Zero knowledge isn't magic; it's math you can verify. The same rule governs Ethereum's supply, except most holders verify nothing and trust the chart instead.
On October 9, 2026, the third-party ledger ethsupply.fyi published a 282-day accounting of Ethereum's money printer. The headline was blunt. Total fee burn — execution-layer base fees plus blob fees — reached 16,524.553 ETH. In the same window, proof-of-stake issuance minted 796,623.377 ETH. The burn offset 2.07% of new supply. Not twenty percent. Two.
I spent a week rebuilding that window from block-level data because a number that clean deserves a second opinion. It holds. Net supply grew by 778,412.846 ETH — roughly 1,007,000 ETH annualized, a net issuance rate near 0.83% per year. Ethereum in 2026 is a mildly inflationary asset wearing the costume of a deflationary one. The costume is stitched from a 2021 memory.
That memory needs correcting before the numbers can land, because a mispriced monetary policy is a mispriced asset.
Context. Ethereum's supply has two moving parts that people routinely conflate. One is issuance: validators receive new ETH for securing consensus, and under proof of stake that issuance scales with the amount staked. More validators, more minted ETH. The other is the burn. EIP-1559 destroyed base fees at the protocol level, and EIP-4844 added blobs, whose data-availability fees are also burned. Priority fees do not burn. They accrue to the block producer. That distinction matters more than most supply dashboards admit, and I will return to it.
The seductive part of EIP-1559 was always its elasticity. The base fee rises with demand and falls with slack, and in a congested network it can destroy more than issuance mints. In 2021 that happened often enough to birth the "ultrasound money" meme. The mechanism was real. The inference drawn from it was not. A burn that exceeds issuance during a high-demand quarter is not the same as a burn that exceeds issuance on average — and holders live on averages, not quarters.
Before the Merge, Ethereum's issuance was a fixed block subsidy that had nothing to do with demand, and the burn was a clean offset against a predictable numerator. The Merge replaced that with a variable subsidy tied to staked ETH. That change is the origin of everything in this ledger. A fixed subsidy can be outrun by a strong burn. A variable subsidy cannot, because the two move together: more confidence in the network means more staking means more issuance means a bigger denominator for the burn to clear. The design made supply neutrality harder, not easier, and the 2026 data is the first full accounting of that trade-off.
Why trust a third-party ledger over the protocol itself? Because Ethereum does not publish a single canonical supply number. Issuance, burn, slashing, and blob fees are computed by clients and aggregated by indexers, and different dashboards count different things. ethsupply.fyi's value is that it separates them, so the offset ratio can be recomputed under different assumptions. When a headline number is contestable, the methodology behind it is the actual finding. Here, the methodology survives scrutiny, and the number it produces is worse than the narrative.
I audited multisig wallets in the aftermath of the 2018 crash and traced Uniswap V2's constant-product swaps through the DeFi Summer of 2020, and the lesson from both is identical: the mechanism is knowable, and the narrative built on top of it usually is not. EIP-1559 is a controller. Read the controller, not the tweet.
Then came Glamsterdam, which added a new variable. It set a conditional gas target of 200 million per block, up from the current 60-million ceiling, with a working target near 100 million under an elasticity multiplier of two. Note the word conditional. That target is not a protocol-enforced switch. It depends on validators actually configuring their clients to raise the gas limit. Consensus does not mandate it. Validators do — and that is a coordination problem dressed as a capacity upgrade.
Core. Start with the break-even math, because the break-even math is where the narrative dies.
For the burn to fully offset issuance at the current 60-million gas limit, the execution-layer base fee needs to sit near 13.85 gwei. Where is it? 0.335 gwei. That is a gap of roughly 41x between reality and the threshold required for neutral supply. Forty-one times. A single order of magnitude would be a slow month. Forty-one is a regime.
Now raise the gas limit, as Glamsterdam contemplates. More gas per block means the same fee pressure spreads across more units, so each unit needs to clear a lower bar — but the absolute burn required is unchanged. At a 200-million gas limit, the break-even base fee lands near 4.16 gwei. Still roughly twelve times current levels. Expanding capacity does not close the gap; in a soft-demand environment it widens it, because more blockspace chasing the same activity pushes the base fee down further. The capacity-maximalist crowd refuses to model this. The controller models it for them.
The mechanics are unforgiving. EIP-1559's base fee is a controller, not a policy. It seeks equilibrium with demand. When L2s absorb the high-value execution that used to land on L1, L1 base fees collapse, and collapse is exactly what the ledger shows. The AMM model hides its truth in the invariant; EIP-1559 hides its truth in the base-fee controller, and the controller is telling us L1 blockspace is oversupplied by an order of magnitude.
Blobs deserve their own line. In one sampled epoch, blob burn came to 0.000442 ETH. That is not a rounding error inside a rounding error — it is the entire data-availability revenue stream for that window, and it is functionally zero. EIP-4844 was sold as a new fee sink. In practice, blob space is so abundant that its market-clearing price sits at the floor. A dedicated DA layer is not earning its keep because there is no scarcity left to price. I have argued for two years that 99% of rollups do not generate enough data to justify a bespoke availability layer, and the blob numbers are the receipt. Every dollar of venture capital chasing "DA as a service" is chasing a fee stream that clears at zero.
Then there is the issuance side, which nobody models when they model burns. Staked ETH sits near 43,754,000, about 35.8% of the 122,116,000 ETH supply, spread across roughly 853,325 active validators. Under proof of stake, issuance rises with the staked amount. So the burn-to-issuance ratio has two levers, and only one of them gets discussed. If staking participation grows, issuance grows, and the denominator of that 2.07% gets bigger — not smaller.
Run the crude APR: roughly 1,031,000 ETH of annual issuance divided by 43,754,000 ETH staked lands near 2.36%, before node, VPS, and cloud costs. Net of operations, a validator might keep 1.5% to 2.0%. That is a thin margin, and thin margins change behavior. A validator deciding whether to keep a machine online at 1.7% net is not the same actor as one earning 5%. The reward is increasingly an inflation transfer rather than a share of protocol revenue, because protocol revenue — the burned fees — is 2.07% of what gets minted.
For scale, Bitcoin's post-2024-halving issuance runs near 0.8% annually. Ethereum's net issuance over this window annualizes to roughly 0.83%. The asset once pitched as the scarcer, more modern monetary network is now printing at the same rate as the one it was supposed to replace. That is not a small irony. It is the collapse of a core pitch, and it happened quietly, in a dataset most holders will never open.
I did institutional custody due diligence ahead of the 2024 spot ETF approvals, comparing threshold-signature and multisig architectures against open-source alternatives, and the supply question never came up. Custodians care about key management, not monetary policy. It should have. An ETF holder buying ETH as a store of value is implicitly underwriting a monetary policy they cannot vote on, enforced by validators they cannot direct, with an issuance rate that rises when confidence rises. That is a strange instrument to package as digital gold.
There is a governance wrinkle worth naming. Gas limit is not set by developers, foundations, or token votes. It is set by validators running clients with a chosen limit. Glamsterdam can suggest 200 million; it cannot enforce it. So the upgrade's actual effect on supply depends on thousands of independent operators making a capacity decision they are not economically punished for getting wrong. If most stay conservative, the 200-million target is a press release. If many chase priority-fee revenue and raise limits into weak demand, they accelerate the burn-issuance imbalance the ledger already shows. Capacity is a choice, not a constant. Neither outcome is coordinated, and both are legitimate.
I verified the accounting method, because a 2% figure is easy to fudge. ethsupply.fyi separates validator-reward issuance, execution-layer base-fee burn, blob burn, slashing, and SELFDESTRUCT burn, and it reports the fee-burn offset separately from the all-burn offset. The all-burn figure is 2.29%, slightly flattering because it folds in slashing. Using the higher number to argue that users are offsetting issuance would be dishonest. The honest figure is 2.07%, and it is the one I would sign my name to.
Contrarian. Here is the blind spot. The market treats Ethereum's supply as a solved problem — EIP-1559 "made it deflationary." What 2026 shows is that EIP-1559 made it conditionally deflationary, and the condition is sustained L1 congestion that L2s were explicitly designed to remove. The roadmap and the burn are pulling in opposite directions. You cannot scale execution off L1 and also count on L1 congestion to burn supply. Pick one, and Ethereum already picked.
Priority fees are a quieter blind spot. They do not burn. They accrue to block producers. So the metric that gets quoted — "fees paid on Ethereum" — conflates two very different things: value destroyed for every holder, and value captured by a subset of validators. When fee revenue "recovers," ask who receives it. If the answer is producers, holders gain nothing except a headline. I have watched this conflation inflate bull cases for two cycles, and it has not become less misleading with repetition.
Then there is the DA narrative. Blob burn at 0.000442 ETH per epoch is not a market; it is an empty room with a price tag. And the L2s paying that near-zero rent are not subsidizing the base layer — they are extracting settlement security at a discount while their own fee revenue stays on their own chains. That is not parasitism in a moral sense. It is parasitism in a mechanical sense, and the host is not being compensated.
The supply-shock trade is the casualty here. For years the bull case rested on a simple mechanic: activity rises, base fee rises, ETH burns, supply falls, price follows. 2026 breaks the first link. Activity can rise and migrate to L2s at the same time, leaving L1 base fees flat while the burn stays decorative. The trade assumed congestion was permanent and monolithic. It was neither. Congestion was a phase, not a policy. Every holder who sized a position on "deflationary by default" now owns an asset whose default is inflation, with deflation as an exception that requires demand L1 no longer reliably receives.
I don't grade mechanisms on their intent. I grade them on their invariants. EIP-1559's invariant is a base fee that tracks demand. It is working exactly as designed, and it is telling us demand is weak. The failure is not in the code. It is in the story we wrapped around the code.
Takeaway. The next two quarters will not be decided by price. They will be decided by whether validators actually raise the gas limit under Glamsterdam — and if they do, the burn-to-issuance ratio gets worse before it gets better, because capacity without demand suppresses the base fee. Watch the base fee, not the burn headline. Watch issuance, not staking yield. Watch who receives the fee, not how much was paid. The mechanism is legible, and it is pointing somewhere uncomfortable: an Ethereum that grows in usage while its native asset quietly inflates. Verify that before you price it.
