An internet exchange point (IXP) is a shared physical infrastructure, typically a layer 2 Ethernet switching fabric in one or more data centers, where independent networks connect to exchange traffic directly with each other. In internet fundamentals, IXPs are the main places where autonomous systems meet for peering, so that traffic between them does not have to travel through paid upstream transit providers.
An IXP does not route traffic itself and does not sell internet access. It provides ports, a switch fabric and usually route servers, while each member network decides with whom it exchanges routes. By keeping traffic local, IXPs reduce latency, lower costs and add resilience to the global internet.
How an internet exchange point works
Each member network installs a router in a data center where the IXP is present and connects it to the IXP switch fabric through a port, commonly 10, 100 or 400 Gbit/s Ethernet. The IXP assigns each member an address from a shared peering LAN, an IPv4 and IPv6 subnet on which all members can reach each other directly.
Over that shared LAN, members establish BGP sessions. Through BGP, each network announces the address prefixes of itself and its customers. Once two networks exchange routes, packets between them cross the switch fabric in a single hop, without passing through any third network.
Large exchanges span several data centers in a metropolitan area, linked by dark fiber or optical transport, so that members in different buildings still share one logical fabric. The fabric itself is usually built with redundant switches and, in larger IXPs, with technologies such as VXLAN with EVPN to keep the layer 2 domain stable.
Peering models at an exchange
IXPs support several ways of exchanging routes. The choice affects operational effort, control and scale, and most members combine more than one model.
| Model | How it works | Typical use |
|---|---|---|
| Multilateral peering via route servers | One BGP session to the route server reaches many members | Broad peering with minimal setup |
| Bilateral public peering | Direct BGP session between two members over the IXP fabric | Important partners needing individual policy |
| Private network interconnect | Dedicated cross-connect between two networks in the same facility | Very high traffic volumes between two parties |
| Remote peering | Member reaches the IXP through a transport provider from another city | Smaller networks without local presence |
Route servers are a key IXP service. They collect routes from participating members and redistribute them according to each member’s filters, but traffic never flows through the route server; it only distributes routing information. Most route servers filter announcements using IRR data and RPKI validation to block hijacked or invalid prefixes.
IXPs in the structure of the internet
The internet is a network of tens of thousands of autonomous systems that connect either by buying transit or by peering. IXPs make peering efficient: one physical port can replace dozens of separate private links.
| Concept | Relationship | Commercial basis |
|---|---|---|
| Internet exchange point | Shared fabric where many networks meet | Port and membership fees paid to the IXP |
| IP transit | Customer buys reachability to the whole internet | Paid per bandwidth to an upstream provider |
| Peering | Two networks exchange only their own and customers’ traffic | Usually settlement-free |
| Internet backbone | High-capacity long-haul networks of large carriers | Owned infrastructure and transit sales |
| Autonomous system | A network with its own routing policy and ASN | Participant in all of the above |
Typical IXP members include ISPs, content providers, CDNs, cloud platforms, hosting companies, universities and enterprises. The presence of large content and CDN networks is what makes an exchange attractive, because a high share of end-user traffic comes from them.
Benefits and governance models
The main benefit of an IXP is keeping local traffic local. Without a nearby exchange, two ISPs in the same city may exchange traffic through a transit provider abroad, adding tens of milliseconds of latency and paying for bandwidth twice.
| Effect | Mechanism | Who gains most |
|---|---|---|
| Lower latency | Shorter, direct paths between networks | Users of real-time applications |
| Reduced transit costs | Peered traffic no longer consumes paid transit | ISPs and content networks |
| Higher resilience | Additional paths independent of upstream providers | All members during outages |
| Local digital ecosystem | Content and caches move closer to users | Regions with weak international links |
Governance varies. Many exchanges, especially in Europe, are neutral non-profit associations owned by their members; others are run by commercial data center operators or by academic and government bodies. Neutrality toward members and data centers is widely considered a core quality criterion, because it keeps the exchange open to competitors on equal terms.
Limitations and operational risks
An IXP is a shared layer 2 domain, so it carries the risks of any large Ethernet network. Operators enforce strict rules for what members may send onto the fabric.
- Broadcast and loop risk: IXPs typically allow only one MAC address per port and filter unwanted protocols.
- Route leaks: a member announcing routes it should not can misdirect traffic, which route-server filtering mitigates.
- Port congestion: an undersized port drops traffic during peaks; members usually upgrade before about 70 to 80 percent utilization.
- Single-site dependence: an exchange in one building is a concentration risk, so large members connect at several sites or IXPs.
- Peering does not replace transit: an IXP provides reachability only to networks that peer there.
For these reasons, networks treat IXP connectivity as one layer in a diverse interconnection strategy, alongside transit from at least two upstream providers and private interconnects for their largest traffic partners.
A non-technical view
Think of an internet exchange point as a busy railway junction shared by many independent railway companies. Instead of sending passengers on long detours through another company’s network, each company lays one track to the junction, and passengers change trains right there.
On the internet, the companies are networks such as ISPs, video platforms and cloud providers. When they meet at an exchange, data travels a shorter route, so websites and calls respond faster and the networks save money on long-distance connections.
A concrete example
A mid-sized regional ISP with around 80,000 subscribers buys all its connectivity from two transit providers. Evening traffic peaks at roughly 60 Gbit/s, most of it video, software updates and cloud services, and the transit bill grows every year.
The ISP connects two 100 Gbit/s ports at the local exchange, one in each of two data centers, and establishes sessions with the route servers plus bilateral peering with several content and CDN networks. Within weeks, more than half of its peak traffic flows over the exchange instead of transit.
Latency to popular services drops by several milliseconds, the transit commitment can be reduced at the next renewal, and when one transit provider suffers an outage, most customer traffic is unaffected because it never depended on that provider in the first place.