What a Layer 2 is, and what risks it adds

At some point someone told you to trade "on Base, it's cheaper", and ever since you have had funds on a network nobody has explained to you. You know transactions cost cents and settle fast. You do not know who produces those blocks, what happens if that person switches off the server, or why withdrawing takes a week.

Layer 2 networks are Ethereum's answer to a problem with no elegant solution. They work, and they work well: as of August 2026 they hold tens of billions of dollars. But they introduce risks the base layer does not have, and at least one of those risks has already materialised — in an orderly, well-documented way.

This guide covers the actual mechanism, the current numbers and, above all, the Polygon zkEVM case, which is the best teaching material available on what trusting an L2 really means.

The trilemma, without the mysticism

The popular formulation says a blockchain can only optimise two of three properties: decentralisation, security and scalability. It is not a proven theorem, but it describes a real engineering tension well.

The concrete bottleneck is this. On a network where thousands of independent nodes validate every transaction, raising capacity means demanding more resources from every node. More bandwidth, more disk, more CPU. And every time you raise that bar, some home operators drop out and the ones left standing are larger entities. You have gained transactions per second and lost decentralisation.

Ethereum chose not to go down that road. Its per-block gas limit rose to 60 million with Fusaka on 3 December 2025, but that is a tuning adjustment, not a change of scale. The structural bet was different: let execution move elsewhere and keep the things only the L1 can provide, which are settlement and data availability.

What an L2 is, exactly, and what it is not

A Layer 2 is a network that executes transactions outside Ethereum but publishes to Ethereum the data needed to reconstruct its state, and settles its results there. The funds live in an L1 contract. The L2 moves balances internally and periodically writes a summary of what happened back to Ethereum.

The consequence that matters: if the team running the L2 vanishes, anyone can in principle rebuild the state from the data published on Ethereum and prove what they are owed. That is the entire value proposition.

A sidechain does not do this. Polygon PoS, the network most people wrongly call "Polygon's L2", has its own validator set and its own consensus mechanism. It is fast and cheap, and it is a perfectly legitimate chain, but its security depends on its validators, not on Ethereum. If that set fails or colludes, Ethereum gives you nothing back. Confusing the two categories is the most expensive conceptual error made in this area.

Optimistic rollups versus zk-rollups

The two families solve the same problem — convincing Ethereum that off-chain execution was correct — with opposite philosophies.

Optimistic: validity is presumed

The operator publishes the new state on Ethereum, asserting it is correct without proving it. In exchange, a seven-day dispute window opens in which any observer can submit a fraud proof pinpointing exactly which execution step is wrong. If the proof succeeds, the state is rolled back and the proposer loses their bond.

This is the model behind Arbitrum One and OP Mainnet, described in detail in Ethereum's documentation on optimistic rollups. Its advantage is compatibility: they run essentially the same bytecode as Ethereum, so smart contracts deploy with almost no changes. Its cost is time — seven days of waiting to withdraw through the native bridge.

Zk-rollups: validity is proven

The operator ships each batch with a cryptographic validity proof (a SNARK or a STARK) that the Ethereum contract verifies. If the proof checks out, the state is correct by construction. There is nothing to dispute, so there is no waiting period: the withdrawal is final as soon as the proof is verified.

The price is paid on the other side. Generating those proofs is computationally expensive, and EVM equivalence has taken years of engineering to reach. This is the model behind zkSync Era and Linea, among others.

What EIP-4844 changed

Before 13 March 2024, L2s published their data in Ethereum calldata — the same expensive space every ordinary transaction competes for. An L2's cost was dominated by what Ethereum charged it to store that data forever.

The Dencun upgrade introduced EIP-4844 and with it blobs: a separate data space with its own fee market, independent of the base fee, which nodes delete automatically after a few weeks. L2s do not need their data available forever, only long enough for someone to reconstruct state or raise a dispute. The design fit was exactly right, and the effect on prices was immediate.

Successive blob capacity expansions followed: Pectra (7 May 2025) doubled blob throughput, and the BPO1 (9 December 2025) and BPO2 (7 January 2026) adjustments raised the per-block blob target to 14 with a maximum of 21. You can see the knock-on effect in Ethereum gas costs and in those of its L2s.

The numbers, as of August 2026

TVL figures from L2BEAT as of 28 August 2026; median costs as of 24 April 2026. TVL is not a security metric: it measures how much money is inside, not how well protected it is.

For context, total DeFi TVL sat at around $99.2 billion on 28 August 2026, with Ethereum holding $49.8B and Base $5.5B inside protocols. In other words, a large share of the value bridged to L2s simply sits idle and never reaches DeFi protocols at all.

The risks nobody explains to you

The sequencer is a single machine

On almost every production L2, the party that orders transactions and produces blocks is a single operator, usually the founding team itself. It cannot steal your funds — the L1 contract does not permit that — but it can censor your transactions, reorder them in its own favour, or simply go offline.

Well-designed rollups include a forced inclusion mechanism: if the sequencer ignores you, you submit the transaction directly to the L1 contract, which compels inclusion after a set delay. Check whether your L2 has this live before assuming you can always exit.

The upgrade keys

The contract holding the funds on Ethereum is almost never immutable. There is usually a multisig or a security council able to upgrade the logic, sometimes instantly. It is a defensible decision — it allows a critical bug to be patched — but it means that whoever controls those keys can, in the limit, change the custody rules governing your money.

The bridges

The historical pattern is damning. Ronin lost around $620M on 23 March 2022 through compromised validator keys. Poly Network, around $612M on 10 August 2021 (of which the attacker returned roughly $578.6M). BSC Token Hub, around $570M on 6 October 2022. Wormhole, more than $320M on 2 February 2022 through a signature verification flaw. Nomad, more than $190M on 1 August 2022 through a faulty initialisation.

Bridges accumulate custody of assets without inheriting the security of the chain backing them. It is the most profitable attack surface in the sector.

The Polygon zkEVM case: what happens when an L2 closes

On 3 July 2026, according to Polygon's own official page, the Polygon zkEVM Mainnet Beta sequencer was sunset and the network stopped producing blocks. This was not a hack and not a collapse. It was a product decision, taken in an orderly way and announced in advance.

Even so, look closely at what it implies:

1. The network produces no blocks. No transactions, no swaps, no closing of positions. Whatever was open stayed exactly as it was.
2. A claims interface was opened to recover assets, running until 31 December 2027. That is a generous deadline, but it is still a deadline: after that point, the route closes.
3. Funds deposited into the network's smart contracts are not recoverable through it. This is the key sentence. If you held ETH in your own account, you claim it. If you had liquidity in a pool, collateral in a lending protocol, or a token issued to you by a contract, the claims interface does not cover that.

The practical conclusion is not "avoid L2s". It is this: do not leave positions inside protocols on low-activity L2s. The less volume and TVL a network has, the closer it is to an internal conversation about whether it is still worth paying for.

Worked example: withdrawing $1,000 from an optimistic rollup

You hold $1,000 on an optimistic rollup and want it back on Ethereum L1 through the native bridge, as of August 2026.

Step 1 — Initiate the withdrawal on the L2. You sign a transaction that burns or locks your funds on the L2 and records the intent to withdraw. Cost: one transaction on that network, on the order of $0.03. Your L2 balance drops to zero immediately. You still have nothing on L1.

Step 2 — Wait for the batch to be published. The sequencer groups your transaction with others and posts the new state to Ethereum. This usually takes minutes to a few hours. Cost to you: zero.

Step 3 — Submit the withdrawal proof on L1. One Ethereum transaction proving your withdrawal is included in that published state. With gas at 0.151 gwei, roughly $0.10–$0.15.

Step 4 — Count seven days. The dispute window opens. During that period anyone can submit a fraud proof against the state containing your withdrawal. You do nothing. Your $1,000 is neither on the L2 nor on the L1: it is in transit, locked in the bridge contract.

Step 5 — Finalise on L1. Once the window closes, you sign a second Ethereum transaction to claim the funds. Another $0.10–$0.15 or so.

Total: around $0.25–$0.35 in fees and seven days of waiting to move $1,000. As a percentage that is 0.03%, which is nothing. In time it is a week during which you cannot react to anything.

The alternative is a third-party liquidity bridge, which fronts you the funds on L1 instantly and absorbs the seven-day wait itself. It charges a fee for that service, varying by bridge and by moment, and adds the bridge contract's own risk. Before using one, ask yourself whether the urgency is worth the exposure.

Checklist before putting money on an L2

• Check on L2BEAT what type of rollup it is and which guarantees are already live: fraud proofs in production, forced inclusion, emergency withdrawals.
• Find out who controls the contract upgrade keys, and whether there is a timelock before an upgrade takes effect.
• Look at TVL and activity. A network with little volume is a candidate for shutdown, even if it works perfectly today.
• Use the native bridge for large amounts. Third-party bridges are faster and add one more contract you are trusting with your money.
• If you are leaving funds idle, leave them in your own account, not inside a protocol. The Polygon zkEVM case shows that this distinction decides whether you get them back.
• Write down which assets you hold on which network. In an orderly exit, your worst enemy is having forgotten you had something there.

Where we stand in late 2026

Ethereum has explicitly embraced the settlement layer role. Glamsterdam, targeting 4 November 2026, and Hegotá, aiming for May 2027, follow the same line: better separation of roles within the block, censorship resistance, statelessness. None of it turns the L1 into a mass execution layer, because that is not what is being sought.

The consequence is that the relevant question is no longer whether to use an L2, but which one and with how much exposure. The sector is concentrating: as of August 2026 Base and Arbitrum One between them hold nearly $24 billion, while everything else competes an order of magnitude below. The closure of Polygon zkEVM, orderly as it was, is the first clear signal of what happens to the ones that fall behind.

Frequently asked questions

Sources and references

1. L2BEAT — Layer 2 TVL, costs and risk analysis
2. Polygon — official Polygon zkEVM page (sunset notice)
3. EIP-4844: Shard Blob Transactions (proto-danksharding)
4. Ethereum.org — Optimistic rollups
5. Ethereum.org — Zero-knowledge rollups
6. Chainalysis — Crypto hacking and stolen funds 2026
7. DefiLlama — TVL by chain

Related guides