Most trading costs are proportional. Spread, price impact and commission all scale with how much you trade, so the cost per dollar is roughly constant and size is broadly neutral.
A network fee is not proportional. It is charged for the transaction, largely irrespective of value, which means the cost per dollar traded falls as size rises — the exact opposite of price impact, which rises with size.
The two costs pull in opposite directions
This is the genuinely useful consequence, and it is a real optimisation rather than a talking point. Trade too small and the fixed fee dominates: a modest position can lose a meaningful share of its value to two transactions before the market does anything. Trade too large and price impact dominates against the available depth.
There is a size in between where the total is least bad, it depends on the specific pool’s depth and the chain’s current fee level, and it moves. Traders who think about impact but treat fees as a rounding error systematically trade too small on expensive chains.
Every leg is a transaction
The round trip is at minimum two. Add an approval, a wrap or unwrap, a hop through an intermediate pair, or a failed attempt that reverted on slippage tolerance — and note that a reverted transaction still costs the fee. You paid to not trade.
That last point deserves weight, because it is the hidden cost of the tight tolerance recommended against sandwich attacks. Protection is bought with occasional reverts, and reverts are not free. It is a genuine trade-off rather than a free lunch, and the right setting depends on the token’s volatility.
Why it shapes whole markets
Fee levels determine which strategies exist at all on a chain. Where transactions are expensive, high-frequency approaches are uneconomic and activity concentrates in fewer, larger trades. Where they are cheap, the opposite — which is why transaction counts are not comparable across chains, one of the differences covered in Solana versus EVM microstructure.
So the fee is not a technicality sitting outside the market. It is part of the microstructure, and it is upstream of what the data even looks like.