Quant Memo
Core

Colocation And Proximity Hosting

Why high-frequency trading firms pay exchanges to place their servers physically inside the exchange's own data center, and what that buys them.

Prerequisites: Latency vs Throughput, Latency & High-Frequency Trading

Signals travel fast, but not infinitely fast — even at close to the speed of light, distance costs time. A trading firm whose servers sit fifty miles from an exchange's matching engine is physically incapable of reacting to a price change as quickly as a firm whose servers sit in the room next door, no matter how good its software is. Colocation is renting rack space inside, or immediately adjacent to, the exchange's own data center, so the distance a signal has to travel — and therefore the time it takes — shrinks to a few feet of cable instead of miles of network.

Why distance is a hard floor on speed

No amount of clever code eliminates the delay imposed by physical distance: information cannot travel faster than the speed of light, and in real fiber-optic cable it travels meaningfully slower than that. A round trip to an exchange fifty miles away costs on the order of several hundred microseconds just from the cable, before any processing happens at either end — often more time than the entire rest of a well-optimized trading system's processing takes combined. Colocation doesn't make software faster; it removes the one delay that software can never fix.

Worked example: the distance-to-latency tradeoff

A firm connects to an exchange from a data center fifty miles away over standard fiber. Signal in fiber travels at roughly two-thirds the speed of light in a vacuum, so the round trip alone — fifty miles there, fifty miles back — costs somewhere in the range of 500-800 microseconds, before the firm's own systems do any work. The same firm colocates a server in a rack a few hundred feet from the exchange's matching engine instead. The round-trip distance is now negligible, and the physical propagation delay drops to a small fraction of a microsecond — the entire several-hundred-microsecond distance cost simply disappears. For a strategy competing to be first to react to a price change, that difference alone can be the entire margin between capturing an opportunity and having it already gone by the time an order arrives.

remote (50mi) ~600us round trip colocated <1us round trip
Physical distance to the matching engine is a hard floor on latency; colocation removes nearly all of it by shrinking the cable run to a few feet.

What this means in practice

Because colocation is such a large lever on speed, exchanges sell rack space and even offer equal-length cables to every colocated customer, specifically to avoid one firm gaining an unfair edge purely from being a few feet closer to the matching engine than a competitor in the same room. This has turned physical proximity into a paid, regulated resource: firms compete on cabinet placement, cable length, and network switch hops, and the arrangement has drawn academic and regulatory attention as a symbol of the broader "latency arms race," where firms spend heavily on speed advantages measured in millionths of a second that offer no benefit to ordinary long-term investors.

Colocation eliminates the one latency cost that no amount of software optimization can touch — the physical time signals take to travel — by placing a firm's servers as close as possible to the exchange's own matching engine, turning proximity itself into a competitive and heavily regulated resource.

Related concepts

Practice in interviews

Further reading

  • Budish, Cramton & Shim, The High-Frequency Trading Arms Race
ShareTwitterLinkedIn