Co-Location

Co-location means placing a trading firm's servers in the same physical building — or the same cage within the same data center floor — as the exchange's matching engine. The physics are simple: signals travel at roughly two-thirds the speed of light through fiber optic cables. A server one mile from the exchange experiences about 8 microseconds of one-way latency from distance alone. A server in a co-location cage a few feet from the matching engine can achieve round-trip latencies of 10-50 microseconds, compared to 1-10 milliseconds for a non-co-located connection.

Exchanges operate dedicated co-location facilities: NYSE uses its Mahwah, New Jersey data center, NASDAQ uses Carteret, New Jersey, and CME Group uses its Aurora, Illinois facility for futures. Each facility charges premium fees for rack space, typically $5,000-$20,000 per month per rack, plus separate fees for direct market data feeds.

Exchanges are required to provide equidistant cable runs to all co-location customers — the same physical cable length from the matching engine to every co-located rack — ensuring that one firm cannot gain a distance advantage over another inside the same facility. This rule, enforced by the SEC, prevents the exchange itself from creating a tiered speed advantage.

Co-location is economically justified only for strategies where microsecond execution edges translate directly into profit. The primary users are high-frequency trading firms running market-making strategies, statistical arbitrage strategies, and latency arbitrage strategies.

Market makers use co-location to quote bid and ask prices on thousands of securities simultaneously and update quotes in response to market moves faster than other participants can trade against stale prices. Without colo, their quotes would be picked off by faster traders every time the market moved.

Statistical arbitrageurs exploit pricing relationships between correlated securities (index futures vs. ETFs, ADRs vs. foreign shares) that only persist for microseconds before equalizing. Getting to these mispricings first requires co-location.

Latency arbitrageurs race to exploit the brief window where the same security shows different prices on different exchanges after a price move. Co-location at multiple exchanges simultaneously is essential.

For any strategy with a holding period longer than a few seconds, co-location provides diminishing marginal value. A momentum trader holding positions for minutes or hours gains nothing meaningful from microsecond execution improvements.

Co-location has been the subject of significant regulatory debate. Critics argue that selling premium speed access creates a two-tiered market where sophisticated firms systematically extract value from slower participants. Proponents counter that co-location improves market quality by enabling tighter spreads and deeper liquidity.

The SEC's Regulation NMS (National Market System) governs co-location practices, requiring exchanges to offer colo services on fair, reasonable, and non-discriminatory terms. This means any firm can rent colo space — it is not exclusive to the largest players — though the cost and operational complexity still creates de facto barriers for smaller firms.

IEX (Investors Exchange) took a different approach by introducing a "speed bump" — a 350-microsecond intentional delay on all incoming orders created by routing through a coil of fiber optic cable. This neutralizes co-location advantages and eliminates latency arbitrage opportunities on IEX. Tradewink's execution engine considers IEX routing for positions where minimizing adverse selection from faster participants is a priority.

Access to exchange co-location services is tiered by cost. A standard co-location arrangement includes: rack space rental (physical server housing), power allocation (servers draw significant electrical power), direct market data feeds (faster than consolidated SIP data), and co-location network connections.

Costs vary by exchange and space requirements. A basic setup at a major exchange data center runs $5,000-$15,000 per month in rack fees alone, plus $1,000-$10,000 per month for direct data feeds per exchange, plus significant capital expenditure for servers, networking hardware, and software development. A full HFT infrastructure deployment can represent $50 million or more in initial investment plus millions annually in operating costs.

For retail and most institutional traders, these economics make direct co-location inaccessible. The practical implication is that retail order flow will always execute after co-located firms have already traded on a given price signal. Understanding this structural reality is why Tradewink focuses on second-to-minute timeframe edges rather than microsecond execution — competing in a space where co-location economics do not apply.