by Fei Wu - This work was conducted during my internship at the EF. I thank my mentor @misilva73 for valuable discussions, feedback, and comments.
Overview
Ethereum’s fee market is moving from a single gas unit that meters and prices every resource to a multi-dimensional design. In Glamsterdam, EIP-8037 introduces multi-dimensional metering, which meters state creation gas separately. EIP-7999 further introduces individual base fee and resource-specific capacity rules (i.e., gas target and limit) for all EVM resources, which, after Glamsterdam, will be state, data, and execution.
This transition requires understanding how the demand of each resource responds to price independently. In this research, we extend the prior state-versus-burst analysis by decomposing the response to historical fee changes into three resource-level demand curves, using three gas-limit increases as the identifying variation.
We then combine the estimated demand curves with counterfactual Glamsterdam gas accounting. Glamsterdam remains a one-dimensional market: execution and data form the regular-gas branch, state creation forms the EIP-8037 state branch, and the bottleneck determines the shared base fee. Studying this equilibrium shows how Glamsterdam repricing and additional capacity jointly affect the effective prices of state, data, and execution.
The estimated elasticities and counterfactual gas accounting of state and execution can be carried into future studies of EIP-7999.
The results presented in this post can be reproduced from this repository.
Main results
-
Resource demand responds differently to the shared fee. Under the independent-demand benchmark, the estimated own-price elasticities span 0.079–0.121 for execution, 0.201–0.229 for data, and 0.254–0.478 for state creation. State is the most price-responsive resource and execution the least responsive.
-
Glamsterdam substantially changes the gas accounting of the same activity. Replaying February–May 2026 activity under the Glamsterdam accounting gives metering multipliers of 1.538 for execution, 1.969 for data, and 5.656 for state.
-
Within the uncapped isoelastic benchmark, the binding branch is stable across elasticity uncertainties. Regular gas determines the shared base fee at 60M and 100M gas limits, while state gas determines it at every tested limit from 150M through 300M. Across the four elasticity estimates, the equilibrium base fee spans 0.004004–0.009999 gwei at 100M, 0.000157–0.001491 gwei at 200M, 0.000065–0.000935 gwei at 250M, and 0.000032–0.000639 gwei at 300M.
-
The high-capacity results depend strongly on extrapolating the state-demand curve. Clearing the 150M, 200M, 250M, and 300M limits requires state-creation activity at 2.53\times, 3.37\times, 4.21\times, and 5.06\times its historical anchor, respectively. An illustrative 2\times state-demand cap instead makes regular gas bind throughout the 150M–300M range and sharply lowers the shared fee. The fees and the binding branch under high-capacity scenarios should therefore be interpreted as conditional on the assumed state-demand tail.
Independent resource-demand model
Let
The notation is shared by the elasticity and equilibrium calculations:
| Notation | Meaning |
|---|---|
| q_i, q_i^0 | Resource-i activity in historical gas-equivalent units, and its February–May 2026 anchor |
| p_i, p^0 | Effective resource price and historical common-price anchor |
| \epsilon_i | Own-price elasticity of resource i |
| Q, s_i=q_i/Q | Total measured gas-equivalent activity and resource share |
| \bar q_i, \bar s_i | Quantity and share in the last clean 45M gas-limit regime used for elasticity recovery |
| b, b^* | Candidate and equilibrium Glamsterdam shared base fee |
| m_i, g_i=m_iq_i | Glamsterdam metering multiplier and resulting metered gas |
| G, T=G/2 | Glamsterdam gas limit and target |
We model the demand curve for each resource as
Under the historical one-dimensional fee market, all three resources face the same base fee:
The historical fee market supplies common-price variation. More generally, resource demand could depend on all three effective prices, q_i=q_i(p_{\mathrm{execution}},p_{\mathrm{data}},p_{\mathrm{state}}). The gas-limit increase events therefore identify a directional response along p_{\mathrm{execution}}=p_{\mathrm{data}}=p_{\mathrm{state}}=p:
The independent benchmark sets the cross-price terms to zero. Under that restriction, the common-price response becomes the resource’s own-price elasticity:
The resource responses relative to execution are
A positive \Delta_i means that resource i expands more than execution when the common fee falls. At the share anchor, aggregate elasticity is
The three resource elasticities therefore satisfy
This specification makes each resource quantity a function of its own effective price. Cross-price substitution and transaction-level bundling are outside this Glamsterdam calculation.
Independence as an identification closure
In general, demand for each resource may depend on all three prices:
Under the current Ethereum one-dimensional fee market, execution, data, and state activity faced the same base fee. The three gas-limit increase events identify each resource’s response to a common price change,
which is the sum of the own- and cross-price responses for each resource. However, it does not separately recover the own- and cross-price terms of a full three-resource demand system. Recovering the individual entries of the full 3\times3 elasticity matrix would require variation that moves execution, data, and state prices separately. Historically, the prices of these resources were never changed separately.[1]
The independent-demand model closes this identification gap by setting the cross-price terms to zero and interpreting each measured common-price response as an own-price elasticity.
This does not claim that transactions literally demand the three resources independently. It is an identification closure: among models consistent with the three observed responses, it introduces no cross-price coefficients that the historical variation cannot estimate. It is useful for Glamsterdam analysis because the fee market still has one shared base fee, although the different metering multipliers create a one-time change in relative effective prices.
The approximation is most plausible when fee changes mainly affect the number of historical-like transactions, leaving the resource composition reasonably stable. It may break down for transactions that consume multiple resources simultaneously. For example, contract deployment and storage writes jointly consume state and execution (moreover, block-level access list as data resource under EIP-7999). A price change in one resource can therefore alter several resource quantities simultaneously
The resulting equilibrium should therefore be interpreted as a reduced-form benchmark conditional on the independence closure, rather than as a structural transaction-demand model.
Empirical elasticity estimates
Data and gas-limit increase events
The daily sample runs from 2025-01-01 through 2026-05-31 and contains three gas-limit increases:
| Event | Date | Gas-limit change |
|---|---|---|
| 30M to 36M | 2025-02-04 | +20% |
| 36M to 45M | 2025-07-21 | +25% |
| 45M to 60M | 2025-11-25 | +33% |
The current-rule accounting separates block gas into three mutually exclusive components:
- State is estimated through a proxy using positive account, storage, and contract-code creation, converted using 25,000 gas per 112-byte account, 20,000 gas per 32-byte storage slot, and 200 gas per contract-code byte.
- Data is charged calldata gas, including the EIP-7623 floor after Pectra.
- Execution is total block gas minus data and state gas. It retains the 21,000 intrinsic gas per transaction so the three components sum to observed block gas.
Across the full sample, the mean resource composition is:
| Resource | Mean gas per block | Share |
|---|---|---|
| Execution | 17.33M | 73.6% |
| Data | 0.81M | 3.4% |
| State | 5.42M | 23.0% |
The elasticity recovery uses the clean pre-Fusaka regime as its share anchor: 75.7% execution, 3.6% data, and 20.7% state, as we explain the reason below.
Event study
Daily resource quantities and shares, the median base fee, and log quantity ratios relative to execution around the three gas-limit changes.
For the event-level table below, each pre- and post-period uses up to 35 days, and the three days surrounding the gas-limit change are excluded. The first event has 31 pre-event days because the sample begins on January 1. Resource gas per block is calculated from the aggregate pre- and post-period totals; the fee measure is the median of daily median base fees. The alternative window lengths are reported in the next subsection.
| Event | \Delta\ln p | \epsilon_{\mathrm{agg}} | \Delta_{\mathrm{state}} | \Delta_{\mathrm{data}} | \epsilon_{\mathrm{execution}} | \epsilon_{\mathrm{data}} | \epsilon_{\mathrm{state}} |
|---|---|---|---|---|---|---|---|
| 30M to 36M | -2.114 | 0.087 | 0.311 | 0.117 | 0.011 | 0.128 | 0.322 |
| 36M to 45M | -0.887 | 0.251 | 0.116 | 0.100 | 0.223 | 0.323 | 0.339 |
| 45M to 60M | -1.254 | 0.234 | 0.233 | -0.094 | 0.180 | 0.086 | 0.413 |
The first two events form the clean-event sample. The third event’s post-period overlaps Fusaka and the first BPO blob-capacity increase, which changed the relative appeal of blobs and L1 calldata during the same period.
A diagnostic over a shorter window from November 26 through December 2 before Fusaka gives a fee decline from 0.160 to 0.047 gwei and a 16% increase in calldata bytes per block, implying \Delta_{\mathrm{data}}\approx0.26. However, over the full 35-day post-period, it gives \Delta_{\mathrm{data}} = -0.094. This inconsistency supports excluding the third event from the elasticity recovery.
Window robustness
The following table averages the two clean-event statistics within each window, then applies the fixed pre-Fusaka share anchor. Each row is one linked elasticity vector used in the equilibrium solver.
| Window | \epsilon_{\mathrm{agg}} | \Delta_{\mathrm{data}} | \Delta_{\mathrm{state}} | \epsilon_{\mathrm{execution}} | \epsilon_{\mathrm{data}} | \epsilon_{\mathrm{state}} |
|---|---|---|---|---|---|---|
| 21 days | 0.195 | 0.085 | 0.361 | 0.117 | 0.202 | 0.478 |
| 35 days | 0.169 | 0.108 | 0.214 | 0.121 | 0.229 | 0.335 |
| 60 days | 0.127 | 0.123 | 0.198 | 0.082 | 0.205 | 0.280 |
| 75 days | 0.119 | 0.123 | 0.175 | 0.079 | 0.201 | 0.254 |
The 75-day row serves as an extended sensitivity. Its first-event pre-period is still limited to 31 days, creating an uneven comparison, and the longer post-periods are more exposed to slow-moving changes.
Across these windows, aggregate elasticity spans 0.119–0.195. Data elasticity is comparatively stable across windows, near 0.202–0.229. State elasticity spans 0.254–0.478. It falls as the window expands, consistent with a strong initial response that partly reverses. Execution elasticity also varies substantially from 0.079–0.121 (roughly 35% difference). Nevertheless, the ordering is stable even though the magnitude of the state response declines as the window expands.
We note that these estimates describe responses around the historical gas-limit increase of 20% and 25%. They do not directly observe activity several times above its anchor or fees orders of magnitude below the reference p^0. Continuing the same elasticity over that distance is an isoelastic functional-form assumption. The four-window range measures sensitivity to the chosen window length, while uncertainty about the shape of demand far from the anchor remains outside that range.
Glamsterdam shared fee market
Glamsterdam retains one EIP-1559-style base fee while tracking regular gas and state gas as separate branches. Regular gas contains execution and data; the state branch contains state creation:
The shared fee responds to the branch with more metered gas:
For gas limit G, the target is T=G/2. The equilibrium base fee b^* solves
The equilibrium calculation is a feedback loop. A trial shared fee b is first multiplied by m_i, giving a different effective price for each resource. Those prices determine execution, data, and state activity through their respective demand curves. The same multipliers then convert activity back into Glamsterdam gas, and the larger of the regular and state branches is compared with the target. The equilibrium calculation adjusts b until the larger branch reaches T. This fixed point is a static comparison of demand and capacity under the stated assumptions.
February–May 2026 anchor
The equilibrium anchor covers February 1 through May 31, 2026: 120 days and 860,505 blocks. The historical price anchor p^0 is the median of daily median base fees, 0.1069 gwei.
| Resource | Historical gas-equivalent quantity per block | Share |
|---|---|---|
| Execution | 23.942M | 78.84% |
| Data | 1.181M | 3.89% |
| State | 5.244M | 17.27% |
| Total | 30.367M | 100% |
This recent anchor is distinct from the pre-Fusaka share anchor used to recover the elasticities. The former sets q_i^0 and p^0 for equilibrium; the latter supplies \bar s_i in the elasticity decomposition.
Why metering multipliers are needed
The demand anchor uses the historical gas schedule, whereas the target is defined under Glamsterdam gas accounting rules. For the same underlying activity, the metering multiplier is
At candidate base fee b, one historical gas-equivalent unit of resource i has effective price
Demand responds to this price ratio, and the resulting activity is converted back into Glamsterdam gas:
Each multiplier is calculated from total counterfactual gas divided by total historical gas for the same February–May activity:
The resulting multipliers are:
| Resource | Multiplier | Accounting replay |
|---|---|---|
| Execution | 1.538 | EIP-8038 and EIP-2780 opcode, transaction-path, and refund accounting |
| Data | 1.969 | EIP-7976 transaction floor plus the EIP-7981 access-list charge |
| State | 5.656 | EIP-8037 byte accounting with CPSB = 1530 |
Full-range aggregates cover February–May 2026 and are constructed from Xatu datasets. We augment them with an Erigon RPC calibration sample targeting 50 uniformly sampled blocks per day (6,000 blocks in total). Erigon node supplies complete access and authorization lists and account-level pre/post traces.
Execution under EIP-8038 and EIP-2780
The execution replay applies EIP-8038’s state-access prices and EIP-2780’s intrinsic-gas paths to the observed transactions. Storage-write repricing is the largest component. EIP-2780 offsets part of that increase by replacing the 21,000-gas transaction base charge with a 12,000-gas sender component, then adding recipient and value-transfer charges according to the transaction path.
| Execution accounting | Gas per block |
|---|---|
| Historical regular execution gas | 23.942M |
| EIP-8038 SSTORE increase | +11.687M |
| EIP-8038 cold SLOAD increase | +2.082M |
| Other EIP-8038 changes | +0.635M |
| EIP-2780 intrinsic-gas changes | -1.033M |
| Counterfactual gas before the new refunds | 37.312M |
| Additional effective EIP-8038 refunds | -0.492M |
| Counterfactual execution gas after refunds | 36.821M |
Thus
Refunds are reconstructed for all 67.64 million refund-positive transactions in the 120-day range. The integer reconstruction identifies 99.82% of the observed refund counter; the remaining amount receives the same daily correction rate as the identified transactions. Small access-list and authorization-write components use the archive node because Xatu does not expose their complete transaction contents.
The left panel decomposes the change in execution-gas accounting under EIP-8038 and EIP-2780. The right panel shows the daily counterfactual-to-historical execution-gas ratio across the 120-day sample.
Data under EIP-7976 and EIP-7981
Under the accounting analyzed here, EIP-7976 raises the calldata floor from 10/40 gas per zero/nonzero byte to 64 gas per byte. EIP-7981 charges 64 gas per access-list content byte, with 20 bytes per address and 32 bytes per storage key.
| Data accounting | Gas per block |
|---|---|
| Historical data gas, including the EIP-7623 floor | 1.181M |
| EIP-7976 floor uplift | +0.776M |
| Calibrated EIP-7981 access-list gas | +0.368M |
| Glamsterdam counterfactual data gas | 2.324M |
Therefore
The EIP-7976 increment is calculated transaction by transaction:
Here B_{tx} is calldata size in bytes and g_{\mathrm{body},tx}^{\mathrm{current}} = g_{\mathrm{used},tx}-21{,}000 is the observed transaction gas excluding the historical base charge. The calculation identifies 19.90 million of 270.78 million transactions, or 7.35%, as binding under the EIP-7976 floor. Those transactions add 0.776M gas per block, equal to 65.7% of historical data gas.
Because the numerator adds access-list gas to a historical calldata anchor, the constant data multiplier assumes that access-list activity scales with the same data demand curve as calldata.
State under EIP-8037
The state-creation proxy specified above is repriced under the EIP-8037 state-gas accounting with CPSB = 1530:
| Accounting convention | State gas per block |
|---|---|
| Historical gas-equivalent proxy | 5.244M |
| EIP-8037 state gas for the same estimated creation | 29.663M |
Hence
Glamsterdam equilibrium results
Fee ranges across event windows
The following ranges hold the three metering multipliers fixed and solve the equilibrium separately for each linked elasticity vector.
| Gas limit | Target | Equilibrium base-fee range | Binding branch in every window |
|---|---|---|---|
| 60M | 30M | 0.553747–1.575098 gwei | Regular |
| 100M | 50M | 0.004004–0.009999 gwei | Regular |
| 150M | 75M | 0.000487–0.002720 gwei | State |
| 200M | 100M | 0.000157–0.001491 gwei | State |
| 250M | 125M | 0.000065–0.000935 gwei | State |
| 300M | 150M | 0.000032–0.000639 gwei | State |
The left panel shows the equilibrium shared base fee for each event-window estimate; the shaded area spans the four results and the dashed 75-day line marks the extended sensitivity. The right panel reports regular- and state-branch gas as a percentage of the shared target. Points show the middle of each four-window range and error bars show its endpoints. A branch determines the fee where it reaches 100%.
Metering and the shared fee
The base fee alone does not show the full price change. One unit of historical gas-equivalent activity for resource i faces effective price m_i b^*, so
This ratio is the effective price of one resource unit. A complete transaction charge sums these prices over the amounts of execution, data, and state it consumes.
Each interval combines the resource’s Glamsterdam metering multiplier with the model-implied shared fee. A value below one means that the lower shared fee more than offsets the metering increase. Ranges span the linked 21-, 35-, 60-, and 75-day elasticity estimates.
At an unchanged base fee, metering alone would raise the effective price by 53.8% for execution, 96.9% for data, and 465.6% for state. The shared fee rises in the tight 60M gas limit case, reinforcing those increases. At a 100M gas limit and above, the fee falls enough to offset all three multipliers. The very small ratios at 200M–300M also show how far those equilibria extend beyond the prices observed in the event study.
Why the binding branch changes
For the independent model, resource-i metered demand can be written as
At a candidate shared base fee equal to the historical anchor, b=p^0, the estimated elasticity ranges imply 36.94–37.63M regular gas and 12.95–19.12M state gas. The effective resource prices at this point are m_i p^0. A 60M limit has a 30M target, so the fee must rise until regular gas contracts to the target. At a 100M limit, both branches expand as the fee falls, but regular gas still reaches the 50M target first because it is initially closer to 50M.
State receives the largest metering increase, which initially makes state creation much more expensive than execution or data. It also has the largest estimated elasticity. As additional capacity lowers the shared fee, state demand rebounds faster than the regular branch. Under the isoelastic curves, state reaches the shared target at every tested limit from 150M through 300M. The regular branch then occupies progressively less of the target, ranging from 77–88% at 150M to 46–58% at 300M.
When state binds, the fee solves
or
The exponent 1/\epsilon_{\mathrm{state}} spans 2.09–3.94 across the four estimated elasticities. This steep response explains why the multiplicative spread between the window-specific fees grows at higher limits.
State-demand Saturation Diagnostic
Once state binds, its historical gas-equivalent activity is fixed by q_{\mathrm{state}}=T/m_{\mathrm{state}}:
| Gas limit | State target | Implied state activity | Multiple of the 5.244M anchor |
|---|---|---|---|
| 150M | 75M | 13.26M | 2.53x |
| 200M | 100M | 17.68M | 3.37x |
| 250M | 125M | 22.10M | 4.21x |
| 300M | 150M | 26.52M | 5.06x |
These quantities are fixed points of the estimated demand curves. The 200M-300M cases extend well beyond the observed state-creation anchor, so their fees should be interpreted as model-implied values under the isoelastic extrapolation.
To show how much these high-capacity results depend on that extrapolation, an illustrative sensitivity caps historical gas-equivalent state activity at 2\times its historical anchor:
The cap equals 10.488M historical gas-equivalent units, or 59.325M metered state gas after applying m_{\mathrm{state}}. State would need to reach 2.53× its anchor to bind even the 75M target at a 150M limit. The 2× cap therefore prevents state from reaching every target from 150M through 300M, and the regular branch determines the shared fee.
| Gas limit | Binding branch: uncapped → capped | Model-implied capped fee | Same-window capped / uncapped fee |
|---|---|---|---|
| 60M | Regular → Regular | 0.553747–1.575098 gwei | 100% |
| 100M | Regular → Regular | 0.004004–0.009999 gwei | 100% |
| 150M | State → Regular | 0.0000465–0.000441 gwei | 8.60–37.20% |
| 200M | State → Regular | 0.00000240–0.0000502 gwei | 1.34–10.01% |
| 250M | State → Regular | 2.72\times10^{-7} - 9.58\times10^{-6} gwei | 0.356–3.718% |
| 300M | State → Regular | 4.99\times10^{-8} - 2.53\times10^{-6} gwei | 0.129–1.689% |
Ratios of capped and uncapped fee are calculated within each event-window elasticity specification and then ranged across the four windows.
At 250M, the capped fees are approximately 27–281 times lower than their uncapped counterparts; at 300M, they are approximately 59–772 times lower. The 60M and 100M results are unchanged because the cap does not bind there.
The wider cap sweep maps the branch result without treating any cap as a forecast. A 3× cap permits state binding at 150M; a 4× cap extends it through 200M; and a 5× cap extends it through 250M. State binding at 300M requires demand to remain unsaturated beyond approximately 5× the anchor.
Each cell reports the branch that determines the shared fee for a gas limit and maximum state-activity multiple. The classification is identical across the four linked event-window estimates. The caps diagnose dependence on the unobserved state-demand tail; they are not estimated saturation points.
Interpretation of the high-capacity results
Robust within the uncapped benchmark. State has the largest estimated response to a common fee reduction among all window-length calibrations. With unrestricted isoelastic expansion, the state branch overtakes regular gas as capacity rises.
Conditional across demand-tail assumptions. State binds from 150M to 300M only when its activity can continue expanding to approximately 2.53–5.06× its historical anchor. A 2× cap makes regular gas bind throughout that range, while the wider sweep below moves the transition as the cap increases.
Most model-dependent output. The precise shared fee at 250M–300M depends strongly on the assumed tail of state demand. These values are functional-form fixed points rather than fee forecasts.
Takeaways and Next Steps
The central result is that we estimate the own-price elasticities of EVM resources: 0.079–0.121 for execution, 0.201–0.229 for data, and 0.254–0.478 for state.
In Glamsterdam fee market, we show that regular gas determines the equilibrium fee at 60M and 100M, while state becomes the candidate bottleneck from 150M onward under the unrestricted isoelastic benchmark — driven by state’s combination of the strongest repricing effect and the highest elasticity. We also note that the high-capacity state equilibria require activity several times above the historical anchor, so both the exact base fees and the binding branch remain conditional on the unobserved tail of state demand.
These results set up the floor for studying the equilibrium of a multidimensional fee market under EIP-7999, where execution, data, and state each have separate base fees, target, and limit.
Measurement limitation
The main empirical limitation is state measurement. Execution and historical data gas come from protocol accounting. State creation is inferred from a calibrated proxy and then translated into EIP-8037 metered gas. Bias in that proxy can affect the state anchor, the recovered state elasticity, and every equilibrium in which the state branch binds.
The event-window range shows sensitivity to timing around the two clean gas-limit changes. All four rows reuse those events. The 75-day elasticity calculation also combines a 31-day first-event pre-period with longer post-periods, making it more exposed to slow-moving changes.
Appendix: data sources and construction
Elasticity estimation
| Source | Use |
|---|---|
| Xatu canonical block and transaction tables | Gas limits, base fees, total gas, transaction counts, and calldata accounting |
| Xatu storage, contract, balance, nonce, and address-appearance tables | Scalable state-creation proxy |
| Daily calibrated accounting panel | Mutually exclusive execution, data, and state quantities used in the event study |
Execution multiplier
| Source | Use |
|---|---|
Xatu default.canonical_execution_block |
Canonical block numbers, dates, and block gas totals |
Xatu default.canonical_execution_transaction_structlog_agg |
Opcode gas, cold-access counts, SSTORE gas, and transaction refund counters |
Xatu default.canonical_execution_traces |
Internal contract creation and positive-value call paths |
Xatu default.execution_transaction and default.canonical_execution_transaction |
Transaction paths and receipt gas used |
Xatu default.canonical_execution_storage_diffs |
Final storage changes used in the EIP-8038 refund reconstruction |
| Erigon RPC sample | Complete access lists and authorization-write counts |
Data multiplier
| Source | Use |
|---|---|
Xatu default.canonical_execution_transaction |
Receipt gas and zero/nonzero calldata counts for historical gas and the EIP-7976 calculation |
Xatu default.canonical_execution_block |
Canonical dates and block coverage |
| Erigon RPC sample | Access-list address and storage-key counts; EIP-7981 uses 20N_{\mathrm{address}}+32N_{\mathrm{key}} content bytes |
State multiplier
| Source | Use |
|---|---|
Xatu default.canonical_execution_storage_diffs |
Newly created storage slots |
Xatu default.canonical_execution_contracts |
Contract accounts and code bytes |
Xatu default.canonical_execution_balance_diffs and default.canonical_execution_nonce_diffs |
New-account candidates |
Xatu default.canonical_execution_address_appearances |
First-seen account filter |
Xatu default.canonical_execution_block |
Dates and block coverage |
| Erigon RPC sample | Account-proxy correction and delegation indicators |