Fee Markets & EIP-1559

Before 2021: a first-price auction

Until August 2021, an Ethereum transaction carried a single gasPrice. Miners sorted pending transactions by that price, filled blocks up to a hard gas cap, and kept the whole fee. EIP-1559’s authors listed the problems with this setup384:

• Guesswork and overpaying. You pay what you bid, so wallets had to predict others’ bids, and users bid high to be safe.
• Volatility. With a fixed block size, a small rise in demand above capacity could make fees jump sharply from block to block.
• Delays. Underbid during a spike and your transaction sat in the mempool for minutes or hours.

How EIP-1559 works

EIP-1559 went live with the London upgrade at block 12,965,000 on August 5, 2021386. Its moving parts384387:

• Target and maximum. Each block has a gas limit and a gas target equal to the limit divided by the elasticity multiplier, which is 2. At London that meant a 15 million target inside a 30 million limit. Blocks can stretch to twice the target when demand surges.
• Base fee. A per-gas price set by the protocol from the previous block alone: it rises if that block used more gas than the target and falls if it used less, in proportion to the gap, by at most 1/8 (12.5%) per block. It started at 1 gwei. It is burned.
• Priority fee (tip). Paid to the block producer, now the validator proposing the block, for including your transaction; it also lets you outbid others when blocks are full.
• Max fee. A ceiling on the total you will pay per unit of gas. Your transaction is valid only while the base fee is at or below it.
• What you actually pay. Per gas, you pay the base fee plus whichever is smaller: your tip, or your max fee minus the base fee. The difference between your max fee and that amount is never charged, and gas your transaction doesn’t use (below its gas limit) is refunded384.

Older “legacy” transactions still work. Their single gas price is treated as both the max fee and the tip, so everything above the base fee goes to the validator384: a 30 gwei legacy bid at a 20 gwei base fee hands over a 10 gwei tip you didn’t need to pay.

Most of the time the base fee drifts slowly and a small tip is enough. In a sudden surge, blocks fill to the maximum and the base fee climbs 12.5% every 12-second slot; until it catches up, tips matter again and the market briefly behaves like an auction.

The limit itself has risen since London: as of October 2026 it is 60 million gas, so the target is 30 million391.

Why burn the base fee?

Paying the base fee to the block producer looks simpler but breaks the mechanism in two ways.

First, a producer who collected the base fee could fill blocks with their own transactions to push the base fee up at no cost, since they would get the money back. EIP-1559 names removing that incentive to manipulate the fee as a reason for the burn, along with making sure only ETH can pay for transactions384.

Second, side deals. Tim Roughgarden’s economic analysis of EIP-1559, published in December 2020, showed that if the producer kept the base fee, producers and users could strike off-chain deals that recreate a first-price auction, bringing back the guesswork. Burning the base fee removes the gain from such deals. The same report argues that tips are needed so producers don’t publish empty blocks, and that tips should go entirely to the producer, since burning them would push the tip market off-chain too385.

The burn does not stop ordering games. Validators earn priority fees plus whatever builders pay them for block-ordering rights; that is MEV, covered in the MEV deep dive.

A worked example

Suppose the base fee is 20 gwei, and you send ETH (a plain transfer uses 21,000 gas). You set a tip of 2 gwei and a max fee of 50 gwei.

• You pay 20 + 2 = 22 gwei per gas: 21,000 × 22 = 462,000 gwei, or 0.000462 ETH.
• Burned: 21,000 × 20 = 0.00042 ETH. To the validator: 21,000 × 2 = 0.000042 ETH.
• Never charged: the remaining 28 gwei of headroom per gas. Your wallet must hold enough to cover the max fee, but you pay only the actual price.

Now the next block. If it was full (60 million gas against a 30 million target), the base fee rises 12.5% to 22.5 gwei. If it was empty, it falls 12.5% to 17.5 gwei. If it sat exactly on target, it stays at 20 gwei.

Why headroom matters: after six full blocks in a row, the base fee is 20 × 1.125⁶ ≈ 40.5 gwei, and after ten it is about 65 gwei. Your 50 gwei max fee survives six full blocks but not ten; the transaction waits until the base fee falls back below 50. Near the ceiling the tip also shrinks: at a base fee of 49 gwei, your effective tip is 1 gwei (50 − 49). Hence a common wallet rule of thumb: a max fee of about twice the base fee plus the tip covers six full blocks in a row, a little over a minute.

Ultrasound money vs. the supply data

Because the base fee is burned, ETH’s supply depends on two flows: new issuance to validators and the burn. The “ultrasound money” meme held that, after the Merge cut new issuance by roughly 88% in September 2022398, the burn would usually exceed issuance and ETH supply would shrink.

For a while it did. Data from the ultrasound.money dashboard, read on October 2, 2026388:

• Supply was about 120.52 million ETH at the Merge in September 2022.
• It fell to a low of about 120.06 million in April 2024.
• It passed its Merge-day level again in February 2025 and stood at about 122.09 million on October 2, 2026, roughly 1.3% higher.
• Over the 12 months to October 2026, supply grew about 0.8%. In the 30 days to October 2, about 1,990 ETH was burned, roughly 66 ETH a day, against issuance of roughly 2,950 ETH a day.
• About 4.64 million ETH has been burned since London in total.

The turn came with the Dencun upgrade in March 2024, which gave rollups a cheap dedicated data lane (next section)225. Activity and fees moved to layer 2s, mainnet congestion fell, and the base fee is now often a fraction of 1 gwei (about 0.1 gwei on October 2, 2026)388. The burn is real, but it tracks demand for mainnet block space. As of 2026, ETH’s supply is growing modestly, not shrinking; treat any “deflationary” claim as a statement about a specific period.

Blobs: a second fee market

EIP-4844, shipped in Dencun, added a second, separate fee market for “blobs”: large chunks of data that rollups post to Ethereum and that nodes keep for about 18 days. Blobs have their own target and maximum per block and their own blob base fee, which rises when blocks carry more than the target number of blobs and falls when they carry fewer. Blob transactions state a separate max fee per blob gas, and the blob fee is burned225.

When blob demand stays below target, the blob base fee keeps falling until it settles at its 1 wei floor; when demand returns, it takes over an hour of near-full blocks to recover, and rollups fall back on first-price bidding meanwhile392. Since Dencun:

• Pectra (May 7, 2025) raised the target from 3 to 6 blobs per block and the maximum from 6 to 9390389. The new 2:3 ratio makes the fee fall faster after empty blocks (about 14.5%) than it rises after full ones (about 8.2%)389.
• Fusaka (December 3, 2025) added EIP-7918, which ties the blob base fee to a minimum set by the execution base fee, so blob space can’t be priced at effectively zero while regular gas costs something391392.
• Blob-parameter-only (BPO) forks then raised capacity without a named upgrade: BPO1 (December 9, 2025) to a target of 10 and maximum of 15, and BPO2 (January 7, 2026) to 14 and 21391. As of October 2026, BPO2’s values are the latest on mainnet; further increases are planned.

The next named upgrade, Glamsterdam, was still in testing in October 2026, with mainnet expected in late 2026. Its gas-repricing proposals include lowering the fixed cost of a basic transaction; check its status before relying on it396.

Bitcoin: still a first-price auction

Bitcoin never adopted anything like EIP-1559. Each transaction offers a fee, quoted as a rate in satoshis per virtual byte (sat/vB); miners fill blocks with the highest rates and keep every fee; nothing is burned. A typical single-input payment of about 140 vB at 5 sat/vB costs 700 sats.

Two tools fix underpriced transactions:

• Replace-by-fee. The sender rebroadcasts a version paying a higher total fee. BIP 125 (2015) defined an opt-in signal for it393. Bitcoin Core 28.0 (October 2024) made its nodes accept replacements by default whether or not the original signaled, so in practice treat any unconfirmed Bitcoin payment as replaceable394.
• Child-pays-for-parent. The recipient (or the sender, via change) spends the stuck output in a high-fee transaction; miners take both to collect the child’s fee. Bitcoin Core 28.0 also added limited package relay, so a low-fee parent can travel with its child394.

The bigger open question is security. Miners are paid the block reward: a subsidy of new coins plus fees. The subsidy halves every 210,000 blocks (it has been 3.125 BTC since the April 2024 halving), and Nakamoto’s whitepaper expected fees to take over eventually1. Fees are volatile. The April 2024 halving block, mined as the Runes token protocol launched, paid about 37.6 BTC in fees, twelve times its subsidy. In the week ending October 2, 2026, fees were about 0.7% of miner revenue, and recommended fee rates that day were 1–3 sat/vB395. Whether fees alone can fund enough hash power once the subsidy has shrunk further is unresolved; this “security budget” question recurs in Bitcoin’s scaling debates.

Setting fees in practice

On Ethereum and its L2s:

• The base fee is not a bid; your wallet reads it from the chain. A small tip is enough unless blocks are full.
• Leave headroom in the max fee. It is a ceiling, not a price.
• To speed up or cancel a pending transaction, send a replacement with the same nonce and a higher tip and max fee. Geth, a major Ethereum client, requires at least a 10% bump by default (100% for blob transactions)397.
• A failed transaction still pays for the gas it used, and one that runs out of gas uses its whole gas limit. Don’t cut a contract call’s gas limit below the wallet’s estimate.
• On a rollup, your fee covers L2 execution plus a share of the cost of posting data to Ethereum, which since Dencun is mostly paid in blob fees.

On Bitcoin:

• Check current fee rates (mempool.space and wallet estimators show them) and pick one for the wait you can accept.
• Use a wallet that supports RBF so you can bump the fee later. If you are receiving a stuck payment, ask whether your wallet supports CPFP.
• Wait for confirmations before treating a payment as final.

Key takeaways

• Before August 2021, Ethereum, like Bitcoin today, used a first-price auction: everyone paid their own bid, which meant guesswork, overpaying, and sharp fee swings.
• EIP-1559 (London, August 5, 2021) introduced, together, a gas target at half the limit, a burned base fee that moves at most 12.5% per block, a tip to the block producer, and a max fee that caps what you pay.
• The burn stops producers inflating the base fee for themselves or striking off-chain deals that recreate the auction.
• The “ultrasound money” story held from the Merge to early 2024. After Dencun moved activity to rollups, the burn fell, and as of October 2026 ETH supply is about 122.09 million and growing about 0.8% a year.
• Blobs run their own burned fee market; Pectra, Fusaka (with a blob-fee floor) and two BPO forks have raised it to a target of 14 and a maximum of 21 blobs per block.
• Bitcoin keeps a first-price auction with RBF and CPFP for stuck transactions, and faces an open question about paying for security as the subsidy shrinks.