A comparison of nine blockchains on transaction costs, application revenue and network fee capture, examining the trade-off between cheaper computation and the economic value retained by infrastructure.
At the Avalanche Foundation, part of our economics research agenda is measuring how value moves through the Avalanche ecosystem: what applications earn on the chain, what the chain itself collects, and how both change over time. This piece in the Economics of Blockchain series compares Avalanche C-Chain, Arbitrum, Base, BSC, Ethereum, Optimism, Polygon, Solana and Tron on transaction costs, application revenue and the share captured through network fees. For the C-Chain, that means two questions: what it costs to use, and how much of the value created on it the protocol captures.
Blockchain networks have spent years competing to make transactions cheaper. Yet lower fees create an economic trade-off: infrastructure that costs less to use may also capture a smaller share of the value created by the applications running on it. That leaves the industry with an open question: what is the right price for blockspace? Priced too high, activity may move elsewhere. Priced too low, a network may collect too little of the value it supports to remain economically sustainable over the long term.
For Avalanche, the question is now a practical one. The Helicon upgrade shipped on September 22, 2026, introducing ACP-283, which for the first time allows validators to adjust the minimum gas price continuously. Where to set it depends on how gas demand responds to price, which is something the past alone can’t answer. That makes Avalanche a detailed case study of changing fee mechanisms and the open question of demand elasticity, trade-offs that face the wider industry.
Our working hypothesis is that low transaction costs and low network-level fee capture are closely linked. Pricing computation cheaply generally means the network collects a smaller share of the fees users pay to applications. That is not, by itself, a measure of success or failure, because the size and composition of application activity matter just as much. Neither transaction prices nor fee revenue alone can establish the economic performance of a network.
Before adjusting the floor, it was important to establish where Avalanche actually stood relative to its peers on cost and on capture. We compared the nine chains on transaction fees and fee capture before Helicon, and tracked how both changed over time.
Key Findings
Average transaction costs fell on all nine networks between July 2024 and August 2026.
The cheapest network depends on the activity measured; a simple transfer, DEX trade and fixed amount of computation produce different rankings.
The ratio of transaction fees to gross application revenue ranges from 3.29% on Arbitrum to 87.81% on Tron, partly reflecting very different activity mixes.
Four networks show a clear downward trend in that ratio over the most recent 12 months examined; none shows a clear upward trend.
The right fee level remains an empirical question: higher prices may raise revenue per unit but reduce demand.
Average transaction costs fell on all nine networks between July 2024 and August 2026.
The cheapest network depends on the activity measured; a simple transfer, DEX trade and fixed amount of computation produce different rankings.
The ratio of transaction fees to gross application revenue ranges from 3.29% on Arbitrum to 87.81% on Tron, partly reflecting very different activity mixes.
Four networks show a clear downward trend in that ratio over the most recent 12 months examined; none shows a clear upward trend.
The right fee level remains an empirical question: higher prices may raise revenue per unit but reduce demand.
The Industry Question: Pricing, Demand and Network Revenue
Since genesis, transaction fees have been the protocol’s only revenue source. Every fee paid on the Avalanche C-Chain is burned, both the base fee and the tip. Validators are paid in newly issued AVAX, not from transaction fees.
Under ACP-283, validators vote on the floor continuously. The network converges on the stake-weighted median, and the floor can double or halve in about an hour. That makes the minimum gas price a lever on how much value the chain collects through transaction fees.
Finding a more economically sensible level means measuring how gas demand responds at different prices (demand elasticity), which requires deliberate experiments, modelling and estimation.
How the C-Chain Sets a Price
Every user-initiated state change on the C-Chain pays for gas. The price has two parts:
Base fee: set by the protocol from recent demand. It rises when blocks fill and falls when they empty. Users don't choose it.
Priority fee: a tip on top, paid for a better position in the block.
Base fee: set by the protocol from recent demand. It rises when blocks fill and falls when they empty. Users don't choose it.
Priority fee: a tip on top, paid for a better position in the block.
A transaction pays (base fee + tip) × gas used. So the bill has a price the chain sets and a quantity the transaction determines.
Two things set the C-Chain apart:
The whole fee is burned. On Ethereum, the base fee is burned and the tip goes to the block proposer. On Avalanche, both are burned, and validators are paid through staking rewards instead.
The tip now makes up most of the fee. In recent periods it has been about 90% of what users pay, up from 22% before the Etna upgrade. That limits how much the protocol's own fee settings can move the total. Most priority fee revenue comes from a small group of actors who pay tips to extract MEV.
The whole fee is burned. On Ethereum, the base fee is burned and the tip goes to the block proposer. On Avalanche, both are burned, and validators are paid through staking rewards instead.
The tip now makes up most of the fee. In recent periods it has been about 90% of what users pay, up from 22% before the Etna upgrade. That limits how much the protocol's own fee settings can move the total. Most priority fee revenue comes from a small group of actors who pay tips to extract MEV.
01 - Weekly Fees by Part
The base fee also has a floor: the minimum gas price. When demand is low, the floor is the price. For most of the C-Chain's history, demand was low enough that this was the norm, not an edge case.
The Upgrades That Reset the Price
02 - Upgrades that Reset Price
ACP-125 (Etna) changed a constant: same algorithm, lower floor.
ACP-176 (Fortuna) changed the algorithm. It replaced the windowed EIP-1559 scheme with a continuous exponential accumulator and a votable gas target. It also cut the floor to one wei, a billionth of a nAVAX, which on a log chart looks the same as no floor at all.
ACP-226 (Granite) appears on the charts because it falls in the window. It introduced dynamic minimum block times and did not change the minimum gas price.
ACP-125 (Etna) changed a constant: same algorithm, lower floor.
ACP-176 (Fortuna) changed the algorithm. It replaced the windowed EIP-1559 scheme with a continuous exponential accumulator and a votable gas target. It also cut the floor to one wei, a billionth of a nAVAX, which on a log chart looks the same as no floor at all.
ACP-226 (Granite) appears on the charts because it falls in the window. It introduced dynamic minimum block times and did not change the minimum gas price.
Each of these was a constant that needed a coordinated upgrade to change. ACP-283 ends that pattern.
03 - C-chain base fee per unit of gas against the protocol minimum
Figure 1. Daily base fee, transaction-weighted (median, interquartile range and mean), against the protocol minimum gas price, September 2024 to August 31, 2026, log scale. Transaction-weighted means each transaction counts as one observation at its block's base fee. Dotted lines mark upgrade activations.
The chart shows three phases: a floor that binds, then a lower floor that binds, then no floor. From September 2024 until Etna, the median base fee is exactly 25.00 nAVAX. That is not a coincidence or rounding. It is the floor, and the market almost never priced above it. After Etna, the median sits at 1.01 nAVAX, pinned to the new, lower floor. A price that tracks its own floor that closely isn't discovering anything.
Only after Fortuna does the base fee behave like a market price. It moves with demand, down to a median of 0.072 nAVAX over the recent period. It didn't fall because demand fell. Daily transactions rose from 0.20 million before Etna to 2.23 million now, about 11x more, while the price per unit of gas fell about 348x. The interquartile range shows the same thing. It stays flat while the floor binds, because almost every transaction pays the same price, and widens only once the floor stops binding.
What a Transaction Actually Costs
Gas price isn't the bill. A user pays the gas price times the gas used, and gas used varies far more between transactions than the price does on any given day.
04 - Total fee paid per transaction cumulative distribution
Figure 2. Cumulative distribution of total fee per transaction, by time period, log scale. The periods are nested and all end August 31, 2026. Post-Fortuna only: before Fortuna the base fee was pinned to a fixed floor, so the distribution would describe the floor, not market pricing.
Over the data from August 2026:
Median transaction: 4,694 nAVAX (0.0000047 AVAX)
90th percentile: 31,413 nAVAX
99th percentile: 2,472,284 nAVAX, about 527x the median
Median transaction: 4,694 nAVAX (0.0000047 AVAX)
90th percentile: 31,413 nAVAX
99th percentile: 2,472,284 nAVAX, about 527x the median
The change is in the tail, not the middle. Compared with the full post-Fortuna period, the 90th percentile fell 5.9x and the 99th fell 2.5x. The median didn't fall by the same amount, and for part of the period it rose. Expensive transactions got much cheaper; typical ones were already cheap.
Keep this in mind for the cross-chain comparison. The mean fee over the same 30 days was 232,257 nAVAX, about 49x the median. Fee distributions are heavily skewed, and the cross-chain data in the next section has no percentiles for any chain. So every figure there is a mean, well above what a typical transaction costs. The rankings are still meaningful, but the levels don't reflect a typical user's cost.
Comparing Transaction Costs Across Nine Chains
Throughout this section, the measure is cost per fee-paying transaction: the mean USD fee across transactions that actually paid a fee. This is not the same as a fee burden per transaction, which averages across every transaction, including those that paid nothing. On chains where most transactions are paid with staked resources rather than cash, the two can differ several times over.
The Cost of Doing Something
05 - Mean USD Cost by task and chain
Figure 3. Mean USD cost of four common actions by chain, transaction-weighted, January to August 2026, log scale. A blank bar is missing data, not zero: Tron and Solana publish no ERC-20 transfer fee, and Tron publishes no native transfer fee.
In the sample period, Avalanche is the cheapest chain in the set for an ERC-20 transfer, at $0.000649. Ethereum costs $0.1010, about 156x more. Across all transaction types, Avalanche averages $0.001890. Rankings differ by task, which is why we show four. For DEX trades, Avalanche is fifth cheapest of nine at $0.0254, behind Optimism, Arbitrum, Solana and Polygon.
How the Comparison Changed
None of Avalanche's fee changes fall within Figure 3's window. So the next two charts start in July 2024, which is early enough to include all three upgrades and late enough that all nine chains have data for every month.
06 - Cost rank among the nine chains
Figure 4. Rank by mean cost per fee-paying transaction (1 = cheapest), monthly, from July 2024. Dotted lines mark Avalanche's fee changes: Etna (December 2024), Fortuna (April 2025), Granite (November 2025).
07 - Mean cost per fee-paying transaction
Figure 5. Mean USD cost per fee-paying transaction by chain, monthly, from July 2024, log scale. Same chain colors and upgrade markers as Figure 4.
Avalanche ranked seventh cheapest of nine in July 2024, first in July 2026 and second in August 2026. The climb follows the fee mechanism changes above.
Rank alone can't show whether Avalanche got cheaper or its peers got more expensive. The levels can. Avalanche fell from $0.1461 to $0.001601 per transaction, about 91x. Ethereum fell from $2.67 to $0.1539, about 17x. Every chain got cheaper over the period, so a falling line doesn't by itself show improvement. Rank is what separates Avalanche's change from the industry-wide trend.
The Price of Computation
Average cost per transaction mixes two things: what a chain charges for computation, and what its users happen to be doing. A chain that carries mostly simple transfers looks cheap even at the same gas price. Pricing a fixed amount of work removes that effect.
08 - USD price of a billion units of EVM gas
Figure 6. USD cost of one billion units of gas, August 2026, log scale, cheapest to most expensive. Solana and Tron are excluded because their compute units measure different work.
On this measure, Avalanche is the second cheapest of the seven chains with usable gas data, at $7.91 per billion units. Optimism is cheapest at $2.65. Ethereum is $1,396, about 177x Avalanche. This is the cleanest of the three comparisons, because it is the only one a chain can't win just by carrying simpler transactions.
What the data does not support is a blanket claim. Avalanche is not the cheapest for everything, and on every chain in the set the means shown here are well above the median transaction.
How Much Value Do Blockchain Networks Capture?
Cost is what a user pays the chain. Capture is how much of everything a user pays ends up with the chain, and most of it doesn't. Applications charge their own fees on top, such as swap fees, borrowing interest and performance fees.
We call the chain's share transaction fees. We call the applications' share protocol revenue: total fees users pay, before the share owed to liquidity providers and depositors is taken out.
The ratio between the two works like a tax rate. It has a clear limit: a charge per transaction isn't a charge on economic value. But it ranks chains usefully and shows where chain-level capture is structurally limited.
09 - Monthly Tx Fees / Protocol Revenue by Chain
Fees as a Share of Application Revenue
Figure 7. Monthly transaction fees divided by protocol revenue, by chain, log scale, January 2025 to August 2026. Avalanche highlighted; other chains muted.
Avalanche is in this low-capture group, in single digits and falling, at 3.39% over the period. Only Arbitrum is lower, at 3.29%. Ethereum is at 6.64%, Base at 7.02% and BSC at 23.30%.
Tron, at 87.81%, says more about the measure than about the chain. Its ratio is about 4x the next highest, and the cause is activity mix. Tron's volume is mostly stablecoin transfers, which pay meaningful gas but generate almost no application fees. Any chain with the same mix would score near the top, which is why the ratio has to be read alongside the absolute figures.
The Size Behind the Ratio
10 - Chain - Revenue - Fees - Fee/Rev Ratio
11 - Tx Fees vs Protocol Revenue by Chain
Figure 8. Cumulative transaction fees against protocol revenue by chain, January 2025 to August 2026, log-log scale. The diagonal lines mark constant ratios: a chain below the 10% line collects less than 10% of user-paid value as gas.
In absolute terms, Avalanche is one of the smaller economies in the set. Over the period it generated about $282.2M in protocol revenue and $9.6M in transaction fees, the second smallest of the nine on both. Solana and Ethereum each generated several billion dollars in revenue, roughly 37x Avalanche.
That changes how to read the low ratio. Avalanche's 3.39% isn't low because revenue is high. Revenue is modest, and fees are very low relative to it. The chain's fee income is about 87x below the largest chain in the set. Raising capture to the level of Ethereum and Base would roughly double it (about 2.0x) at today's revenue, and still leave it about 43x below. Growing what applications earn on the chain GCP (Gross Chain Product) matters at least as much as raising the chain's share.
The Trend
We tested the downward drift in Figure 7 instead of judging it by eye.
Falling (four of nine): Tron, Optimism, Avalanche C-Chain and Arbitrum. Tron is the steepest, at -12.87 percentage points per year.
Rising: no chain shows a clear upward trend.
No reliable direction (five): BSC, Base, Ethereum, Polygon and Solana. Base and Polygon end the period higher than they started but still fit a slight downward slope.
Falling (four of nine): Tron, Optimism, Avalanche C-Chain and Arbitrum. Tron is the steepest, at -12.87 percentage points per year.
Rising: no chain shows a clear upward trend.
No reliable direction (five): BSC, Base, Ethereum, Polygon and Solana. Base and Polygon end the period higher than they started but still fit a slight downward slope.
For Avalanche, the ratio fell from 5.79% in January 2025 to 1.13% in the latest month. The breakdown matters more than the slope. Monthly protocol revenue fell about 30% over the period, while monthly transaction fees fell about 74%, from a peak of $1.7M in October 2025. The share fell because the chain's own take shrank faster than application activity, not because applications earned more.
The Pricing Trade-off: What the Evidence Can and Cannot Tell Us
Low cost and low capture are one finding, not two. A chain that prices computation near the bottom of its peer group will collect a small share of the value passing through it, almost automatically, whatever its activity mix. Avalanche's 3.39% isn't a mistake in the fee schedule. It's what a low fee schedule produces. The two moved together: the gas price fell about 348x, daily transactions grew about 11x, and the chain's share has been falling since.
ACP-283 made the fee schedule adjustable without an upgrade for the first time, so it's worth being clear about what kind of tool it is. A floor only applies when demand is low. When demand is high, the fee algorithm takes over. So the floor prices the quiet periods, not all activity, which limits both the harm a bad setting can do and the revenue a good one can raise. Figure 1 shows the scale. In the 30 days before Helicon, the median base fee was 0.072 nAVAX. A floor well below that would rarely apply; one near it would apply often.
Where to set the floor is an empirical question this data can't answer. Fee revenue is price times quantity: a higher floor earns more per unit of gas but may reduce how much gas is used. Nothing here shows which effect is bigger, because the floor has never changed on its own. It has always changed inside an upgrade that changed other things too, while prices were falling on every chain in the set. Answering it takes experiments: set the floor at several levels, hold each long enough for demand to settle, and vary it enough to separate its effect from the wider trend.
We're tracking the impact closely, especially which users and transactions are affected and how. We'll share detailed analysis as more data comes in.
Industry-wide implications
The comparison does not identify a universally cheapest blockchain or a single optimal fee model. Rankings change with the task measured, and transaction-fee-to-application-revenue ratios reflect activity mix as well as network pricing. The nine-chain evidence instead points to three questions for future analysis: how comparable computation is priced, how much gross application activity networks support, and how demand changes when prices change. Historical observations alone cannot determine the optimal fee floor; controlled experiments, modelling and estimation are needed. The findings should therefore be read as a baseline for comparing blockchain economics, not as a ranking of overall network quality.
Appendix
Base fee. The part of the gas price the protocol sets from recent demand. It is burned, not paid to a validator.
Priority fee. The tip a user pays on top of the base fee to get included sooner. On the Avalanche C-Chain it is burned along with the base fee. On Ethereum it goes to the block proposer.
Minimum gas price. The protocol floor under the base fee. A block's base fee can never go below it, so when demand is low the floor is the price.
Sample History: 25 nAVAX until ACP-125 (Etna), then 1 nAVAX until ACP-176 (Fortuna), then one wei, which on a log chart looks the same as zero.
Sample Unit: nAVAX per unit of gas, where 1 nAVAX = 10^9 wei = 10^-9 AVAX.
How it works since ACP-283: validators set a preferred floor in their node configuration, and the network converges on the stake-weighted median. A validator with no preference inherits the previous block's value. The floor can double or halve in about an hour of sustained voting.
Effect at activation: ACP-283 activated at the existing value, so what changed is how the floor moves, not where it sits. The measurements here describe the regime it inherited.
Stated motivation: the proposal targets spam, not revenue. At a floor of one wei, protocols running proof-of-work on-chain stay profitable during quiet periods.
Sample History: 25 nAVAX until ACP-125 (Etna), then 1 nAVAX until ACP-176 (Fortuna), then one wei, which on a log chart looks the same as zero.
Sample Unit: nAVAX per unit of gas, where 1 nAVAX = 10^9 wei = 10^-9 AVAX.
How it works since ACP-283: validators set a preferred floor in their node configuration, and the network converges on the stake-weighted median. A validator with no preference inherits the previous block's value. The floor can double or halve in about an hour of sustained voting.
Effect at activation: ACP-283 activated at the existing value, so what changed is how the floor moves, not where it sits. The measurements here describe the regime it inherited.
Stated motivation: the proposal targets spam, not revenue. At a floor of one wei, protocols running proof-of-work on-chain stay profitable during quiet periods.
Transaction-weighted statistics. Each transaction in a day counts as one observation at its block's base fee, so a block with 400 transactions counts 400 times. This measures what users experienced rather than what blocks did. The two differ because transactions cluster in busier blocks, which are more expensive.
Cost per fee-paying transaction. The mean USD fee across transactions that actually paid one, excluding system and zero-fee transactions. In other words, cost per payer. It is not the same as a fee burden per transaction, which divides total fees by every transaction, including those that paid nothing. Where most transactions are paid with staked resources, the two can differ several times over. Daily means are combined as a transaction-weighted mean, so a quiet day counts for less than a busy one.
Gas quantum (USD per billion units). Total fees paid divided by total gas used, times one billion, for a single month. Pricing a fixed amount of gas isolates gas price times native token price, which is what a chain actually sets, without mixing in transaction type. One billion units of Avalanche gas buys about 47,600 native transfers at the 21,000-gas minimum cost.
Protocol revenue. Total fees users pay to applications, before the share owed to liquidity providers, depositors and other supply-side participants is taken out. This is the gross, user-paid measure.
Transaction fees. Gas users pay on a chain's base layer: base fee plus priority fee. On the Avalanche C-Chain, both are burned.
Fee/revenue ratio. Monthly transaction fees divided by monthly protocol revenue. It approximates the share of user-paid value the chain collects, as opposed to the applications and participants above it. It is calculated monthly to smooth out daily price noise, so the monthly figure depends on how the AVAX/USD price moved within the month.
Retained revenue. The share of user-paid application fees that the protocol treasury, governance or token holders keep after supply-side participants are paid. It is the closer equivalent of profit. Over this period it was $87.1M on Avalanche, about 31% of gross revenue. Against that figure, transaction fees are 10.98%.
