Each block has room for a limited amount of computation and data, that room expires when the block is produced, and it cannot be stored for later. Users bid for inclusion, and the clearing price is what everyone calls a transaction fee. Very little about a chain's economics makes sense until this auction does.
A scarce good that perishes every few seconds
The limit is deliberate. If blocks could be arbitrarily large, the cost of running a full node would rise until only well-funded operators could verify the chain, and open verification is the property the system exists to provide. So capacity is capped, in bytes on Bitcoin and in units of gas on Ethereum and similar chains, where gas prices computation and storage rather than raw size, so a simple transfer and a complex contract call consume very different amounts.
Waiting transactions sit in the mempool, a loose set of pending messages each node keeps for itself. There is no single queue and no ticket order. Whoever builds the next block selects from what they can see, ordinarily by fee density, so a transaction competes against everyone else who wants the same slot at the same moment. When demand is low the auction clears near the floor; when a popular mint or a wave of liquidations arrives, it does not, and the price can move by an order of magnitude within minutes.
This is why fee spikes are short and sharp rather than gradual. Supply is perfectly inelastic in the short run: no amount of willingness to pay creates additional room in the block being produced right now.
How the auction is actually run
Bitcoin uses a straightforward pay-what-you-bid market in satoshis per virtual byte. Underbid, and the transaction waits, possibly for days, until demand falls or it is replaced by a higher-fee version of itself.
Ethereum changed its mechanism with EIP-1559 in August 2021. Each block carries a protocol-computed base fee that rises when the previous block was more than half full and falls when it was less, moving in bounded steps so the price adjusts over a few blocks rather than through a bidding scramble. The base fee is burned, permanently removing that ether from supply. On top of it, a priority fee goes to whoever produces the block, and that is where competition for position within a block now lives.
The mechanism did two distinct things, and they are often conflated. It made fees far easier to estimate, because the base fee is known in advance for the next block rather than guessed. Separately, it redirected a share of fee income from block producers to all holders through burning. The first is a user-experience change; the second is a change in who captures the money.
Who receives the money
Split every payment by destination, because the two halves mean different things. The portion paid to miners or validators is supply-side revenue: it compensates the parties who secure and operate the chain, and it is a cost of production rather than a benefit to holders. The portion burned, or otherwise directed at the asset itself, behaves as holder revenue, because it reduces supply in a way every holder shares proportionally without anyone deciding to distribute it.
Here the equity vocabulary needs care. Burned fees resemble a share repurchase in arithmetic, but there is no board authorizing it, no obligation to continue, no accounting statement recognizing it, and no residual claim standing behind it. Fee income is not earnings: nobody is required to collect it on holders' behalf, and a governance vote or a client upgrade can change the split. A ratio such as market cap to fees is not a price-to-earnings ratio and does not carry the same meaning; it is a rough activity multiple whose denominator can move by an order of magnitude with a single change in demand, and whose numerator is a market opinion rather than a book value.
There is also a security dimension that pure income framing misses. On a chain whose issuance declines over time, fees are eventually expected to fund most of the payment to block producers. How much security spending a chain can sustain from fees alone is an open question about the long-run design, not a settled fact, and it is the reason fee series are watched by people with no trading interest at all.
Rollups purchase block space wholesale
Fees also explain why Layer 2 networks exist. A rollup executes transactions off the main chain and posts data back to it, so thousands of user transactions share the cost of one settlement footprint. The economics are those of a wholesaler: a rollup's sequencer charges users, pays the base layer for data availability, and keeps the difference. Ethereum's introduction of blobs, a separate and lower-cost data lane whose contents expire after a short window, gave rollups a dedicated market rather than forcing them to compete with ordinary transactions for identical space.
One consequence matters when reading any fee chart. Headline fee totals on a base layer can fall while total usage rises, because activity moved to networks that purchase space in bulk. A fee series read without knowing which layer captured the activity produces the wrong conclusion about demand, and the same applies to transaction counts.
Ordering has value of its own
Fees pay for inclusion, but position within a block is separately valuable. Observing a pending trade and placing transactions around it can be profitable, which is maximal extractable value, and its unpleasant forms include front-running and the sandwich attack. Specialized builders now assemble blocks and bid for the right to have them proposed, so part of what looks like a fee market is really an auction for ordering rights. This is a cost borne by users that rarely appears in any fee metric, and its size is estimated rather than measured directly.
Reading fees without over-reading them
The useful series are fees over 24 hours for level, median transaction fee rather than the average for what a normal user pays, since a handful of complex contract calls drag an average upward, and the amount burned where a burn exists. Fee yield expresses annualized fees against market capitalization, which is a comparison device rather than a yield anyone receives. All of these are denominated in a currency whose own price moves, so a fee series in dollars mixes activity with price and is best read next to the native-unit series.
The fees section holds these series with the burn and supply-side split shown separately, and every definition behind them is written out in the metric catalog. The next lesson follows a single payment from signature to confirmation, which is where the fee market stops being abstract.