Internet Exchange Points and Carrier Hotels: Why They Shape Data Center Markets

The internet is a network of independently run networks, each identified by an autonomous system number (ASN). For traffic to move between two of them, they have to connect somewhere. PeeringDB, the community database most network engineers use to plan these connections, defines an internet exchange point as “an infrastructure for interconnecting three or more Autonomous Systems.”¹ In practice an IXP is usually a switching platform inside a data center, where members plug in once and then exchange traffic with many peers across that shared fabric.⁴

A carrier hotel is the building, not the switch. Data Center Dynamics describes the category as facilities where metro fiber carriers meet long-haul carriers, with network providers numbering in the dozens, often in large multi-story buildings in the center of a major metro, some of them more than a century old.² Inside, carriers and tenants meet in a meet-me room, a neutral space where their cables terminate and are patched together with physical cross-connects.

The two often overlap. Large exchanges tend to run their switches in carrier hotels and other network-dense colocation buildings, because that is where their members already are. A modern colocation campus can also become a carrier hotel in function if enough networks build into it.

Interconnection hubs are mostly accidents of history that became self-reinforcing. Networks go where other networks already are, because every new connection in the same building is cheaper and faster than one across town. Once a building or metro holds enough of them, it becomes hard for any other location to compete.

• New York. 60 Hudson Street, whose lobby has barely changed since the 1930s, has long been a telecom hub and is now a major data center interconnection point.²
• Los Angeles. One Wilshire was built in the 1960s as an office building and renovated in 1992 for telecom use; subsea cables to Asia terminate there, and more than 300 network providers are available in its main meet-me room.⁵
• Chicago. 350 East Cermak Road, a former printing plant of about 1.1 million square feet, is a key Midwestern carrier hotel.⁵
• Northern Virginia. MAE-East, one of the first large public exchanges, had a key hub at a UUNET facility in Ashburn. Carriers judged its scalability limited and sought bids for a separate carrier-neutral building nearby; Equinix’s founders won that work and opened their first Ashburn data center in 1998.⁶

The Ashburn story is the clearest example of how an exchange shapes land use. The neutral interconnection campus attracted networks, the networks attracted cloud providers, and the cloud providers attracted the wholesale campuses that now define Loudoun County. Our guide to the Northern Virginia data center market follows that growth into its current power constraints.

Carrier hotels face their own limits. Data Center Dynamics notes the open question of whether historic legacy buildings can keep up as the industry shifts toward purpose-built, high-density facilities.² The usual result is a hub-and-spoke pattern: the old building keeps the meet-me room and the network density, while large compute moves to newer campuses connected back to it by fiber.

Counting exchanges is harder than it sounds, because databases use different rules for what counts as active. In a June 2025 comparison of three of them published by the Internet Society, IXPDB listed 1,092 exchanges worldwide, PCH’s directory 1,192 and PeeringDB 1,226.⁴ The Internet Society’s own IXP Tracker counts an exchange as active only if it lists at least three members in PeeringDB.³

The Ashburn exchange shows how concentrated membership is. It had 346 member networks in February 2026 and 339 in May, and the combined port capacity of its members rose from about 27.2 to 27.5 terabits per second over the same months.⁹ Membership churns, too: in the 12 months to May 2026, 23 networks left and 17 joined.⁹

Neutral, member-driven exchanges have also grown in the United States. DE-CIX reported that its New York exchange had become the third largest in North America, and that connected customer capacity across its North American exchanges rose 40% in 2025 to 43 terabits.¹⁰

An exchange lowers two costs at once: the money networks pay to carry traffic and the time that traffic takes. The clearest measurement comes from emerging markets, where an exchange replaced international detours. An Internet Society study found the Kenya Internet Exchange Point cut local traffic latency from 200–600 milliseconds to 2–10 milliseconds and saved local providers almost $1.5 million a year in international connectivity costs, with a similar effect in Nigeria.¹¹ The same study found that the exchanges prompted Google to place a content cache in both countries.¹¹

U.S. metros do not see effects that dramatic, but the mechanism is the same. A data center near a dense exchange can reach hundreds of networks with short cross-connects or a single metro circuit, as Ashburn’s membership count shows.⁹ A site far from one pays for longer circuits and takes longer to reach users. That is why network-dense buildings stay in demand and why enterprises and cloud on-ramps cluster around them.

The effect compounds at the market level. A metro with deep interconnection attracts cloud regions, which attract enterprises that want low-latency access to those clouds, which attract more colocation. Our guide to major U.S. data center markets shows how closely the largest markets track the historic network hubs.

Not every data center needs to sit beside a carrier hotel. The answer depends on who the users are and how quickly they need a response.

Even remote campuses need good paths back to hubs. A 2025 Fiber Broadband Association study reports that hyperscale operators want low latency, achieved with the most direct routes possible and limited switching, along with very high uptime.⁸ It projects that tripling U.S. hyperscale capacity by 2029 would require roughly double the fiber route miles, from about 95,000 in 2024 to about 187,000.⁸ So a site three hours from Ashburn or Dallas may work for training, but only if diverse fiber can reach the hub within the buyer’s latency budget. Our guides on fiber route diversity and latency and latency requirements by workload work through those budgets.

PeeringDB is a free, user-maintained database of networks, exchanges, the data centers where they peer and the carriers that link them.⁴ It shows which networks are present in which buildings and which exchanges they peer at. The Internet Society’s IXP Tracker builds on PeeringDB data to show active exchanges, membership and capacity by country and by exchange.³

1. 01Find the nearest facilities listed in PeeringDB and note how many networks each one hosts.
2. 02Identify which exchanges operate in those facilities and how many members each has.
3. 03Check whether major clouds and content networks are present, since their presence often signals a mature hub.
4. 04Estimate the route distance, not the straight line, from the site to the two closest network-dense buildings.
5. 05Ask fiber providers whether dark fiber or lit capacity is available on at least two physically separate routes.

The data has limits. PeeringDB entries are self-reported, membership counts change month to month, and a listing does not tell you about available capacity, pricing or whether a building has space and power for new tenants. Treat it as a map of where networks meet, then confirm with the carriers. Our guide to GIS data sources for site selection covers how to layer it with power and land data.

For most greenfield land, the practical question is not “Is there an exchange next door?” but “Can this site reach a hub fast enough, on diverse routes, for the buyer it is meant for?” A short checklist:

• Define the target workload first. A training campus, an inference region and a retail colocation building have very different needs.
• Measure route distance to the nearest two or three network-dense facilities, and convert it to round-trip latency.
• Confirm at least two physically diverse fiber paths, ideally from different carriers in different rights-of-way.
• Ask whether nearby long-haul routes have access points where a lateral could be built, and who would build it.
• Note planned new exchanges or neutral facilities in the region; hubs do grow, as recent U.S. exchange growth shows.¹⁰

Fiber can usually be built, but it takes time and money, and rights-of-way can be harder to secure than the distance suggests. Our guides to fiber connectivity for data center sites and dark fiber vs. lit services cover the build options. If you want an outside view of how a parcel scores on connectivity alongside power and land, you can get a site reviewed.