Colocation and What It Actually Buys a Firm

Colocation buys a shorter cable to the matching engine and nothing else, a distinction that separates the strategies that need it from those that do not.

Colocation means physically hosting a firm's trading servers inside or immediately next to an exchange's own data center, so that the cable carrying an order to the matching engine is as short as physically possible. What it buys is a fixed, small reduction in propagation delay, nothing about strategy, signal quality or fill probability beyond that single physical fact.

An exchange's data center, places like the Equinix NY4 and NY5 facilities in Secaucus, New Jersey, or the NYSE's own facility in Mahwah, rents out rack space to trading firms alongside the exchange's own matching engines. A colocated firm pays for power, space and a direct cross-connect, and the exchange in turn treats every colocated participant as equally close, since the whole point of the facility is that no firm inside it has a further physical advantage over another firm inside it.

Exchanges typically go further and equalize cable length within the facility itself, running each colocated cabinet's cross-connect through a patch panel built so that every participant's cable to the matching engine measures the same length regardless of which physical rack they occupy. This levels the field among colocated firms specifically, which shifts the competitive question from where a server sits to how fast the software running on it can process a message once it arrives.

Light travels through fiber optic cable at roughly two-thirds the speed it travels through a vacuum, close to 200,000 kilometers per second rather than the roughly 300,000 kilometers per second of light in open space, so every additional kilometer of cable adds a small but real amount of one-way delay. A firm running its servers from an office miles from the exchange, routing over a metro fiber connection, faces a round trip measured in the hundreds of microseconds beyond what a colocated rack sees, purely from the extra cable length.

Inside the data center itself, the cross-connect between a colocated server and the exchange's matching engine can be as short as tens of meters, which is where the bulk of the latency saving actually comes from.