On this page
Co-Location Exchange Economics: Renting Microseconds
Co-location is the service that lets a trading firm rent rack space inside or directly adjacent to an exchange's matching engine. Cable lengths are matched across customers, but every co-located firm sits microseconds closer to the engine than anyone trading from a remote site.
Key Takeaways
- Co-location exchanges require all customers to use length-matched cables, so the edge is physical presence in the building, not preferential wiring.
- A full market-making operation across five US venues can cost $11–33 million per year, with engineering headcount typically the largest line item.
- A common mistake is treating co-location as an unfair advantage; SEC rules require non-discriminatory access at published, approved fee schedules.
- Co-location economics explain why electronic market makers dominate liquidity provision and why high-frequency trading has such high fixed-cost barriers to entry.
Key Takeaways
- Co-location exchanges require all customers to use length-matched cables, so the edge is physical presence in the building, not preferential wiring.
- A full market-making operation across five US venues can cost $11–33 million per year, with engineering headcount typically the largest line item.
- A common mistake is treating co-location as an unfair advantage; SEC rules require non-discriminatory access at published, approved fee schedules.
- Co-location economics explain why electronic market makers dominate liquidity provision and why high-frequency trading has such high fixed-cost barriers to entry.
What It Is
A modern equity exchange is a physical building. Nasdaq's primary matching engine runs out of Carteret, New Jersey. NYSE runs out of Mahwah, New Jersey. Cboe BZX/EDGX runs out of Secaucus, New Jersey. CME runs out of Aurora, Illinois. Each operates an on-site co-location facility: a data center where exchange members can lease cabinets, power, cross-connects to the matching engine, and direct market data feeds.
The economic value of co-location is the elimination of public-internet and metro-fiber latency between trader and venue. Round-trip times measured from a co-located cabinet to the matching engine are typically in low single-digit microseconds, compared with hundreds of microseconds to milliseconds for off-site connectivity.
The Intuition
Two ideas drive the market for co-location. First, every exchange operator earns a recurring revenue stream from selling rack space, power, and cross-connects. Second, every trading firm earns higher fill rates and lower adverse selection by being closer to the matching engine. Both sides are willing to pay because the alternative (trading from outside the building) is structurally disadvantaged.
US exchanges are obligated under SEC rules to provide co-location on a fair and non-discriminatory basis. Cable lengths inside the data hall are normalised so no customer gets a shorter wire to the matching engine than another. The "edge" is access to the building at all, not preferential access within it.
How It Works
A typical co-location package includes:
- Cabinet space. Quarter-, half-, or full racks (often 4 kW to 17 kW power per rack), measured in U-spaces.
- Cross-connects. Length-matched fiber pairs from the customer's cabinet to the exchange's order-entry gateway and market-data feed handlers.
- Direct market data. Multicast feeds of the venue's full order book (e.g. Nasdaq TotalView-ITCH, NYSE Integrated Feed) delivered with sub-microsecond fan-out.
- Power and cooling. Engineered for high-density compute, including FPGA accelerator cards and ultra-low-latency switches.
Costs vary by venue and configuration. Public exchange filings and SEC orders approving co-location fees describe annual cabinet costs that can run into the tens of thousands of dollars per cabinet, plus per-port cross-connect fees, plus market data subscription fees that scale with redistribution and use rights. Total all-in spending for a serious latency-sensitive operation across multiple US exchanges runs into the millions per year, before headcount and FPGA development.
A typical co-located trade path:
Multicast market data hits feed handler card ~ 0.2 us
Strategy decision in FPGA ~ 0.3 us
Order generated and placed on wire ~ 0.1 us
Cross-connect to exchange gateway ~ 0.3 us
Gateway-to-matching-engine ~ 0.5 us
Total tick-to-trade ~ 1 to 5 us
That sub-microsecond layer is what colocation customers buy. A non-colocated participant, even one sitting in the same metro area, faces 100 us to 500 us of additional one-way latency from external fiber, network equipment, and the absence of length-matched cabling.
Worked Example
A market-making firm wants to quote two-sided in the top 200 US equities across Nasdaq, NYSE Arca, NYSE, BZX, and EDGX. A representative annual cost outline:
Co-location, 5 cabinets x 5 venues ~ $3 to 5 million
Direct market data feeds (all venues) ~ $1 to 2 million
Cross-connects and switching ~ $0.5 to 1 million
FPGA hardware refresh ~ $0.5 million
Engineering headcount (10 to 30 people) ~ $5 to 20 million
Network connectivity (microwave/fiber) ~ $1 to 5 million
Total annual run-rate ~ $11 to 33 million
Against that cost, the firm needs spread capture, rebate income, and adverse-selection management to clear the bar. Tier-one HFT market makers routinely earn revenues many multiples of these costs, which is why the activity persists. Smaller firms, or strategies whose alpha does not scale with speed, find the economics impossible and trade from non-co-located venues or via brokered low-latency networks.
Common Mistakes
-
Confusing co-location with proximity hosting. Proximity hosting is a third-party data center near the exchange, not inside it. The metro-fiber leg from a proximity facility to the exchange adds several hundred microseconds. For latency-sensitive strategies, that gap is decisive. Marketing copy sometimes blurs the distinction.
-
Assuming all colocation customers are equal. Length-matched cabling is the rule, but customers can still differ by switch generation, network card, FPGA design, and software stack. Two firms in the same data hall can have very different tick-to-trade times even with identical cable lengths.
-
Treating colocation cost as the full barrier to entry. The cabinet bill is typically the smallest line item. Engineering talent, FPGA development, and ongoing testing are far larger ongoing costs. A firm that buys racks but underinvests in engineering will not capture co-located edge.
-
Ignoring fee filings. US exchanges file colocation fee changes with the SEC. Those filings (and the SEC orders approving or suspending them) describe the actual price schedule, fairness analysis, and competitive landscape. A firm modelling colocation economics should read the filings rather than relying on press summaries.
-
Forgetting Reg NMS context. Co-location is fully consistent with Reg NMS as long as access is non-discriminatory and prices are transparent. Periodic political pressure to ban or restrict colocation surfaces every few years; a serious operator monitors the regulatory calendar rather than relying on the status quo.
Frequently Asked Questions
Q: What is co-location in simple terms? Co-location means renting a server cabinet inside the same physical building as a stock exchange's matching engine. The closer your hardware is to the engine, the fewer microseconds your orders take to arrive and execute.
Q: How does co-location affect investment decisions? For most investors it is invisible but relevant: firms with co-location can quote tighter spreads and fill orders faster, which can improve execution quality for everyone. It also creates a structural cost barrier that concentrates professional liquidity provision among well-capitalized firms.
Q: What is a real-world example of co-location exchange economics? A market-making firm quoting 200 stocks across Nasdaq (Carteret, NJ), NYSE Arca (Mahwah, NJ), and Cboe venues (Secaucus, NJ) would rent cabinets at each site. Five cabinets across five venues can cost $3–5 million per year before any data feeds or staffing.
Q: How can investors evaluate the impact of co-location on their trades? Check execution quality reports (Rule 606 disclosures) to see where your broker routes orders and whether fills are at or inside the NBBO. Brokers routing to co-located market makers generally produce tighter spreads than those routing to slower venues.
Q: How is co-location different from proximity hosting? Proximity hosting is a third-party data center near an exchange, not inside it. The additional metro-fiber hop adds several hundred microseconds, which is decisive for latency-sensitive strategies. Marketing copy sometimes uses the terms interchangeably, but they are not equivalent.
Sources
- Nasdaq. "Equity 7 Section 130, Co-Location Services and Pricing." https://listingcenter.nasdaq.com/rulebook/nasdaq/rules
- NYSE. "Mahwah Data Center Co-location Service Description." https://www.nyse.com/markets/connectivity
- SEC. "Order Approving NYSE Co-location Fees (Release No. 34-62960)." https://www.sec.gov/rules/sro/nyse/2010/34-62960.pdf
- SEC. "Concept Release on Equity Market Structure (Release No. 34-61358)." https://www.sec.gov/rules/concept/2010/34-61358.pdf
Disclaimer
This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.