If your broker’s trade server lives in Secaucus, New Jersey, a VPS in Chicago is the wrong box. I’ll say that even though we sell Chicago VPSes. Your latency is set by the distance between the VPS and the server that takes your orders, so pick the metro first and the host second.
Here’s where the engines are, what the round trips look like, and how to test a host before you pay for a month you’ll regret.
Trading happens in a handful of buildings
People talk about “the market” like it’s everywhere. It isn’t. Your order ends up at a matching engine or at your broker’s trade server, and those sit in a small number of data centers.
CME Globex runs in CyrusOne’s Aurora I building on Diehl Road in Aurora, Illinois, about 30 miles west of downtown Chicago. CME sold the building to CyrusOne in 2016 and signed a 15-year lease to keep running Globex there. That’s where ES, NQ, CL and the currency futures like 6E actually match.
Cboe FX runs its New York matching engines in Equinix NY5 in Secaucus and its London engines in Equinix LD4 in Slough, according to Cboe’s own FX connectivity manual. Participants can cross-connect from NY2, NY4 and NY6 on the same Secaucus campus. Cboe’s US equities and options primary data center is in Secaucus too.
In Tokyo, the building to know is Equinix TY3. LMAX put a matching engine there in 2014, and Cboe FX added a point of presence in 2023.
Retail brokers follow the liquidity. IC Markets, for example, says its MT4 and MT5 servers are in NY4 and its cTrader servers are in LD5. That pattern is typical.
So for spot FX during US hours, the center of gravity is northern New Jersey. For CME futures, it’s Aurora. Know which one your orders go to before you shop.
What the round trips actually look like
Light in fiber covers roughly 200 km per millisecond, which means every 100 km of cable route adds about a millisecond to the round trip, and real routes are never straight lines. The published numbers line up with that.
| Route | Round trip | Path |
|---|---|---|
| Aurora ↔ downtown Chicago | ~0.6 ms | Best-case fiber (287 µs each way) |
| GigeNET Chicago ↔ Aurora I | ~2.5 ms | Public internet, via Zayo |
| Chicago ↔ New Jersey | 8.2–8.5 ms | HFT microwave |
| Chicago ↔ New Jersey | 13 ms | Spread Networks dark fiber |
| Chicago ↔ New York | ~18 ms | Public internet |
| Washington area ↔ New York | ~7.5 ms | Public internet |
| New York ↔ London | ~70 ms | Public internet |
| Chicago ↔ London | 85–92 ms | Public internet |
| NY5 ↔ TY3 | 137 ms | Cboe’s own link |
Where the numbers come from: Databento for the Aurora fiber row, our own looking glass for the GigeNET row, Wikipedia’s Spread Networks entry for the two New Jersey rows, WonderNetwork’s ping tables for the public-internet rows, and The Full FX for the Cboe link.
The microwave rows are there for scale. You won’t get those paths from a VPS.
What you will get is the public-internet column. A VPS inside the Secaucus campus sees its broker in about a millisecond. The same broker from Chicago is around 18 ms away, and from Chicago to a London broker you’re near 90 ms before your EA has done anything. The same distance math decides where a web server belongs; Ashburn vs Chicago applies it to websites and apps.
Why a VPS beats your home fiber
A gigabit line at home in Phoenix still has to cross the country to reach Secaucus. Bandwidth doesn’t shorten distance.
Home setups also fail in boring ways: Wi-Fi retries, an ISP that congests every evening, a Windows update that reboots the PC at 2 a.m. while your EA holds a position. A VPS in the broker’s metro fixes the distance and stays on through all of that, and your home connection is left carrying only the remote desktop session, where 60 ms of lag is an annoyance rather than slippage.
Ping is the headline. Jitter and loss do the damage.
A steady 18 ms is fine for an EA that trades on 15-minute bar closes. A path that averages 5 ms but spikes to 80 ms at the New York open is worse, because the spikes land exactly when price moves.
Packet loss hurts more than either. An order is a small message, so there’s often nothing queued behind it to trigger a fast retransmit. The sender waits for a timeout instead, and on Linux the minimum retransmission timeout is 200 ms. So one dropped packet at the wrong second costs you 200 times what shaving a millisecond off the ping saves.
When you compare hosts, compare the 99th percentile and the loss count over a full day. The average hides everything that matters.
What matters on the VPS itself
The CPU you actually get. On an oversold host, other customers’ load shows up as late ticks and slow order sends, and it gets worse during releases, when everyone’s busy at once. GigeNET’s VPS page says each plan “delivers dedicated CPU, guaranteed RAM”. Test it under load anyway. Don’t take anyone’s word for it, ours included.
Enough memory for Windows. MT4 and MT5 are Windows programs. Microsoft lists 2 GB of RAM and 32 GB of disk as the minimum for Windows Server, with 4 GB recommended, which rules out the smallest plans for a Windows trading box. Our VPS plans run from 2 vCPU with 1 GB RAM up to 10 vCPU with 10 GB. For two terminals and an EA, I’d start at the 4 vCPU, 4 GB, 50 GB tier.
Uptime during news. Non-farm payrolls, CPI and FOMC days are when your EA has to work and when shared infrastructure gets tested. Ask any host what its SLA covers (network, host hardware or both) and when maintenance windows fall.
How to test a provider before you commit
You can do this in a day.
- Find your broker’s server. Ask support for the data center name, not the city. “New York” can mean NY4 in Secaucus or a building in Manhattan. In MT4 and MT5, the connection indicator in the bottom-right corner shows the current ping to the trade server.
- Measure from each candidate metro. GigeNET runs a public looking glass where you can ping, traceroute or run MTR from our Chicago, Los Angeles and Washington, D.C. locations to any server you name. Other hosts have their own. Use them.
- Run it for 24 hours. From a trial VPS on Linux:
# one probe per second for 24 hours,
# per-hop summary at the end
mtr --report --report-wide \
--report-cycles 86400 --interval 1 \
203.0.113.10 > mtr-24h.txt
# timestamped ping log, so you can
# pull out specific minutes later
ping -D -i 1 203.0.113.10 \
| tee ping-24h.log
On Windows, WinMTR does the same job. If the broker’s server drops ICMP, measure TCP connect time to the port your terminal uses with Sysinternals PsPing.
- Read the ugly minutes. Pull the samples from 8:25 to 8:40 a.m. Eastern on a release day and from the London and New York overlap. If the 99th percentile in those windows is several times the median, or you see any loss, keep looking.
Where GigeNET fits
We run VPSes in three places: Chicago, Ashburn in Virginia, and Los Angeles. Here’s the plain version of who each one suits.
CME futures traders. Chicago is the right call. Our Chicago facility is in Arlington Heights, in the same metro as the Aurora building. From our Chicago looking glass, a public market-data gateway inside Aurora I (Databento’s dc3.databento.com) came back in about 2.5 ms round trip on average, with no loss, over Zayo, measured on a Monday afternoon during US market hours in late September 2026. That’s a public-internet path, not a cross-connect, so it won’t win a race against firms racked in Aurora. For an EA or a bot that cares about milliseconds, it’s plenty, and you can rerun the same ping yourself during the open. If you run several terminals or a data-heavy bot, a Chicago dedicated server gives you the whole machine.
FX traders with a New Jersey broker. Ashburn is our closest location, and Ashburn is not Secaucus. Public data puts the Washington area about 7.5 ms from New York. For swing EAs or discretionary trading you won’t see that in your fills; for news scalping, where a few milliseconds decide the fill, buy a VPS inside the Secaucus campus from a provider that has one.
FX traders with a London broker. None of our locations is right. Chicago to London is 85–92 ms. Get a London VPS.
Current plans and pricing are on the VPS page. DDoS protection is included with every plan. Before you trust any host with live positions, ours included, get two things in writing: how large an attack the included protection absorbs, and what happens to your server when one goes past it.
FAQ
Only if your broker’s trade server is in the Chicago area, which is uncommon for spot FX. Most retail FX brokers host MT4 and MT5 in the Secaucus Equinix campus or in the Slough campus near London. Chicago is the right metro for CME futures, including currency futures like 6E, because Globex matches in Aurora.
For EAs that act on bar closes, anything steady under about 20 ms is plenty. For scalping around news, you want low single digits, and consistency matters more than the average. Judge a host on the 99th percentile over a full day, not the number the terminal shows at a quiet moment.
No. Our VPS locations are Chicago, Ashburn and Los Angeles. For a broker in Secaucus, Ashburn is our closest option at roughly 7 to 8 ms from New York on public routes, which suits most EAs but not latency-sensitive scalping.
Ask support for the data center name. Many brokers publish it; IC Markets lists NY4 for MT4 and MT5 and LD5 for cTrader. You can also run mtr to the server’s IP and read the last few hop names, which often carry a city or airport code.
If you run more than one terminal or a heavy EA, yes. On shared-CPU hosts a neighbor’s load shows up as slow ticks and delayed order sends, right when volume spikes. Whatever the host promises, check CPU behavior during a release before you trust it with live money.
Before you pay anyone, run a 24-hour MTR from our looking glass to your broker’s server. If Chicago or Ashburn fits, the plans are on the VPS page.
