Quick definition Co-location is renting space at an exchange's data centre to host your trading servers. Co-located servers are physically close to the exchange's matching engine, minimizing latency. What it is Exchanges rent space in their data centres to traders who want to minimise latency. A co-located server can receive market data and submit orders in under 100 microseconds. A remote server might take 1-5 milliseconds. Co-location is expensive. Traders pay monthly fees (typically thousands of dollars) for the space, power, and connectivity. Only large trading firms and hedge funds can afford it. Why it matters In high-frequency trading, microseconds matter. A trader with a 500 microsecond latency advantage over competitors can profit from price moves before others. Co-location is the primary way to achieve this latency advantage. Co-location has also raised concerns about fairness. Traders who cannot afford co-location are disadvantaged by slower latency. Co-location tiers Exchanges offer different co-location tiers at different prices. A basic tier might offer 500 microseconds of latency. A premium tier might offer 100 microseconds. The price difference can be substantial. Practical example An HFT firm rents a co-located server at NYSE. They feed the server real-time market data from NYSE and route orders through the co-located connection. Another HFT firm operates from an office in Connecticut, 50 miles away. Their latency is 5 milliseconds. They cannot profitably trade against the co-located firm because they receive market information 5 milliseconds later. Kernel bypass Some co-located servers use kernel bypass technology, which bypasses the operating system's network stack to achieve even lower latency (under 10 microseconds). This is the frontier of latency optimisation. Regulation Exchanges offer co-location to all traders equally. There is no restriction on who can co-locate. However, the cost limits access to large firms. See also - Latency - High-Frequency Trading - Kernel Bypass - Exchange Data Centre
