caching program for Facebook, Instagram, and WhatsApp content. Apple (AS714) has built out its own CDN for iCloud, App Store, Apple TV+, and software updates, deploying edge caches in ISP networks globally. The trend is clear: every major content provider is converging on the same strategy of embedding inside access networks.

Regional Variations in Peering Economics

Peering economics vary significantly by geography, and these variations have real consequences for end users.

North America and Europe

These regions have the most mature peering ecosystems, with dozens of IXPs, low transit prices, and a strong culture of settlement-free peering. Competition among transit providers has driven prices to commodity levels. An ISP in Frankfurt or Ashburn can choose from many IXPs and transit providers, keeping costs low and performance high.

Australia

Australia has historically been an outlier among developed markets. The dominant incumbent, Telstra, has maintained a restrictive peering policy and high transit prices relative to other markets. This has created situations where traffic between two networks in the same Australian city might route via the United States because the domestic peering or transit path is unavailable or too expensive. Australian peering economics have improved in recent years with the growth of exchanges like IX Australia, but they remain atypical.

Africa and Developing Regions

In many African countries, international transit costs remain high because submarine cable capacity is limited and expensive. Local IXPs, such as the Kenya Internet Exchange Point (KIXP) and the Nigeria Internet Exchange (IXPN), play an outsized role in these markets by keeping local traffic local. Without IXPs, a packet traveling between two networks in Nairobi might route through London or Amsterdam, incurring intercontinental transit costs and hundreds of milliseconds of additional latency.

Transit Pricing Trends

The economics of transit have shifted dramatically over the past two decades, and the trajectory is still downward.

In competitive markets like Ashburn (Virginia), London, Frankfurt, and Amsterdam, transit prices have fallen below $0.50 per Mbps/month for high-volume commitments on 10 GbE or 100 GbE ports. This is a decline of over 99% from the $100+ per Mbps/month that was common in the early 2000s.

Several forces have driven this decline:

• Fiber overbuilding — The late-1990s telecommunications bubble left a glut of dark fiber that took over a decade to absorb, but it dramatically lowered the cost of long-haul transport.
• Moore's Law for networking — Router and switch capacity has grown exponentially while per-port costs have fallen. A 100 GbE port costs less today than a 1 GbE port did in 2005.
• CDN disintermediation — As content providers moved traffic off backbone transit links and onto local peering, demand for long-haul transit decreased, putting downward pressure on prices.
• IXP growth — The proliferation of IXPs has given networks more options for peering, further reducing reliance on paid transit.
• Competition — Low barriers to entry for transit providers in major markets mean aggressive price competition.

However, transit prices remain high in underserved regions. In parts of Africa, Central Asia, and the Pacific Islands, transit can still cost $20-50+ per Mbps/month due to limited submarine cable capacity and less competitive markets.

Paid Peering: The Middle Ground

Between settlement-free peering and full transit lies paid peering. In this model, one network pays the other for a peering relationship, but the scope of routes exchanged is limited — typically only the receiving network's customer routes, not a full routing table. Paid peering is common in situations where the traffic exchange is highly asymmetric but both parties benefit from direct interconnection.

Paid peering became prominent through ISP-CDN disputes. An ISP might refuse settlement-free peering with a CDN because the traffic ratio is 10:1, but full transit pricing would be excessive for the limited set of routes exchanged. Paid peering threads the needle: the CDN pays a fraction of what transit would cost, and the ISP receives compensation for the asymmetric traffic load.

The exact terms of paid peering arrangements are typically confidential. Estimates suggest that paid peering rates are typically 10-30% of equivalent transit rates, though this varies widely by market and negotiating leverage.

The Future of Peering Economics

Several trends are reshaping peering economics:

• Regulatory intervention — South Korea and some European countries have introduced or proposed "sending party pays" regulations that would require content providers to pay ISPs for traffic delivery. This remains controversial and its long-term impact is uncertain. The EU's attempt at similar regulation under the European Data Act generated significant debate before being set aside.
• Consolidation — Mergers among ISPs and content providers reduce the number of independent peering decisions. When a large ISP acquires a content company, internal traffic that previously crossed a peering link now stays within a single AS.
• 400 GbE and beyond — As port speeds increase, the per-bit cost of peering continues to fall, making settlement-free peering practical for ever-smaller networks.
• Remote peering — Technologies that allow networks to connect to IXPs without physically colocating are expanding access to peering, particularly for networks in underserved regions.
• Subsea cable investments by content providers — Google, Meta, Microsoft, and Amazon now own or co-own a growing share of the world's submarine cable capacity. This vertical integration further reduces their dependence on third-party transit and reshapes the bargaining dynamics of peering.

Observing Peering Economics in BGP Data

You can see the effects of peering economics directly in BGP routing data. When you look up an ASN in the looking glass, the neighbor relationships reveal the network's position in the economic hierarchy. A network with many upstream providers is paying for transit. A network that appears as an upstream of many others is selling transit. Networks connected laterally, with short AS paths between them, are likely peers.

Try exploring these networks to see different peering profiles:

• AS3356 (Lumen/Level 3) — A Tier 1 transit provider with thousands of downstream customers and peering with all other Tier 1 networks
• AS13335 (Cloudflare) — One of the most heavily peered networks, with minimal transit and extensive IXP presence
• AS6939 (Hurricane Electric) — Famous for its open peering policy and one of the most connected ASes by number of peers
• AS174 (Cogent) — A large transit provider known for aggressive pricing and peering disputes
• AS2906 (Netflix) — A major content source that has shifted from transit to embedded caching
• AS15169 (Google) — Extensive peering plus embedded cache infrastructure inside thousands of ISP networks

The AS paths you see in the routing table are, ultimately, the product of economic decisions. Every hop in a path represents a business relationship — a contract, a handshake, or a port fee. Understanding peering economics is understanding why the internet's routing table looks the way it does.