Every fee schedule you read is a decision somebody made and could change. The network fee is not one of those. It is the clearing price of an auction that runs every few seconds among everybody who wants their transaction included, and no platform sets it, discounts it, or keeps it. Understanding whose money it is explains most of the confusion around it.
When you move an asset on a blockchain, somebody has to include your transaction in a block and every node has to store it forever. The fee compensates whoever does the including, and it goes to them or is partly destroyed depending on the chain's rules.
It does not go to the platform you are withdrawing from. A platform can charge its own withdrawal fee on top, and many do, but that is a separate line with a separate justification, and the two are frequently presented as one number without being one thing.
This matters practically because the two behave completely differently. A platform's fee is a policy you can compare across providers. A network fee is a market price that is the same for everybody at that moment, and no provider can be cheaper on it than another for the same transaction on the same chain.
A platform fee is a policy. A network fee is a market price. Presenting them as one number hides which half anybody can actually compete on.
Block space is finite by design. Each block holds a limited amount of computation and data, and there are usually more transactions waiting than fit. The mechanism that resolves this is an auction: transactions offering more get included sooner, and those offering less wait.
That is why the fee moves so violently. Demand for inclusion is not smooth, and a period of intense activity can raise the clearing price by an order of magnitude within minutes, then return. The number is not a tariff being adjusted, it is a price being discovered continuously.
It also explains why quoted network fees go stale so fast. A figure published in a help page describes whatever the auction was doing when somebody wrote it, and the only current figure is the one an estimator produces at the moment you look.
The fee is not proportional to the value you are moving. It is proportional to the computational work and the storage the transaction imposes, which is why sending a very large amount and a very small amount cost the same, and why a transaction that touches several contracts costs more than a simple transfer.
This is the single most counterintuitive property and it has a direct consequence: the fee as a percentage of what you are moving is enormous for small amounts and negligible for large ones. The same absolute cost that is irrelevant on a large transfer can exceed the value of a small one entirely.
The practical rule follows immediately. Small, frequent movements are the expensive pattern on any chain with meaningful fees, and consolidating them is the largest saving available. Nothing about the fee schedule changes; the behaviour does.
Many assets exist on several chains, and the same balance can usually be withdrawn over any of them. The network fee then depends entirely on which one you choose, and the difference between the cheapest and the most expensive option for the same asset is routinely large.
The choice is not free of consequences. The asset has to be received on the same network by whoever you are sending it to, and sending on a network the destination does not support is one of the most common ways balances are lost permanently. The saving is real and it is conditional on the destination.
The habit worth building is to check the destination's supported networks before choosing, rather than after. That check takes seconds, it is available on any receiving platform, and it converts a routine decision into an informed one at no cost.
Three levers actually reduce network cost and they are all behavioural. Consolidating many small withdrawals into fewer large ones, because the fee does not scale with value. Choosing a cheaper supported network when the destination allows it. And moving at a time when the auction is less contested, which is visible in any fee estimator.
Platforms apply a fourth on their side by batching: combining many users' withdrawals into a single transaction, which spreads one network fee across all of them. That is why a platform's stated withdrawal fee can be lower than the network fee at that moment, and it is a genuine efficiency rather than a subsidy.
None of these are clever. They are the whole of what is available, which is worth stating plainly because the subject attracts a great deal of advice that is more elaborate and less effective. There is no technique that makes an uncongested chain cheaper than it is.
Trading fees and network fees are charged at different moments for different things, and conflating them produces bad comparisons. A trading fee is charged when an order executes, is set by the venue, and is a proportion of the value traded. A network fee is charged when an asset moves on a chain and is proportional to computation.
A trader who moves assets rarely and trades often pays almost entirely the first. A holder who trades rarely and moves assets often pays almost entirely the second. These are different cost structures, and advice written for one is frequently misapplied to the other.
The published trading schedule and what it returns to the client are here for reference, and they are the part a venue actually controls. The network fee is not on this table, because nobody at a venue decides it.
| Market | Maker | Taker | What it means |
|---|---|---|---|
| Buying and selling assets outright | 0.100 % | 0.100 % | 0.0100 % |
| Leveraged positions on contracts | 0.020 % | 0.060 % | 0.0060 % |
Second-layer networks reduce the fee by doing the execution elsewhere and writing a compressed record to the base chain, so the base chain cost is shared across many transactions at once. A transaction on such a network therefore carries a small share of a base chain fee rather than a whole one.
That share is not fixed. It moves with base chain congestion and with how full the batch is, which means a layer's fee is lower than the base chain's but tracks it. A period of expensive base chain activity makes layers more expensive too, in proportion.
The practical point for anybody moving assets is that layers are a genuine reduction rather than an illusion, and they come with their own considerations about which destinations accept them. The saving is real and the compatibility check is the same one as above.
A transaction that offers too little is not rejected, it waits. It sits pending until the auction price falls to meet it or until it is dropped, and during that time the assets are neither in the old place nor the new one. Most interfaces do not explain this well and it is the source of a great deal of unnecessary alarm.
Some chains allow replacing a pending transaction with a higher-fee version, which resolves this directly. Others do not, and the only option is waiting. Knowing which situation you are in before you need to is worth the two minutes it takes to find out for the chains you actually use.
The other common failure is sending to an address on the wrong network. The transaction succeeds on its own chain, the destination never sees it, and recovery depends entirely on whether anybody controls the receiving address on that chain. It is not a rare event and it is completely preventable by the check described above.
A transaction offering too little is not rejected. It waits, and during that time the asset is in neither place, which is where most of the alarm comes from.
Fee estimators return a range and a confidence, and the useful reading is the whole range rather than the headline number. The low end is what you pay if you are willing to wait, the high end is what buys immediate inclusion, and choosing between them is the actual decision.
The estimate is also a forecast of an auction rather than a quoted price, and it can be wrong in both directions. Paying the high end when the auction cools is overpaying, and paying the low end during a rising period is the pending-transaction situation described above.
For anything not urgent, the low end and patience is the cheaper choice most of the time. For anything time-sensitive, the premium for immediacy is a real cost of the urgency rather than a fee to be optimised away, and treating it as one leads to the worst outcome available.
Network fees are one of the few costs in this industry that nobody markets, discounts or negotiates, because there is nobody to negotiate with. That makes them unusually honest and unusually hard to reduce by choosing a provider, which is the reverse of every other line on the bill.
What remains available is entirely behavioural: fewer and larger movements, the right network for the destination, patience where it is affordable, and reading the whole range rather than a headline. Those four are the complete list and they are worth more than any comparison.
The broader point is worth keeping. A cost you cannot negotiate is a cost to be designed around rather than shopped for, and treating it as the second is how people end up paying the first many times over in small amounts they never added up.
Whoever includes your transaction in a block, or it is partly destroyed depending on the chain's rules. It does not go to the platform you are withdrawing from. A platform can charge its own withdrawal fee on top, and the two are frequently presented as a single number without being one thing.
Block space is finite and there are usually more transactions waiting than fit, so inclusion is decided by auction. Demand is not smooth, and a period of intense activity can raise the clearing price by an order of magnitude within minutes. It is a price being discovered, not a tariff being adjusted.
No. The fee is proportional to the computational work and storage the transaction imposes, not to the amount moved. The consequence is that the fee is enormous as a percentage of a small transfer and negligible on a large one, which makes small frequent movements the expensive pattern.
Because many assets exist on several chains and the fee depends entirely on which one you use. The saving is real and conditional: the destination must support the network you choose, and sending on an unsupported one is among the most common ways balances are lost permanently.
Consolidating many small withdrawals into fewer large ones, choosing a cheaper supported network, and moving when the auction is less contested. Platforms add a fourth on their side by batching many users' withdrawals into one transaction, which spreads a single network fee across all of them.
A trading fee is charged when an order executes, is set by the venue, and is a proportion of value traded. A network fee is charged when an asset moves on a chain and is proportional to computation. Someone who trades often and moves rarely pays almost entirely the first.
The transaction is not rejected, it waits, and during that time the assets are in neither place. Some chains let you replace a pending transaction with a higher-fee version; others do not, and the only option is waiting. Knowing which applies to the chains you use is worth finding out in advance.
As a range rather than a headline number. The low end is what you pay if you are willing to wait, the high end buys immediate inclusion, and choosing between them is the decision. It forecasts an auction rather than quoting a price, so it can be wrong in both directions.