Trading firms place execution infrastructure in the carrier hotels that host, or sit closest to, exchange matching engines, and put everything else somewhere cheaper. In the NYC Tri-State market that means cabinets at $200–240 per kW per month with under 1% vacancy (StackedAI analysis, carrier-hotel market analysis, Apr 2026) and no capacity above 5 MW (CBRE, NA Data Center Trends H1 2025 NY Tri-State, Sep 2025).
What is a carrier hotel, and why do exchanges anchor it?
A carrier hotel is a multi-tenant building whose value comes from who is already inside it. Carriers, exchanges, market data vendors, cloud on-ramps and trading counterparties terminate fiber in a shared meet-me room, and any tenant can reach any other with a short cable called a cross-connect. The operator sells space, power and cross-connects; the tenants bring the network effect.
For trading, the anchor tenant is the exchange. A matching engine pairs buy and sell orders, and the time an order takes to reach it is bounded by physics: light in fiber travels a fixed distance per microsecond, so the firm whose cabinet is nearest (or, under equidistant rules, equally near) is acknowledged first. Once an exchange places its engine in or beside a carrier hotel, market makers follow, data vendors follow them, and the building becomes the ecosystem. The NY4 versus 350 East Cermak comparison covers the two largest US examples.
Interconnection demand shows in operator results. Equinix reported a record 9,700 net interconnection additions in the second quarter of 2026, with bookings up 23% (Equinix, Q2 2026 Results, Jul 2026). Digital Realty, which operates 350 East Cermak, reported $108 million of quarterly bookings in the 0–1 MW and interconnection segment and cash renewal spreads of +25.4% (Digital Realty, Q2 2026 Results, Jul 2026), which is what incumbents in scarce buildings pay to stay.
Which carrier hotels matter for trading firms?
Six addresses account for most low-latency placement decisions in North America and London, and their roles differ.
| Hub | Operator | Ecosystem role | Market condition (cited) |
|---|---|---|---|
| Equinix NY4 / NY5, Secaucus, NJ | Equinix | Interconnection hub of the New Jersey equities and options ecosystem; hosts venues; fiber reach to Nasdaq (Carteret) and NYSE (Mahwah) | No capacity >5 MW; record-high pricing (CBRE, NA Data Center Trends H1 2025 NY Tri-State, Sep 2025); $200–240/kW/mo, <1% vacancy (StackedAI analysis, carrier-hotel market analysis, Apr 2026) |
| 350 East Cermak, Chicago, IL | Digital Realty | Chicago’s carrier hotel and historic home of the futures ecosystem; reach to CME Aurora and East Coast venues | Vacancy 1.9%, no 5 MW+ contiguous (CBRE, NA Data Center Trends H2 2025 Chicago, Feb 2026); Q1 2026 rents $200–230/kW/mo (CBRE, Global Data Center Trends 2026, Jun 2026) |
| CME Group colocation, Aurora, IL | CyrusOne (facility); CME Group (program) | Purpose-built site housing the CME Globex matching engine; equidistant cabling rules | Priced by the exchange program; 2026 price lists not published |
| Equinix LD4, Slough, UK | Equinix | Anchor of the Slough trading campus west of London; hub for FX and European derivatives venues | London vacancy 8.6% in Q1 2026 (CBRE, Global Data Center Trends 2026, Jun 2026) |
| 56 Marietta Street, Atlanta, GA | Multi-operator carrier hotel | The Southeast’s principal interconnection point; secondary and disaster-recovery footprints | $180–185/kW/mo (StackedAI analysis, carrier-hotel market analysis, Apr 2026); Atlanta vacancy 1.0% (CBRE, Global Data Center Trends 2026, Jun 2026) |
| 151 Front Street West, Toronto, ON | Multi-operator carrier hotel | Canada’s primary carrier hotel; Canadian equities and derivatives participants, carriers and cloud on-ramps | $160–200 CAD/kW/mo (StackedAI analysis, carrier-hotel market analysis, Apr 2026) |
Two clarifications keep the table honest. The New Jersey equities ecosystem is a triangle, not a building: Nasdaq’s engine is in Carteret, NYSE’s in Mahwah, and Secaucus hosts other venues plus the fabric that ties all three together, so firms place their primary footprint in Secaucus and buy fiber to the other corners. And CME’s Globex engine is not in 350 East Cermak; it sits in Aurora, where firms wanting the shortest path colocate under CME’s program, while 350 East Cermak is the place to reach everyone else. Latency figures between sites are marketed by network vendors and vary by route; StackedAI benchmarks them per engagement.
How should different types of trading firms think about placement?
StackedAI’s trading-firm work covers three profiles (StackedAI analysis, carrier-hotel market analysis, Apr 2026), and each weights the execution path differently.
Prop trading and market-making. These firms live on the execution path. Order gateways, feed handlers, FPGA or kernel-bypass servers and risk checks belong in the carrier hotel or exchange hall, in adjacent cabinets so cabling stays short. Everything off that path (order management, compliance archives, analytics) belongs outside, because carrier-hotel power is the most expensive the firm buys.
Quant and systematic funds. Execution latency matters less, but a fund that wants full-depth market data at the source still needs a capture footprint in the hub. Research and backtesting are batch workloads, and a GPU cluster should not sit near a matching engine: AI racks run in the 50–70 kW band (StackedAI analysis, Tier-2 conversion density roadmap citing Goldman Sachs and Dell’Oro, Aug 2026) while legacy halls were engineered for 5–15 kW per rack (StackedAI analysis, Cooling Solutions engineering note, Aug 2026). See the GPU research cluster placement page.
Crypto-native firms. Crypto venues run in public cloud regions and commercial data centers rather than exchange-owned halls, so colocation for a crypto market maker often means the same cloud region or carrier hotel as the venue’s cloud on-ramp. The anchor is a cloud region, and the footprint may need to move when the venue does.
What are the market conditions in the major hubs in 2026?
Carrier hotels are a narrow slice of a market short of everything. North American vacancy sat at 1% for the third consecutive year, rents are up roughly 70% since 2020, and vacancy is expected near zero through 2028 (JLL, NA Data Center Report Midyear 2026, Aug 2026), with no meaningful easing before 2030 (Cushman & Wakefield, Americas Data Center Market Shifts to Managed Growth, Feb 2026).
The trading hubs are tighter still, as the table above shows: no capacity above 5 MW in the Tri-State market, under 1% carrier-hotel vacancy, and no 5 MW+ contiguous block in Chicago, where ComEd delays run to 2032 or later and a ten-year letter of credit is required at 50 MW or more (CBRE, NA Data Center Trends H2 2025 Chicago, Feb 2026). Atlanta and Toronto price below the two primary hubs (StackedAI analysis, carrier-hotel market analysis, Apr 2026), which is why both appear as the secondary leg in two-site designs.
These are space-and-power rates. Cross-connects, exchange ports, installation and remote hands are billed separately, and no 2026 price list for NY4, NY5, 350 East Cermak, Aurora or LD4 has been published; the low-latency colocation cost page breaks down the full bill.
What do cross-connect and equidistant fairness actually mean?
Cross-connect. A dedicated physical cable, usually single-mode fiber, run by the operator between two tenants through the meet-me room, billed as a one-time install plus a monthly charge. It is the unit a trading footprint is built from: one to each exchange gateway, market data feed, network provider and counterparty. Counts in the dozens per cabinet are common for market makers and the line can rival the power bill, so the number bundled into the contract is a negotiation point.
Equidistant fairness. Exchanges that sell colocation run cabling of equal length from every participant’s cabinet to the access switch in front of the matching engine. A cabinet physically closer than others has surplus fiber coiled on a spool (a delay spool) so its path matches the farthest cabinet. CME’s Aurora program and the New Jersey exchange sites operate this way, so inside an exchange-run hall a better position cannot be bought, and competition shifts to hardware, feed handling and inter-hub paths. Inside a general carrier hotel such as NY4 or 350 East Cermak, where the operator is not the exchange, no such rule applies, and cabinet location relative to the meet-me room and the firm’s other cabinets does matter.
How do trading firms negotiate carrier-hotel contracts?
In a building with under 1% vacancy (StackedAI analysis, carrier-hotel market analysis, Apr 2026), the operator holds the leverage on rate; the firm’s leverage lies in term, volume and the operator’s interest in keeping tenants that generate cross-connect revenue.
| Negotiation item | What to ask for | Why it matters |
|---|---|---|
| Term | Three to five years with renewal options at capped escalators | Cash renewal spreads of +25.4% (Digital Realty, Q2 2026 Results, Jul 2026) show what uncapped renewals cost |
| Committed power density | Per-cabinet kW commitment with stepped increase rights; committed versus metered billing made explicit | Density headroom is the difference between adding a server and adding a cabinet, and cabinets are scarce |
| Cross-connect bundle | A stated number included per cabinet; a fixed rate card for additions over the term | Cross-connects recur monthly and grow with venues and feeds; a fixed card prevents mid-term repricing |
| Right of first refusal on adjacent cabinets | Written ROFR or option on the next one or two cabinets in the row | Adjacency keeps intra-firm cabling short |
| Install and non-recurring charges | Waiver or amortization of cabinet, cross-connect and cage NRCs against term | The most negotiable line, because NRCs do not affect recurring revenue |
| Remote hands | A monthly block at a fixed rate; defined response times during market hours | Response time during trading hours is an operational control |
| Sublease and assignment rights | Right to sublease or assign to an affiliate or successor | Creates an exit if strategies retire; sublease inventory is how newcomers get in |
| Power and cooling SLA | Explicit uptime, temperature and humidity commitments with meaningful credits | Legacy halls at 5–15 kW per rack (StackedAI analysis, Cooling Solutions engineering note, Aug 2026) can be thermally marginal for dense servers |
Operators do not publish rate cards for these buildings, and quotes vary with relationship, hall utilization and timing, so the most effective single move is a structured request for proposal across every viable hub and operator at once.
What is the two-site pattern, and why does it dominate?
Firms that both trade and research run two sites. Site one is the carrier hotel or exchange hall: a few high-value cabinets holding only what must be there. Site two is a Tier-2 market data center holding research compute, storage, backtesting, order management and everything that tolerates milliseconds instead of microseconds.
Tier-2 colocation prices 15–30% below primary-market averages, and utility power in candidate markets runs from about 7.15 cents per kWh in Dallas-Fort Worth to roughly 9.25 cents in Northern Virginia (StackedAI analysis, Tier-2 brownfield conversion thesis, Aug 2026). Density matters more than rate: AI racks in the 50–70 kW band (StackedAI analysis, Tier-2 conversion density roadmap citing Goldman Sachs and Dell’Oro, Aug 2026) exceed the air-cooling ceiling of roughly 30–50 kW per rack (StackedAI analysis, Cooling Solutions engineering note, Aug 2026), and a carrier hotel is the wrong place for liquid cooling, as the liquid cooling retrofit page explains.
The sites are joined by dedicated fiber or wavelength services terminating on a cross-connect in the carrier-hotel cabinet. Market data is captured at the hub and shipped out; models are trained at the Tier-2 site and their parameters shipped back; the execution path never leaves the hub.
How does a placement advisor add value?
Colocation placement is a small-ticket transaction in a market that rewards non-public information. An advisor brings four things: benchmarking, because with no 2026 rate cards a quote can only be judged against other quotes and a maintained pricing dataset (StackedAI analysis, carrier-hotel market analysis, Apr 2026); availability intelligence, because in a hub with no capacity above 5 MW (CBRE, NA Data Center Trends H1 2025 NY Tri-State, Sep 2025) knowing which halls have cabinets and when a block returns is worth more than any discount; access to sublease and secondary cabinets, which firms retiring strategies leave under contract with cross-connects installed and which are often the only near-term way into a full building; and negotiation, meaning a request for proposal structured so operators compete on comparable terms, then term and density traded for expansion rights, NRC waivers and cross-connect bundles.
Key terms
- Carrier hotel: a multi-tenant building where carriers, exchanges and enterprises interconnect through a shared meet-me room.
- Matching engine: the exchange system that pairs buy and sell orders; its location anchors the low-latency ecosystem around it.
- Cross-connect: a dedicated physical cable run by the operator between two tenants in one building, billed as a one-time install plus a monthly charge.
- Equidistant cabling: an exchange colocation policy giving every participant equal cable length to the matching-engine access switch.
- Tier-2 market: a secondary data center market priced 15–30% below primary averages (StackedAI analysis, Tier-2 brownfield conversion thesis, Aug 2026).
How StackedAI applies this
StackedAI advises proprietary trading, market-making, quantitative and crypto-native firms on where to place execution and research infrastructure, and negotiates the resulting contracts. The firm benchmarks quotes against its own carrier-hotel pricing data, tracks availability and sublease inventory across the Tri-State, Chicago, Atlanta and Toronto hubs, and designs two-site architectures that keep GPU clusters out of the latency path. Andre van Zijl’s operating background includes running Cologix’s Canadian business, anchored on the Toronto carrier-hotel ecosystem. Placement engagements are described on the advisory services page; the Tier-2 site thesis is covered under Tier-2 brownfield conversion.
Frequently asked questions
What is a carrier hotel?
A multi-tenant building where carriers, exchanges, cloud on-ramps and enterprises interconnect physically through a meet-me room. Its value comes from who is already inside. For trading firms the relevant carrier hotels are those that host, or sit closest to, exchange matching engines and their market data feeds.
Where should a trading firm colocate for US equities?
The New Jersey equities ecosystem spans Carteret (Nasdaq), Mahwah (NYSE) and Secaucus, where Equinix NY4 and NY5 host other venues and the densest cross-connect fabric. Most firms place primary execution in Secaucus and reach the other two by fiber. The Tri-State market has no capacity above 5 MW and record pricing (CBRE, Sep 2025).
How much does colocation in a carrier hotel cost?
StackedAI’s April 2026 analysis places NYC Tri-State cabinets at $200–240 per kW per month with under 1% vacancy, Atlanta’s 56 Marietta at $180–185, and Toronto’s 151 Front Street West at $160–200 CAD. Cross-connect, exchange port and install fees sit on top; no 2026 price lists are published, so StackedAI benchmarks them per engagement.
What is an equidistant cross-connect?
Exchanges that offer colocation run cabling of equal length from every participant’s cabinet to the matching-engine access switch, whatever the cabinet’s floor position. Surplus fiber is coiled on spools so nobody is physically closer. Colocation therefore buys equal access, not a private advantage.
Can a quant fund run its GPU research cluster in the carrier hotel?
It usually should not. AI racks run in the 50–70 kW band while legacy halls were built for 5–15 kW per rack (StackedAI analysis, Aug 2026), and carrier-hotel power is the most expensive the firm buys. Research and backtesting tolerate latency that execution cannot, so the cluster belongs in a Tier-2 site linked back by fiber.
Sources
- StackedAI analysis, carrier-hotel market analysis, Apr 2026 (internal; hub pricing and vacancy; target firm profiles)
- StackedAI analysis, Cooling Solutions engineering note, Aug 2026 (internal; legacy hall and air-cooling density limits)
- StackedAI analysis, Tier-2 conversion density roadmap citing Goldman Sachs and Dell’Oro, Aug 2026 (internal; AI rack density band)
- StackedAI analysis, Tier-2 brownfield conversion thesis, Aug 2026 (internal; Tier-2 pricing discount and utility rates)
- CBRE, North America Data Center Trends H1 2025, New York Tri-State market profile, Sep 2025, https://www.cbre.com/insights/local-response/north-america-data-center-trends-h1-2025-market-profiles-new-york-tri-state
- CBRE, North America Data Center Trends H2 2025, Chicago market profile, Feb 2026, https://www.cbre.com/insights/books/north-america-data-center-trends-h2-2025/chicago-data-center-market
- CBRE, Global Data Center Trends 2026, Jun 2026, https://www.cbre.com/insights/reports/global-data-center-trends-2026
- Equinix, Second Quarter 2026 Results, Jul 2026, https://www.prnewswire.com/news-releases/equinix-reports-second-quarter-results-raises-2026-guidance-and-long-term-outlook-302838047.html
- Digital Realty, Second Quarter 2026 Results, Jul 2026, https://investor.digitalrealty.com/news-releases/news-release-details/digital-realty-reports-second-quarter-2026-results
- JLL, North America Data Center Report Midyear 2026, Aug 2026, https://www.jll.com/en-us/insights/market-dynamics/north-america-data-centers
- Cushman & Wakefield, Americas Data Center Market Shifts to Managed Growth, Feb 2026, https://www.cushmanwakefield.com/en/united-states/news/2026/02/americas-data-center-market-shifts-to-managed-growth