Strasmore Research
Learn Matt ConnorBy Matt Connor

What Exchange Colocation Costs Per Month

What exchange colocation costs per month: cabinet, power, cross connects, ports and entitlements, with every figure pinned to a dated public fee schedule.

Exchange colocation costs start near $1,320 a month for the smallest power circuit on a published US equity fee schedule and run past $19,000 a month for a single bundled cabinet at the NYSE, before one byte of market data is switched on. Colocation means renting space for your own servers inside the building that houses an exchange's matching engine, the computer that pairs buy and sell orders. The rack is rarely the largest line on the invoice: ports, cross connects and data entitlements usually are.

Every dollar below is pinned to a named filing with a date. US exchanges must file fee changes with the SEC, so these schedules are public documents and any figure here can be re-checked against the version in force when you read it.

What exchange colocation actually buys you

An exchange matches orders in one primary data centre. The main US equity venues sit within a few miles of each other in New Jersey:

  • Nasdaq matches at Carteret, in the halls it calls NY11 and the NY11-4 expansion.
  • NYSE matches at Mahwah.
  • Cboe matches at Secaucus.

A colocation customer rents a cabinet, a locked rack roughly the size of a large refrigerator, on the same floor as that engine, powers it, and runs fibre from it into the exchange's network. What you are buying is the removal of the wide-area leg. An order sent from a Manhattan office crosses tens of miles of fibre in each direction. An order sent from a cabinet in the same building crosses a few hundred feet.

The colocation invoice, line by line

Cabinet and power

Power is the meter that matters. Nasdaq replaced its cabinet-density pricing with a flat rate of $550.00 per kilovolt-ampere for each power circuit at Carteret in filing SR-NASDAQ-2026-005, submitted 15 January 2026 and published in the Federal Register on 2 February 2026. Multiplying each circuit by that rate produces the published examples: $1,320.00 a month for a 2x20 amp 120V circuit, $1,980.00 for 2x30 amp 120V, $5,944.22 for a three-phase 2x30 amp 208V circuit and $11,888.45 for a three-phase 2x60 amp 208V. Under the schedule it replaced, a low-density cabinet of 2.88 kW or less billed $3,850 to install and $2,200 a month.

NYSE sells the same thing pre-bundled. Its 2 kW Partial Cabinet Solution carries a $10,000 initial charge and a monthly charge the exchange raised from $15,000 to $16,500 in a filing published 16 June 2026, operative no later than 31 October 2026. The same round of filings added a 4 kW option at $12,000 initial and $19,000 a month, and withdrew the 1 kW option. The 2 kW bundle is not bare space: it carries one LCN connection, one IP network connection, two NMS network connections, two fibre cross connections and a time feed.

Cross connects

A cross connect is a single physical fibre pair patched between two points inside the building by the data centre operator. NYSE's co-location schedule has billed an individual fibre cross connect at $500 to install plus $500 a month, with a bundle of six at $500 to install plus $1,500 a month. Those amounts appear in NYSE co-location fee filings from 2012 onward and in the NYSE Arca schedule filed in 2015. Counts add up fast. A firm with two servers, two exchange handoffs and a backup path holds a dozen of them.

Ports and gateways

Ports are where the bill compounds. Cboe raised its monthly fee per 10 Gb physical port from $7,000 to $8,000 effective 2 January 2026. Each physical port then carries logical ports, the individual sessions your software logs into, billed on their own: a Cboe bulk order-entry logical port costs $3,000 a month for each increment of up to 30 million orders per trading day. Options quoting is priced per port and tiered. On Nasdaq's options market a FIX port is $650 per port per month, while quoting ports run $1,620 each for the first five, $1,080 for ports six through twenty, and $540 for each port above twenty.

Market data entitlements

Ports move orders. Data is a separate purchase. Nasdaq's depth-of-book entitlement billed a $2,224 monthly direct access fee, a $1,125 monthly internal distributor fee and a $2,820 monthly external distributor fee, all effective 1 January 2026. Per-user and per-device fees sit on top, and every venue prices its own book the same way. The full structure is in what real time market data costs, the archives of the same feeds are priced in what historical tick data costs, and whether you are billed as a professional at all is settled in professional versus non professional market data.

Add the published line items for one venue and one rack and the monthly number lands in the mid five figures, before hardware, staff and a link to anywhere else. A firm present on several venues with redundant paths is into six figures a month.

Fiber equalization: colocation buys the floor, not an edge

Light covers about a foot per nanosecond in a vacuum and roughly two thirds of that in glass, so 100 extra feet of fibre is about 150 extra nanoseconds. Exchanges take that variable off the table inside the room. The fibre from every client cabinet to the exchange access point is cut to the same length, whichever cabinet it is. Nasdaq has described an equalization project covering the NY11 and NY11-4 halls at Carteret, and NYSE has equalized connectivity at Mahwah since its earliest co-location filings in 2010. The cabinet beside the engine and the cabinet at the far wall are the same number of nanoseconds away.

The practical consequence for a new tenant: colocation is a floor, not a ranking. Paying the bill puts you level with the other tenants. The differences that remain sit in your own hardware and software, and in the paths between data centres. Public comment letters filed with the SEC on Nasdaq's 2024 co-location filings dispute how complete the equalization is in practice, which is worth reading before treating the word as absolute. Once inside, the contest is position in the order queue at each price level, and how to estimate queue position covers how that is measured.

What message rates colocated systems are built for

The fee stack is sized for a message rate most people never see. The panel below counts SPY trade messages across one fixed session, 10 June 2026, in half-hour buckets on the New York clock, running from the early premarket through the close of after-hours trading.

QuerySPY trade messages per second by half hour, 10 June 2026
32 rows (showing 20)
et_timemessage_counttrades_per_second
04:0031861.8
04:308270.5
05:0011030.6
05:308520.5
06:006700.4
06:306860.4
07:00113656.3
07:3031121.7
08:0029751.7
08:30118536.6
09:0035842
09:309147350.8
10:006807037.8
10:307850243.6
11:007824843.5
11:308686648.3
12:004541725.2
12:303405318.9
13:004433124.6
13:305168428.7
The exact SQL behind every number
SELECT
    formatDateTime(
        toStartOfInterval(toTimeZone(window_start, 'America/New_York'), INTERVAL 30 minute),
        '%H:%i')                          AS et_time,
    sum(transactions)                     AS message_count,
    round(sum(transactions) / 1800, 1)    AS trades_per_second
FROM global_markets.delayed_stocks_minute_aggs
WHERE ticker = 'SPY'
  AND window_start >= '2026-06-10 08:00:00'
  AND window_start <  '2026-06-11 00:00:00'
GROUP BY et_time
ORDER BY et_time
Run this yourself

The earliest half hour in view, 04:00 ET, averaged 1.8 trades a second across the bucket, and the final one, 19:30 ET, averaged 0.8. Between them stand the opening and closing buckets, the two tallest points on the chart. The date is fixed in the SQL, so the panel draws the same session every time.

Averages hide the bursts. The next panel takes a five-minute slice of that session, 10:00 to 10:05 ET, and groups every SPY quote update by the millisecond it arrived in.

QuerySPY quote updates per millisecond, 10:00 to 10:05 ET on 10 June 2026
updates_in_one_millisecondmillisecond_countpct_of_active_ms
1 update2847941.3
2 updates1475921.4
3 updates888012.9
4 updates56058.1
5 or more updates1129316.4
The exact SQL behind every number
WITH active_ms AS
(
    SELECT
        toUnixTimestamp64Milli(sip_timestamp) AS ms,
        count()                               AS updates
    FROM global_markets.cache_stocks_quotes
    WHERE ticker = 'SPY'
      AND sip_timestamp >= '2026-06-10 14:00:00'
      AND sip_timestamp <  '2026-06-10 14:05:00'
    GROUP BY ms
)
SELECT
    multiIf(updates = 1, '1 update',
            updates = 2, '2 updates',
            updates = 3, '3 updates',
            updates = 4, '4 updates',
                         '5 or more updates')              AS updates_in_one_millisecond,
    count()                                                AS millisecond_count,
    round(100 * count() / (SELECT count() FROM active_ms), 1) AS pct_of_active_ms
FROM active_ms
GROUP BY updates_in_one_millisecond
ORDER BY updates_in_one_millisecond
Run this yourself

Of the milliseconds that carried any quote update at all, 41.3% carried 1 update, while 11293 of them carried 5 or more updates. A system that falls a few milliseconds behind during clusters like those is working from a book the exchange has already moved past.

Message rate is also wildly uneven across names, which is most of the answer to who colocation is economic for. The panel below counts quote updates for five household names over one hour of the same pinned session.

QueryQuote updates in one hour, five names, 10:00 to 11:00 ET on 10 June 2026
symbolquote_update_countms_between_updates
SPY15675392.297
NVDA8527214.222
AAPL27989812.862
KO15369423.423
ADP34311049.257
The exact SQL behind every number
SELECT
    ticker                        AS symbol,
    count()                       AS quote_update_count,
    round(3600000 / count(), 3)   AS ms_between_updates
FROM global_markets.cache_stocks_quotes
WHERE ticker IN ('SPY', 'NVDA', 'AAPL', 'KO', 'ADP')
  AND sip_timestamp >= '2026-06-10 14:00:00'
  AND sip_timestamp <  '2026-06-10 15:00:00'
GROUP BY symbol
ORDER BY quote_update_count DESC
Run this yourself

SPY posted 1567539 quote updates in that single hour, one every 2.297 milliseconds on average. The quietest of the 5 names, ADP, updated once every 1049.257 milliseconds. The colocation invoice is identical for both.

Who colocation is economic for

  • Market makers and latency-sensitive proprietary firms quoting continuously, where queue position at each price level is the product itself.
  • Options market makers quoting thousands of series at once. The Nasdaq quoting-port tiers above, which only begin discounting past twenty ports, are priced for exactly those firms.
  • Broker-dealers running order routers that must reach every venue quickly on behalf of clients.
  • Vendors and extranet providers who buy space, ports and feeds once, then resell slices of them.

These are fixed monthly charges. They do not scale down with your volume, which is what makes them punishing for a small operation: a $16,500 cabinet bundle bills the same whether it carries 30 million orders a day or thirty. For anyone holding positions for hours or days, the microseconds colocation removes do not show up in the results at all. The holding period swamps them by many orders of magnitude.

Realistic alternatives for everyone else

  • Proximity hosting in a nearby carrier hotel. The Secaucus campus sits a short fibre run from several matching engines, priced by a landlord rather than by an exchange, with the last leg bought as a cross connect.
  • Managed access through an extranet provider, who holds the rack, the ports and the entitlements and bills you one invoice instead of five fee schedules.
  • Sponsored or direct market access through a broker that is already colocated. The broker's infrastructure does the work and the cost arrives inside the commission.
  • Consolidated data for research, with a direct feed only where it changes a decision. The SIP versus direct exchange feeds sets out exactly what the cheaper option leaves out.
  • Research in the cloud, execution elsewhere. A backtest that assumes fills only a colocated system could win is fiction, and latency models in HFT backtests covers how to keep that assumption honest.

FAQ

How much does exchange colocation cost per month?

Published US equity schedules put a minimal footprint in the low thousands per month and a bundled cabinet well into five figures. Nasdaq's Carteret rate is $550.00 per kVA per power circuit, pricing a 2x20 amp 120V circuit at $1,320.00 a month (filed January 2026). NYSE's 2 kW Partial Cabinet Solution is $16,500 a month plus a $10,000 initial charge (filed June 2026). Ports, cross connects and market data entitlements are billed on top.

Does colocation make me faster than other firms in the same building?

No. Exchanges equalize the fibre length from every client cabinet to their access point, so every tenant sits the same distance from the matching engine whichever cabinet they rent. Colocation removes the wide-area leg between an office and the exchange. It does not rank you against your neighbours in the room.

What is a cross connect, and why is it billed every month?

A cross connect is one physical fibre pair patched between two points inside the data centre, for example from your cabinet to an exchange handoff. It is sold as a recurring service rather than a one-time installation. NYSE's schedule lists $500 to install plus $500 a month for an individual fibre cross connect, with six for $1,500 a month.

Are exchange colocation fees public?

Yes. US exchanges file fee changes with the SEC, and they appear in the Federal Register and on exchange rule-filing pages, usually with the old and new amounts side by side. Figures age quickly: several quoted on this page changed during 2026, which is why each one here carries a filing name and a date.

Can an individual trader buy exchange colocation?

Space is sold to exchange members and their service providers, and the fixed monthly cost reaches five figures before any hardware. Retail platforms reach the same venues through brokers that are already colocated, and an individual's fill quality tracks that broker's routing rather than their own distance from New Jersey.


Every panel here ships with the SQL that produced it. Open one, swap the ticker or the date, and run it on the Strasmore terminal to see what another name's message rate looks like.

#colocation#exchange fees#latency#market data#infrastructure