Anycast Routing Contracts Favor the Largest CDN Buyers

Anycast routing is sold as a neutral technical choice: one IP address, many locations, packets go to the nearest one. In practice the routing table is a contract, and the contract favors whoever owns the most real estate. This piece explains how anycast economics tilt toward large CDN buyers and what smaller networks can still put in writing.

Anycast Rewards Scale Not Neutrality

Anycast is an addressing and routing method where a single IP address is announced from many locations at once. Routers forward each packet to whichever announcement their normal BGP decision process prefers, usually the one with the fewest network hops. The protocol has no opinion about fairness. It picks the shortest path it can see.

That design turns geographic reach into a routing advantage. A CDN with hundreds of points of presence can announce the same prefix from dozens of cities, so almost every end user sits a short hop from one of them. A smaller CDN with three or four locations cannot match that density, no matter how good its software is.

The result is a self-reinforcing loop. More PoPs means better anycast performance, which wins larger content contracts, which funds more PoPs. The mechanism is not hidden; it is just rarely stated plainly in a sales deck.

Consider Cloudflare's public anycast network, which by its own documentation spans hundreds of cities. That scale lets it announce a single prefix from dozens of locations, ensuring most users hit a nearby PoP. A regional CDN with a handful of sites simply cannot replicate that density. The performance gap is structural, not a matter of engineering skill.

Peering Contracts Encode Anycast Advantage

Internet exchange points like AMS-IX in Amsterdam exist so networks can exchange traffic without paying each other. They are typically non-profit and neutral, and they made settlement-free peering the default for well-connected networks. The catch is that "well-connected" is doing a lot of work in that sentence.

Large CDNs peer privately at scale, connecting directly into access networks and bypassing exchanges entirely. Those private interconnects carry no per-bit fee and often come with better routing preference. Smaller CDNs usually cannot get the same deals because they do not bring enough traffic to the table to interest a large access network.

So the smaller CDN buys transit instead, paying a carrier for reach it cannot peer for. That cost lands in the contract it signs with its own customers. The routing asymmetry and the contract asymmetry are the same fact viewed from two angles. A related piece on rebalance cost at scale makes a similar point about infrastructure economics: the protocol is neutral, the bill is not.

CDNI Interconnection Promises But Underdelivers

Content delivery network interconnection, or CDNI, is a set of interfaces that lets two independent CDNs hand traffic to each other. The pitch is straightforward: a small CDN extends its footprint by borrowing a partner's, and content providers get more redundancy without buying from one giant.

The technical interfaces exist and are reasonably well specified. The commercial incentives mostly do not. A large CDN has little reason to federate with a rival that could use the arrangement to win accounts. Footprint extension tends to favor the bigger partner, which already has the reach and can set terms.

CDNI shows up in standards documents and pilot projects more often than in production contracts. That gap is the story. Interconnection is a business decision wearing a protocol's clothes, and the business case for the incumbent is thin.

The Hidden Costs of Anycast for Small Buyers

Anycast routes can shift mid-session. A BGP withdrawal, a peering dispute, or a link failure can move a flow to a different PoP while TCP state lives on the old one. The connection resets. For a large buyer with redundant contracts, that is a blip absorbed by a second provider. For a small buyer on a single CDN, it is an outage.

Service-level agreements rarely cover the revenue lost during a reset. Credits are typically calculated against the monthly bill, not against the sales the downtime cost. A buyer moving a few thousand dollars a month through a storefront can lose more in one bad hour than the credit returns in a year.

Large buyers negotiate around this by design. They run multi-CDN setups, hold failover contracts, and treat anycast instability as a routing problem to be engineered out. Smaller buyers inherit the same protocol without the same leverage.

There is also an operational trade-off for providers. Anycast can concentrate traffic onto a single PoP during a route change, creating hot spots that degrade performance for everyone in that region. Large CDNs mitigate this with sophisticated traffic engineering and real-time telemetry, but those tools are expensive and require staff that smaller providers cannot justify. The result is that anycast's failure modes are themselves scale-dependent: the big player absorbs the shock and reroutes; the small player absorbs the outage and explains it to customers.

What Edge Buyers Can Actually Negotiate

Contract terms are where anycast risk becomes manageable. A buyer who asks for the right clauses before signing has more room than one who discovers the problem during an incident.

• Require anycast prefix diversity, so a single PoP failure cannot take the whole service down.
• Ask for BGP community visibility, which lets your network engineers see and influence path selection.
• Contract for explicit failover to unicast addresses when anycast routing misbehaves.
• Push for CDNI-style interconnect clauses, even if the provider currently has no federation partners.

None of these guarantees performance. They do move the burden of proof onto the provider, which is the point. A vendor that cannot document prefix diversity is telling you something about its architecture.

Practical Moves for Smaller CDN Customers

Map your anycast prefixes across every provider you use, and record which PoPs announce them. That inventory is the baseline for every failover conversation you will have later.

Test failover with synthetic traffic before you need it. Injecting a route withdrawal in a controlled window tells you more than any SLA document. Then negotiate peering transparency into the contract itself, so routing changes are disclosed rather than discovered.

Join or form a regional internet exchange point if one is nearby. Even modest exchanges reduce transit spend and give smaller networks a place to peer on neutral terms. Budget for multi-CDN redundancy as a line item, not an emergency expense. The cost of a second provider is usually smaller than the cost of a single point of failure, and the second contract is the only real leverage a small buyer holds.

Why Anycast Economics Persist

The tilt toward scale is not an accident to be patched. Anycast works by letting the routing system pick the topologically closest announcement, and topological closeness correlates with physical build-out. A network that has spent a decade placing racks in carrier hotels across six continents will win the shortest-path race by default. That advantage compounds because the same build-out that improves anycast performance also improves cache hit ratios and reduces origin egress, both of which lower the CDN's own unit costs. The large buyer sees those savings in the form of lower per-GB rates, while the small buyer sees only the list price.

There is a countervailing force, though a weak one. Anycast routing can concentrate traffic onto a single PoP during a route change, creating hot spots that degrade performance for everyone in that region. Large CDNs mitigate this with sophisticated traffic engineering and real-time telemetry, but those tools are expensive and require staff that smaller providers cannot justify. The result is that anycast's failure modes are themselves scale-dependent: the big player absorbs the shock and reroutes; the small player absorbs the outage and explains it to customers.

For buyers, the practical lesson is that anycast is not a feature you purchase in isolation. It is a property of the entire interconnection strategy, and the contract is the only place where a smaller network can claw back leverage. The clauses that matter are the ones that force disclosure of routing policy and guarantee a path to unicast when anycast misbehaves. Without those, the buyer is simply trusting that the provider's incentives align with its own. They rarely do, because the provider's incentive is to maximize the value of its own footprint, not to subsidize a competitor's reach.