Co-location
Co-location is a service offered by stock exchanges and trading venues that allows market participants to physically install their trading servers within the exchange's own data center, minimizing the latency of order transmission to microseconds and enabling high-frequency trading strategies that depend on speed advantages.
Key takeaways
- Co-location reduces round-trip latency for order submission and market data receipt from milliseconds (remote connection) to single-digit microseconds (physically adjacent servers).
- Exchanges offer co-location as a commercial service, charging monthly fees ranging from tens of thousands to hundreds of thousands of dollars per server rack.
- Co-location is a prerequisite for competitive high-frequency trading: market-making algorithms, statistical arbitrage, and latency arbitrage all require co-located infrastructure.
- Regulatory concerns focus on fairness: co-location is sold on equal terms to all who can pay, but critics argue it creates a two-tiered market where speed buyers have structural advantages over retail and long-term investors.
- The arms race in latency reduction has extended beyond co-location to microwave and millimeter-wave transmission networks that route market data across long distances (Chicago to New York) faster than fiber optic cable.
Explanation
Co-location emerged as a commercial service from exchanges in the late 2000s, formalized as part of the transition to electronic limit order books. When exchanges moved from floor-based to electronic trading, the physical location of a market participant's computers relative to the exchange's matching engine became the primary determinant of execution speed. Exchanges monetized this geography by building data center facilities adjacent to (or housing) their matching engines and renting server rack space to market participants.
The physics of latency are straightforward: signals travel at approximately two-thirds the speed of light through fiber optic cable (200 km per millisecond). A participant with servers in New Jersey connecting to NYSE Arca's matching engine in Mahwah, NJ (40 miles) faces roughly 0.2 millisecond (200 microsecond) one-way latency — too slow for competitive market-making in the highest-frequency strategies. A co-located server 40 meters from the matching engine faces less than 1 microsecond latency — a 200x speed advantage. This difference determines whether a market-maker can update quotes before being 'picked off' by faster traders reacting to correlated market moves.
The regulatory treatment of co-location has been carefully structured to avoid exchange market manipulation. The SEC requires that exchanges offer co-location services on fair and non-discriminatory terms: the same latency for the same price, without preferential access for affiliated parties. Exchanges publish their co-location price schedules and connectivity specifications, and auditors periodically verify that equal cable lengths are used for all co-located participants (a practice called 'equidistant connectivity' or 'cross-connect equalization'). NYSE and NASDAQ have both been investigated by the SEC for unequal co-location practices and paid fines in the 2014-2017 period.
The technology arms race has extended beyond co-location to transmission media. Once co-located servers achieve similar latency at the exchange, the bottleneck shifts to market data distribution across exchanges. Getco (now KCG/Virtu) and Jump Trading pioneered microwave networks between CME Group's data center in Aurora, IL and NASDAQ's data center in Carteret, NJ — a 1,200-kilometer path. Microwave transmission travels through air at closer to the speed of light than through fiber, reducing Chicago-New York latency from 8 milliseconds (fiber) to 4.5 milliseconds (microwave). Subsequent improvements using millimeter-wave technology and experimental laser networks have pushed latency toward the theoretical minimum.
From a market quality perspective, co-location and HFT's presence in CLOBs has a dual character. Bid-ask spreads have narrowed dramatically since electronic market-making replaced specialist/dealer models, benefiting all investors through lower transaction costs. However, market depth (the size of displayed liquidity) has become more fragile: co-located market-makers update quotes at microsecond speed during volatility events, potentially withdrawing liquidity precisely when it is most needed. The Flash Crash of 2010 and various mini-crashes in single stocks have been linked to this dynamic.
Formula
Latency (one-way) ≈ Distance / (Speed of Light × 0.67 for fiber); Signal Speed ≈ 200,000 km/s (fiber) vs. 300,000 km/s (microwave)
Example
A high-frequency trading firm manages quantitative market-making strategies across 50 equity markets. They rent 40U of rack space in NYSE's Mahwah data center for $35,000/month and equivalent space in NASDAQ's Carteret facility for $40,000/month. Total co-location costs of $900,000 annually are dwarfed by the revenue generated: by providing continuous two-sided markets and capturing the bid-ask spread on 200 million shares per day at $0.0015 average spread capture, the firm earns approximately $10,000-15,000 per trading day — $2.5-3.75 million annually — from NYSE alone. Without co-location, competitors with microsecond speed advantages would consistently 'pick off' the firm's stale quotes whenever a correlated market moves, turning the market-making business from profitable to loss-making.
Related terms
Best Execution Bid Ask Spread Central Limit Order Book Electronic Trading Equalization Equity Exchange Fill Or Kill Order Floor High Frequency Trading Latency Limit Order