Latency is the subject traders spend money on before measuring whether it is their problem. For a narrow set of strategies it is the whole business. For everybody else it is a small component of a cost dominated by two other things, and the money spent on it would have produced more elsewhere.
The number that matters is round-trip time from your decision to a confirmation, and it decomposes into pieces of very different sizes. There is the time for your software to construct and send an order, the network transit to the venue, the venue's own processing and matching, and the return trip carrying the result.
For a participant connecting over the public internet from an ordinary location, transit dominates and is largely a function of physical distance and the route packets happen to take. For a participant with a direct connection in the same building as the matching engine, transit is negligible and the venue's own processing dominates. Those two situations are so different that they barely share a vocabulary.
The decomposition matters because it says where an improvement is available. Halving a component that is five percent of your total changes almost nothing, and most spending on speed is aimed at components that were never the constraint. Measuring the parts before buying anything is the step people skip.
Latency is a sum of four things, and most money spent on speed targets whichever one the seller happens to sell.
The prize is queue position. Orders resting at the same price are filled in the order they arrived, so being first to a price level means being filled before everybody else at it. That is the entire mechanical benefit of speed on a resting order, and it is worth something specific: the difference between being filled and not being filled when only part of the level trades.
For an order that crosses rather than rests, the prize is different. It is arriving before the resting order you wanted is cancelled or taken by somebody else, which matters when a price is briefly available and disappears. That is a real advantage and it applies to a narrower set of situations than the queue one.
The reason this is worth stating precisely is that both benefits require competing with somebody for the same specific thing at the same specific moment. A trader whose orders are not in a race with anybody gains nothing from winning a race, and most orders are not in a race with anybody.
Decomposed honestly, the time between an ordinary trader deciding and their order arriving is dominated by things that have nothing to do with their connection. The decision itself, the interface rendering, the clicks, and the confirmation dialogue account for far more than the network does. A human deciding takes hundreds of milliseconds at best and usually seconds.
Against that, the difference between a good consumer connection and an excellent one is a small fraction of the total. Optimising it is optimising a component that a human sitting in front of the screen has already dwarfed, and no amount of network improvement compresses the part of the delay that is the person.
This is why the honest answer for most traders is that latency is not their problem. The strategies where it is are ones where no human is involved in the decision at all, which is a different activity with different economics and a different capital requirement.
The upgrade people buy is a faster home connection, and the reason it disappoints is that consumer connections are optimised for bandwidth rather than for latency. Downloading faster does not mean a small packet reaches a distant server sooner, and the marketing measures the first while the trading depends on the second.
The route also matters more than the speed, and it is not something a consumer contract controls. Packets from a home connection take whatever path the providers between you and the venue have arranged, which can be substantially longer than the geographic distance and varies from day to day. Paying more for the same route buys nothing.
The genuine improvement available to an individual is unglamorous: connecting to the venue's nearest access point rather than a default one, using a wired connection rather than wireless, and running the software on a machine that is not doing anything else. Those three cost nothing and remove more delay than any consumer upgrade.
The professional version of this is to place your machines physically in the same facility as the matching engine and connect directly to it. That removes the network from the equation almost entirely, and it is why the firms that compete on speed are all in a small number of buildings.
The cost is a business expense rather than a purchase: rack space, direct connectivity, and the engineering to use it. That threshold is the reason the activity is concentrated among firms rather than individuals, and it is also why the competition among them is so intense: everybody at that level has removed the same delays, so the remaining differences are in software and hardware design measured in microseconds.
It is worth knowing this exists even if it will never be relevant to you, because it explains the shape of the market. When a price appears and disappears before a retail interface has repainted, nobody has done anything improper. A participant in the same building acted on it, and the gap between their position and yours is measured in physics.
The firms competing on speed have all removed the same delays. What remains between them is microseconds, which is a different sport from the one most traders think they are in.
Speed competition has a specific economic property: it is a race where being marginally faster than the next participant captures most of the benefit, so the return on investment is extremely nonlinear. That structure produces continuous escalation, because being second fast is worth much less than being first.
The costs of that escalation are paid by the firms involved and passed through in complicated ways. The argument in favour is that competition among fast market makers narrows spreads for everybody, which is true and measurable. The argument against is that some of the speed advantage is used to trade ahead of slower participants rather than to quote to them, which is also true in specific documented ways.
The reasonable position is that both effects exist and their net is genuinely disputed among people who study it seriously. What is not disputed is that a slower participant cannot win the race, which makes the question of who benefits less useful than the question of how to trade without needing to.
Two features change the picture relative to traditional markets. Venues are geographically dispersed rather than concentrated in a few financial centres, so there is no single location that is fast to everywhere, and a participant optimising for one venue is not thereby fast to another. That fragments the advantage.
The second is that a large share of crypto trading happens through public interfaces rather than direct connections, which compresses the range between the fastest and slowest participants relative to markets where everybody serious is colocated. The advantage is still real and the distribution of it is flatter.
Against that, blockchains introduce a category of latency that has no traditional equivalent: the time for a transaction to be included in a block, which is not something anybody can optimise beyond paying more for priority. Where an order involves an on-chain action, that delay dominates everything else and it is measured in seconds rather than microseconds.
The strategies that survive without speed are the ones that do not require being first. Resting orders that wait rather than chase, positions held over hours or days rather than seconds, and instruments where the competition for queue position is not intense are all viable without any latency advantage at all.
The framing that helps is to stop competing on the dimension where you lose. A trader whose approach requires reaching a price before somebody else has chosen an approach that a faster participant will beat consistently, and no amount of connection improvement changes that. A trader whose approach requires being patient is not in that race.
This is not a consolation. It is a description of where an edge is actually available to somebody without institutional infrastructure, and it is a larger space than the coverage of speed suggests. Nobody with a colocated server is competing for a position held over three days.
There is a deliberate version worth mentioning because it inverts the assumption. Some venues have introduced small mandatory delays applied to incoming orders, on the reasoning that a uniform delay removes the advantage of being marginally faster without removing anybody's ability to trade. The mechanism is contested and it exists.
The individual equivalent is choosing instruments and moments where speed is not the deciding factor, which is a decision rather than a limitation. Trading during the busiest hours means competing with everybody; trading a less contested instrument means competing with fewer participants, and the second is available to anybody.
It also inverts the usual advice about execution. A patient order that rests and waits is not disadvantaged by latency at all, because nothing about it is a race. Choosing that mode wherever it is acceptable removes an entire dimension of competition rather than losing it.
Before spending anything, measure. Time from when you send an order to when the confirmation arrives, over many orders, and compare the median against the outliers. If the median is small relative to your holding period, latency is not your constraint and no purchase will change your results.
Then compare it against the other two components of execution cost. If the spread you cross and the impact you cause are each many times your latency cost, which they are for the large majority of traders, then the ranking of what to work on is settled by the measurement rather than by what is interesting.
The measurement usually ends the discussion, which is why it is worth doing first. Traders who run it typically discover that their execution cost is dominated by the hour they trade and the size they send, and that both of those are free to change while a faster connection is not.
For most traders, no. Consumer connections are optimised for bandwidth rather than latency, the route is not something a consumer contract controls, and the delay is dominated by the human deciding rather than by the network. A wired connection and a machine doing nothing else remove more delay than any upgrade.
Queue position on resting orders, since orders at the same price fill in arrival order, and arriving before a resting order is cancelled on orders that cross. Both require racing somebody for the same thing at the same moment, and most orders are not in a race with anybody.
Placing your machines in the same facility as the venue's matching engine and connecting directly, which removes the network almost entirely. It is a business expense rather than a purchase, which is why speed competition is concentrated among firms and why the remaining differences between them are measured in microseconds.
Both effects are real and the net is genuinely disputed. Competition among fast market makers narrows spreads, which is measurable. Some speed advantage is used to trade ahead of slower participants, which is documented. What is not disputed is that a slower participant cannot win the race.
Yes, in two ways. Venues are geographically dispersed rather than concentrated, so no single location is fast to everywhere, which fragments the advantage. And blockchains add a delay measured in seconds for on-chain actions, which nobody can optimise beyond paying for priority.
Anything that does not require being first: resting orders that wait rather than chase, positions held over hours or days, and instruments where competition for queue position is not intense. Nobody with a colocated server is competing for a position held over three days.
On the reasoning that a uniform delay applied to incoming orders removes the advantage of being marginally faster without removing anybody's ability to trade. The mechanism is contested and it exists, and the individual equivalent is choosing instruments and moments where speed is not the deciding factor.
Time your orders from send to confirmation, over many orders, and compare the median against your holding period and against your other execution costs. For the large majority of traders the spread crossed and the impact caused are each many times larger, which settles the ranking of what to work on.