A portion of every Ethereum transaction fee is destroyed rather than paid to anyone. The design solves a specific incentive problem, and the effect on supply is a consequence rather than the purpose.
The fee has two parts with different destinations
Each block carries a base fee per gas unit, computed by the protocol from how full recent blocks have been.
Users may add a priority tip, which goes to the block producer as payment for inclusion and ordering.
The base portion goes nowhere. It is subtracted from the sender's balance and removed from the supply entirely.
Paying it to producers would corrupt the mechanism
If block producers received the base fee, they would benefit from it being high, and the base fee rises when blocks are full.
They could fill their own blocks with self-generated transactions to push the fee upward, paying themselves and raising the cost for everyone else.
Burning removes the payoff. A producer that manufactures congestion pays the base fee and receives nothing back, so the manipulation costs money instead of earning it.
Predictable pricing was the goal
Before the change, users bid blindly in a first-price auction and routinely overpaid or were left waiting through several blocks.
A protocol-computed base fee that adjusts with demand gives wallets a figure they can estimate reliably, with the tip handling short-term competition.
Making that fee credible required it to be unpayable to anyone with influence over block contents, and destruction is the simplest way to guarantee that.
Supply becomes usage-dependent
New issuance rewards validators, while the burn removes currency in proportion to how much the network is used.
When activity is heavy enough for the burn to exceed issuance, total supply falls for that period. When activity is light, supply grows.
Net issuance therefore varies with demand for block space rather than following a fixed schedule, which is a different property from a capped supply and is often conflated with one.
Rollups changed where the burn comes from
As activity moved to layer two networks, individual users stopped paying base fees on the main chain directly.
Rollups pay instead, in bulk, when they post data and settle, so the burn now depends heavily on how much those networks are publishing.
Cheaper data availability lowers that contribution, which is why the relationship between network usage and supply is looser than it was when most transactions happened on the base layer.