Latency math: why exchange-proximity matters for futures execution
2026-05-06 · ~6 minute read
“Low-latency” is the most over-claimed phrase in retail crypto trading. Most consumer-grade bots route every order through a server in the wrong continent and call it fast because the UI feels snappy. This post is the math we wish someone had written for us five years ago.
What round-trip time is actually doing
Round-trip time (RTT) is the wall-clock interval between “our order leaves our process” and “the exchange's acknowledgement reaches our process.” It includes the outbound network leg, exchange match-engine queueing, and the inbound network leg. It does not include the time we spent deciding to send the order. (That is a separate budget; we measure both.)
For a passive limit order resting on the book, RTT does not really matter — once the order is queued, the exchange's clock is the only one that counts. For an aggressive order taking out a quote you saw a moment ago, RTT is everything: every millisecond you spend in flight is another millisecond someone else can pull their quote.
Where region-of-deployment buys you something
Exchange match engines for the major crypto venues live in a handful of physical regions — typically Singapore, Tokyo, and a US-East zone. From a server in the same region you can expect sub-50 ms round trips. From a server in the wrong region, the same call will routinely take 150 to 250 ms, dominated by the speed of light through fiber, not by the exchange.
The practical consequence: if your bot is pinging a Singapore match engine from a US data center, you have given up roughly 200 ms on every order. On an aggressive order, that is the difference between filling at the displayed price and filling at the next tick — or not filling at all.
Where it stops buying you anything
Past somewhere around 30 to 50 ms, the marginal value of further latency reduction collapses for retail-style strategies. Real high-frequency shops are competing on micro-second co-located cross-connects — a budget that does not pay back at retail size, and we are honest about not playing that game (see the post on what high-frequency actually means in retail crypto).
The realistic target for a retail-grade execution bot is: be in the right region, on a low-jitter network path, with a well-tuned client. That gets you to the floor where the bottleneck stops being you and starts being market structure.
How we measure (and why we will publish it)
We measure RTT continuously from every region we deploy in, against every exchange we trade. The number you see on the proof page is not the “best ever recorded” — it is a rolling measurement, which is the only honest version. Marketing teams love to print the floor; traders care about the median, the 95th percentile, and the worst minute of the day.
When the live data feed lights up in Phase 5, the latency table on the proof page will refresh continuously and label the region currently selected by the router for live order flow.