Ethereum and Layer 2s: How the Rollups Work, What Their Numbers Measure, and Why Two Trackers Disagree by Four Times

Ethereum is the largest single venue in crypto on the one measure that can be independently recounted, and the second-largest thing in crypto is the collection of chains built on top of it. Neither of those sentences tells you how big either one is. The numbers published for Ethereum and for its Layer 2s measure different quantities, and the gap between them is not a rounding error. On 6 August 2026 the two most credible public trackers of Ethereum rollups reported figures for the same fourteen chains that differed by a factor of 3.91. Both were right. This page sets out what Ethereum and its rollups actually are, what each published figure counts, which ones you can check yourself, and the one ranking in this sector that we think is worth more than all the dollar totals put together.

What Ethereum actually is

Ethereum is a public ledger with a general-purpose computer attached. Bitcoin’s ledger records who holds what; Ethereum’s records who holds what and what programs are running, what state they are in, and what happens next time somebody pokes them. Those programs are called contracts, they are deployed by anybody who pays the fee, and once deployed they execute whenever a transaction calls them. Nobody has to approve the deployment and nobody can stop the execution.

Two consequences follow, and the whole sector is downstream of them. The first is that most of what people mean by “crypto” other than the coins themselves — lending, exchange, stablecoins, staking, tokenized funds — is contracts on Ethereum or on something built to look like it. The second is that Ethereum’s throughput is a hard ceiling shared by everyone using it. Every transaction competes for the same space in the same blocks, so when demand rises the price of inclusion rises for all of it. That constraint is the entire reason Layer 2s exist.

Since September 2022 Ethereum has run on proof of stake. Validators post collateral in ether, propose and attest to blocks, and lose part of the collateral if they misbehave. This replaced mining, cut the network’s energy draw by orders of magnitude, and introduced a different problem: staking is a service most holders buy rather than run, which means it concentrates in the hands of whoever is best at selling it. On DefiLlama’s numbers on our pull date, 76 staking, liquid-staking and restaking protocols reported $47.79bn of Ethereum-side value, 19.6% of everything reported on the chain, and the largest single one held more than a third of that. We treat validator concentration as a separate measurement and it is named as this page’s sequel below.

What a rollup actually is, and what it is not

A rollup executes transactions somewhere else and posts the results to Ethereum. That is the whole idea. Users transact on the rollup, cheaply, because the rollup does not have to fight for Ethereum block space per transaction; it batches thousands of them, compresses the data, and writes the batch to Ethereum once. Ethereum stores the data and holds the money. The rollup does the work.

The security question is therefore not “is the rollup honest” but “what happens when it is not.” There are two answers in production, and the difference is genuine.

• Optimistic rollups assume batches are valid and give anybody a window to submit a fraud proof. If nobody objects, the batch stands. This works if the objection mechanism is real and someone is watching.
• Zero-knowledge rollups submit a cryptographic proof that the batch was executed correctly. Ethereum verifies the proof. Nobody has to be watching, because nothing invalid can be accepted in the first place.

On paper zk rollups are the stronger design. In practice, on our pull date, the nine optimistic rollups on L2Beat’s list held $23.66bn, 68.2% of all value secured, while the thirteen zk rollups held $3.23bn, or 9.3%. The better architecture holds a seventh as much money as the weaker one. That is worth sitting with, because it tells you what the market is actually pricing, and it is not cryptographic guarantees.

Two further categories exist and both matter for a reason people gloss over. Validiums use validity proofs but keep transaction data off Ethereum; optimiums use fraud proofs and keep data off Ethereum. In both cases, if the operator withholds the data, users cannot reconstruct their own balances to exit. Between them they held $0.46bn, 1.4% of the total, across seven projects. And of the 103 projects on L2Beat’s list carrying any value at all, 74 are not rollups by any of these definitions — they hold $7.34bn, 21.2% of the total, and are frequently described in their own marketing as Layer 2s. The word has stopped doing work.

The stages, and why they are the only ranking worth reading

L2Beat classifies every rollup into one of three stages according to how much the operator can still override. It is the most useful classification in the sector and almost nobody quotes it, because it is unflattering.

• Stage 0. The operator can do essentially what it likes. There may be a security council, an upgrade key, a pause button, or all three. Your exit depends on the operator continuing to behave.
• Stage 1. Fraud or validity proofs are live and users can exit without the operator’s cooperation, but a council retains override powers for emergencies.
• Stage 2. The proof system is the final authority. No council can override it. This is what a rollup is supposed to be.

The measurement: how much money sits behind a system nobody can override

We pulled L2Beat’s scaling summary on 6 August 2026 and grouped every project by stage, weighting by value secured. The answer is the single most important number on this page.

1. Stage 2 holds $511,551. Four projects have reached it. Across all of them combined they secure just over half a million dollars, which is 0.0015% of the $34.69bn tracked. Facet holds $509,426 of that. Cartesi’s PRT Honeypot v2 holds $1,560. Aztec Network holds $565. Ethscriptions holds nothing at all.
2. Stage 1 holds $23.85bn, or 68.8%, across six projects: Base at $11.75bn, Arbitrum One at $10.09bn, OP Mainnet at $1.40bn, Starknet at $0.37bn, Ink at $0.16bn and Unichain at $0.07bn.
3. Stage 0 holds $5.58bn, or 16.1%, across 74 projects — the long tail, where the operator retains full control.
4. Projects to which the staging framework does not apply — twenty of them, mostly validity-proof sidechains and things marketed as Layer 2s that are not — hold $5.25bn, or 15.1%.

Read that again in plain terms. After roughly six years of rollup development, the amount of money sitting on a rollup that its own operator cannot override is 0.0015% of the money sitting on rollups. The remainder sits behind a governance body that can, in the right circumstances, change the rules. That is not a scandal and it is not fraud; every one of the large projects publishes its stage openly and L2Beat exists precisely to make it legible. It is, however, flatly incompatible with how this sector describes itself. “Trustless scaling” is a roadmap, not a description of where the money is.

We would go further. Stage is the only Layer 2 statistic that is both meaningful and hard to game, because it is assessed against published contract permissions rather than self-reported. If you are choosing between rollups on any basis other than stage and fees, you are choosing on marketing.

Where the value actually sits

Concentration in this sector is severe and getting less discussed as it gets worse. Of the 103 projects with any value secured on our pull date, the top three hold 73.7%, the top five hold 81.4%, and the top ten hold 91.6%. The Herfindahl index across all 103 reads 2,168, which in an antitrust setting would describe a highly concentrated market. Ninety-three projects share the last 8.4% between them.

• Base — $11.75bn, 33.9%
• Arbitrum One — $10.09bn, 29.1%
• Polygon PoS — $3.73bn, 10.8%
• OP Mainnet — $1.40bn, 4.0%
• Mantle — $1.25bn, 3.6%
• Robinhood Chain — $0.98bn, 2.8%
• Hyperliquid — $0.98bn, 2.8%
• Lighter — $0.88bn, 2.5%
• Starknet — $0.37bn, 1.1%
• Linea — $0.35bn, 1.0%

Two observations we will stand behind. First, the top two are both optimistic rollups built by companies with large consumer businesses attached — one of them a listed exchange — which suggests distribution is beating technology decisively. Second, fourteen of the 103 entries are Layer 3s, chains settling to another rollup rather than to Ethereum, and they hold $1.06bn, 3.0% of the total. The recursive-scaling thesis is, so far, a rounding error.

What “Layer 2 TVL” is actually made of

The headline figure for rollups is not a measure of activity, adoption or usefulness. It is a measure of assets sitting on the chain, and L2Beat breaks it into three parts that behave very differently.

• Canonical — $12.94bn, 37.3%. Assets bridged in through the rollup’s official bridge. The originals are locked on Ethereum; what you hold on the rollup is a claim on them.
• External — $12.45bn, 35.9%. Assets bridged in by a third party. The claim runs against that bridge, not against Ethereum, and the security of the rollup is irrelevant to whether the bridge pays.
• Native — $9.30bn, 26.8%. Assets issued on the rollup itself, which mostly means the rollup’s own token.

So slightly over a quarter of “value secured” is not secured by Ethereum in any meaningful sense, and a further third is only as good as a third-party bridge. When a chain announces that it secures several billion dollars, the honest reading is that a plurality of it is a promise made by somebody other than Ethereum.

The asset mix is more surprising and, we think, underreported. Stablecoins account for $16.02bn, ether for $5.77bn, and bitcoin-denominated assets for $4.07bn. There are nearly three dollars of dollar-substitutes on Ethereum’s own rollups for every dollar of ether. Ethereum built the scaling layer; the dollar is what people are using it to move. Anyone modeling Layer 2 growth as a driver of ether demand should reconcile that against these figures first.

The measurement: two reputable trackers, one set of chains, four times apart

Here is the part that should change how you read every Layer 2 headline. We took the fourteen chains that both L2Beat and DefiLlama track under recognizably the same name, and put their published figures side by side. Both were pulled on 6 August 2026, minutes apart.

Chain	L2Beat	DefiLlama	Ratio
Base	$11.75bn	$4.64bn	2.5×
Arbitrum One	$10.09bn	$1.20bn	8.4×
Polygon PoS	$3.73bn	$0.80bn	4.7×
OP Mainnet	$1.40bn	$0.31bn	4.5×
Mantle	$1.25bn	$0.07bn	17.8×
Robinhood Chain	$0.98bn	$0.43bn	2.3×
Starknet	$0.37bn	$0.17bn	2.2×
Linea	$0.35bn	$0.03bn	12.6×
World Chain	$0.31bn	$0.03bn	11.5×
ZKsync Era	$0.20bn	$0.01bn	14.3×
Ink	$0.16bn	$0.12bn	1.4×
Unichain	$0.07bn	$0.03bn	2.5×
Blast	$0.07bn	$0.03bn	2.3×
Scroll	$0.04bn	$0.01bn	3.7×
All fourteen	$30.78bn	$7.87bn	3.91×

Three point nine one, and the per-chain spread runs from 1.4× to 17.8×. If you have read a sentence beginning “Arbitrum holds” and it was followed by a number, you have no way of knowing which of two figures eight times apart the writer meant, and a fair chance the writer did not know either.

Why they disagree, and why both are right

This is not a case of one source being wrong. They are answering different questions, and each answers its own question well.

L2Beat measures total value secured: everything sitting on the chain, whether it is doing anything or not. Bridge in a million dollars of stablecoins and leave them in your wallet and L2Beat counts them, because the question it is asking is how much users stand to lose if the chain’s security assumptions fail. DefiLlama measures total value locked: assets deposited into DeFi protocols on that chain. Stablecoins sitting in a wallet count for nothing, because the question it is asking is how much capital the applications on the chain have attracted.

Both are legitimate. Neither is “the size of the chain.” And the ratio between them is itself informative: the wider the gap, the more of a chain’s balance is parked rather than deployed. Mantle at 17.8× and ZKsync Era at 14.3× are chains where almost nothing is being used; Ink at 1.4× is a chain where nearly everything on it is deployed into something. That ratio is a better measure of whether a rollup has an economy than either number alone, and we have not seen anyone publish it. We intend to keep publishing it.

The rule we would apply, and would like to see applied to us: if you quote a Layer 2’s size, name the tracker and name the measure in the same sentence. “$10.09bn secured, L2Beat” and “$1.20bn locked, DefiLlama” are both fine. “$10.09bn in Arbitrum” is not a statement about anything.

What we could not reconcile

One thing we could not make add up, disclosed because we would want to know it if somebody else had published this. L2Beat’s summary endpoint returns both an aggregate time series and a per-project breakdown. Summing the per-project figures on our pull gives $34.69bn. The aggregate series on the same response, at its most recent point, gives $33.43bn. That is a gap of $1.26bn, or 3.6%, inside one response from one source.

We suspect the series covers a narrower set of projects than the table, or is synced to a slightly earlier timestamp. We do not know, and until we do, treat every total on this page as carrying a 3–4% band. Where we quote $34.69bn we mean the sum of the per-project figures, and we have said so each time. If a number on this page ever gets quoted back to us to two decimal places, that is our fault for printing it that way.

Along the same lines: five of the 108 projects L2Beat lists carry no value at all, and four are flagged as under review, meaning their classification may change. We excluded the zeroes from percentage calculations and left the under-review projects in, which is a judgment call that moves nothing material.

Ethereum mainnet, and the shape of the last five years

The rollups exist to relieve Ethereum, so it is worth being blunabout what has happened to Ethereum itself. On DefiLlama’s chain numbers on our pull date, Ethereum held $41.53bn, which is 55.1% of the $75.38bn tracked across all 461 chains. It remains, by a distance, the largest venue in the sector: BSC is second at 6.5%, Tron third at 6.4%, Solana fourth at 6.3%, and Base — a chain built on Ethereum — fifth at 6.2%.

The trend is less comfortable. Ethereum’s chain total was $81.48bn a year earlier, so it is down 49.1% over twelve months. Against its peak of $107.45bn on 9 November 2021 it is down 61.4%. It is now roughly level with where it stood in August 2024, at $42.77bn. Five years of scaling work, a consensus change and an entire rollup ecosystem have not moved the total value on the base layer past where it was two years ago.

We are wary of over-reading that, and here is the honest complication. Some of the decline is capital moving to rollups, which is what the rollups were built to do — so the same event reads as failure on the mainnet line and success on the Layer 2 line. What we can say is that the two numbers do not add up to a growth story: $41.53bn on the base layer and $34.69bn secured on rollups today, against $81.48bn on the base layer alone a year ago. Even crediting every dollar on every rollup as migrated rather than lost, the combined figure is below where the base layer was on its own.

One more, from L2Beat’s own thirty-day series. Between 7 July and 6 August 2026 total value secured fell 3.1%, from $34.50bn to $33.43bn, while the ether price rose 6.0%, from $1,797.47 to $1,905.64. Dollar totals that fall while the denominating asset rises mean assets left. That is a small window and we would not build a thesis on it, but it is the opposite of the direction the price implies, and it is the sort of thing that gets lost when only the dollar total is quoted.

What it costs, and what it does not cost

The economics of a rollup are simple enough to state. You pay two things: a small fee to the rollup for execution, and a share of the rollup’s cost of posting data to Ethereum. Since the blob-carrying transaction format arrived in March 2024, that second component has been paid in a separate fee market from ordinary Ethereum gas, which is why rollup fees fell by an order of magnitude and stayed there.

We are deliberately not printing a table of current fees per chain. Rollup fees move with Ethereum blob demand hour to hour, a figure pulled today would be wrong tomorrow, and a standing reference that quietly goes stale is worse than one that declines to guess. What we will state as durable: transacting on a major rollup costs cents rather than dollars, transacting on Ethereum mainnet costs dollars, and the gap is structural rather than promotional. Check the current number yourself before you move size — and check the withdrawal cost too, which is charged on Ethereum and is the expensive half of the round trip.

The other cost is time. Withdrawing from an optimistic rollup through its official bridge means waiting out the fraud-proof window, conventionally seven days. Third-party bridges will front you the money for a fee, which is a real service and also the reason 35.9% of the value on these chains is a claim against a bridge rather than against Ethereum.

How to read a rollup before you put money on it

Six checks, in the order we would run them. All six are answerable from public sources in about ten minutes, which is the point.

1. Stage. Look it up on L2Beat before anything else. Stage 0 means the operator can override the system. That may be an acceptable risk for a small balance and a short stay; it is not the same product as Stage 1.
2. Where the data goes. If transaction data is posted to Ethereum, you can reconstruct your balance and exit without the operator’s help. If it is posted elsewhere, you cannot. This distinction is what separates a rollup from a validium, and it is disclosed.
3. Who holds the upgrade keys, and how fast they act. A security council with a multi-day timelock is a different animal from one that can upgrade contracts instantly. Both exist. The delay is in the contracts and is checkable.
4. Whether the proof system is actually running. Several chains describe themselves by the design they intend to ship. Ask whether fraud or validity proofs are live in production today, not whether they are on a roadmap.
5. The composition of the balance. If most of the value is the chain’s own token, the headline total will collapse with the token price and tells you little about adoption.
6. The locked-to-secured ratio. Divide the DefiLlama figure by the L2Beat figure. A chain where nearly all the value is deployed into applications has an economy. A chain at a tenth has a balance sheet and a marketing budget.

Where rollups actually fail

The failure modes here are not the ones people worry about, and it is worth being specific because vagueness is how this sector avoids accountability.

• Sequencer downtime. Almost every rollup in production runs a single sequencer. When it stops, the chain stops. Funds are not lost, but you cannot transact and you cannot exit quickly, which during a price move is the same as a loss.
• Upgrade risk. At Stage 0 and Stage 1, a contract upgrade can in principle change the rules under which your assets are held. This is the risk the staging framework was invented to describe, and it applies to 84.9% of the value on this list.
• Bridge failure. Historically the largest losses in this part of crypto have come from bridges, not from rollups. A third-party bridge holding the originals is a single counterparty; if it is drained, your balance on the rollup is a claim on nothing.
• Data withholding. On validiums and optimiums, an operator that withholds data can prevent users from proving what they own. This is a live design risk, not a theoretical one.
• Token-denominated collapse. Where native tokens are a large share of the balance, a fall in the token both shrinks the headline number and removes the collateral behind whatever was borrowed against it. The two happen together, which is the definition of a bad day.

Notice what is absent from that list: an optimistic rollup being defrauded through its proof system. As far as we can establish, that has not happened in production at scale. The designs are working. It is everything around the designs — single sequencers, upgrade keys, bridges — that has cost people money.

What the data does not show

We would rather list this than have a reader infer more from the numbers above than they can carry.

• Nothing here measures usage. Value secured and value locked are both balances. A chain can hold billions and process almost nothing, and several on this list do.
• Nothing here measures users. Addresses are free and one person can hold thousands. We have not published an address count and will not treat one as a user count.
• The stage framework measures permissions, not competence. A Stage 1 chain with excellent operators may in practice be safer than a Stage 2 chain with a subtle bug. Stage tells you who can intervene, not who will, and not whether the code is correct.
• We did not verify a single contract ourselves. Every stage classification on this page is L2Beat’s reading of the deployed contracts. It is the best public assessment we know of and it is still a third party’s judgment, restated.
• The 3.91× gap is not a measure of anything being hidden. It is two definitions producing two answers. We are not alleging misreporting by either source.
• Twelve-month comparisons on dollar totals are contaminated by price. Ethereum’s 49.1% fall is partly assets leaving and partly assets being worth less. We have not decomposed it, so we have not claimed which dominates.

What would change these conclusions

Four things, named in advance so that we can be held to them.

1. Stage 2 taking real money. If a single Stage 2 rollup crosses $1bn secured, the central finding on this page — that trustless scaling holds 0.0015% of the money — is obsolete and we will say so in the changelog rather than quietly editing the number.
2. A zk rollup entering the top three by value. That would mean the market has started paying for the stronger design, reversing the current position where it holds a seventh as much as the weaker one.
3. The two trackers converging on a shared definition. If L2Beat and DefiLlama publish a reconciled measure, the 3.91× finding becomes a historical note and the case for naming your tracker weakens.
4. Ethereum’s base-layer total plus rollup total exceeding the base-layer figure of a year ago. $81.48bn is the bar. Clearing it would make the migration story hold together in a way it currently does not.

And one thing that would not change our mind: a large rise in transaction counts. Transactions are cheap to manufacture on a chain that costs cents to use, and we will not treat a count of them as evidence of adoption without a cost-weighted view alongside it.

Where this sits in the rest of our coverage

This page is the standing reference for Ethereum and its rollups. It is re-measured quarterly at this URL, with the revision date and a changelog line, rather than being republished as a new post each time.

• DeFi Explained — how the applications running on these chains work, and why the published DeFi total is a third of what the protocol pages add up to.
• Crypto Exchanges Explained — how centralized venues work, what they charge, and what proof of reserves does not prove.
• Self-Custody and Wallets — holding your own keys, and how to check whether anyone is maintaining your hardware.
• Stablecoins Explained — the $16.02bn of dollar-substitutes on these rollups, and who issues them.
• Bitcoin Explained — the other half of the market, and the one measurement in crypto that anybody can recount.
• Layer 1 vs Layer 2 — the shorter introduction to the distinction, if this page is more than you need.

This page’s sequel: Ethereum’s validator set. Staking now accounts for 19.6% of everything reported on the chain and the largest liquid-staking protocol alone holds 7.4% of it. We intend to measure how concentrated the validator set actually is, using the same method we applied to Bitcoin mining pools — block attribution that anybody can recount — and to publish the failure if the attribution turns out not to be recountable, as we did with exchange volume.

Revision history

6 August 2026 — first publication. All figures pulled 6 August 2026. L2Beat scaling summary: 108 projects, 103 with non-zero value secured, $34.69bn total by summation. DefiLlama: 461 chains, $75.38bn total, Ethereum $41.53bn. Next scheduled re-measurement: November 2026.

Sources. All figures pulled 6 August 2026 and stated as of that date. Value secured, staging, category and composition data: L2Beat scaling summary endpoint. Chain totals and the comparison figures: DefiLlama chains endpoint. Ethereum history: DefiLlama historical chain TVL. Ethereum-side protocol and staking figures: DefiLlama protocols endpoint. Every endpoint above is public and unauthenticated; the totals we quote are sums we computed from the raw responses, and anyone can repeat the calculation. Comments are closed sitewide. Corrections to corrections@coinagereport.com.