Most assets have a supply schedule set in advance. Ether has one side set by protocol and the other set by how much the network is used, which produces behaviour with no direct equivalent.
The base fee is destroyed rather than paid
Under the current fee mechanism, each transaction pays a base fee determined by congestion, and that amount is removed from supply permanently.
The block producer receives only the priority tip, so the bulk of what users pay during busy periods goes to nobody.
The original purpose was to make fees predictable and to stop producers from manipulating a pure auction, and the supply effect was a consequence rather than the goal.
Issuance and destruction now offset
New ether is created to pay validators, at a rate that depends on how much is staked rather than on activity.
The burn depends entirely on activity. When usage is high enough, more is destroyed than created and total supply contracts.
When the network is quiet the opposite holds and supply grows, so the net change alternates rather than following one direction.
Fee revenue became a shared claim
Before the change, transaction fees accrued to whoever produced the block, benefiting a specific participant.
Burning distributes the value across every holder proportionally, by making each existing unit a slightly larger share of the total.
This is closer to a buyback than to a dividend, and it happens continuously without any decision by anyone.
Layer two activity reduced the burn
As transactions moved to rollups, the fees paid on the base layer fell, because a rollup posts compressed data for many transactions in one place.
The base layer earns less per user transaction than it did when everything executed directly on it.
Dedicated data space for rollups reduced that further, which lowered costs for users and reduced the destruction that had been offsetting issuance.
Supply is now a usage measure
The net change in supply over a period is effectively a summary of how heavily the network was used relative to what it pays validators.
That makes it informative about activity, and it also means it cannot be projected forward the way a fixed schedule can.
Anyone modelling future supply is really forecasting demand for block space, which is a substantially harder question than reading a predetermined curve.