Author: Ryan Y Yi, Head of Investment, Coinbase Ventures; Michael Atassi, Chartered Financial Analyst, Coinbase; Translation: Golden Finance xiaozou

Key points:

  • While L2’s lower gas fees/increased throughput make them central to on-chain activity and a sizable portion of the ETH economy, they may be subject to pressure to maintain decentralization and alignment with ETH L1.

  • Builders who want to experiment and customize their own applications and distribute them in alignment with L2 are now choosing to build L3 - application chains based on the underlying L2.

  • The focus of this article is to build a common understanding of L3.

1 Introduction

（1）What is L3?

If L2 is the on-chain hub, then L3 can be thought of as "on-chain servers" that have isolated state environments and fee markets, but are based on the underlying L2 and utilize its deposit/distribution mechanisms. This provides applications with customizable block space while still leveraging the existing liquidity and user base of L2.

  • Cost: 1000x lower due to a combination of 1) lower entry costs (directly from centralized exchanges to L2), 2) slightly lower settlement/execution costs (trades settle to L2 instead of L1), and most importantly 3) fungible data availability (DA), which is how chains verify data accuracy (DA is 95%+ of the total cost of L2 using ETH L1 data). Gas fees are also more predictable because L3 has its own fee market (e.g., on L2, a surge in activity for one application will increase fees for all other applications).

  • Customizability: L3 adopts a lower decentralization standard than L2, which unlocks the ability to experiment with new token economics (i.e. custom gas tokens), virtual machines (i.e. Solana VM on ETH L2), and alt-DAs (i.e. Celestia instead of ETH L1).

（2）What is the difference between L3 and L2?

L3 is rollup, so it has many similarities to the way people understand L2.

  • Settlement: Similar to L2 settling to L1, L3 settles to L2.

  • Bridging: Just like how assets are typically (or third-party) bridged from L1 to L2, the same is true for bridging assets from L2 to L3.

  • Virtual Machines: The software used by L3 does not necessarily need to run on the same technology stack as its underlying L2. For example, many existing L3s run on Arbitrum Nitro, but ultimately settle with Base (which runs on the OP Stack). In addition, most L3s are modified versions of existing popular L2s. For example, Arbitrum (Nitro) and OP Stack have released technology stacks that have been adjusted to the needs of L3 builders.

  • Data availability: This is the biggest differentiator. L3 will choose to use alternative DA layers (e.g. Celestia, EigenDA, Arbitrum AnyTrust), while L2 must use ETH L1 for alignment/decentralization. Therefore, L3 achieves an extremely low-cost gas environment.

(3) How to release L3?

Since L3 primarily leverages a permissionless open source technology stack, developers can: 1) run the technology stack/infrastructure themselves, 2) leverage a RaaS (Rollup-as-a-Service) provider that provides managed services (e.g. Conduit, Caldera) to deploy and host L3, or 3) consult a white label service provider (e.g. Syndicate) who will “subcontract” to various infrastructure providers (e.g. RaaS, Bridges, DevTools).

(4) Will L4 appear?

  • Since L3 provides dedicated blockspace and the ability to natively bridge to L2 “hubs” liquidity/users, we believe this will cover all true on-chain use cases.

  • Even if L2 transaction costs fall, L3 may be the “last” frontier for vertical scaling (i.e. there is no L4).

Integration with L2 - The core assumption of L3 is to be able to leverage the liquidity/users of the underlying L2 "hubs". If you create an "L4", you will be further away from L2 and deviate from the original goal.

No cost improvements - Alternative DAs are meant to reduce costs. Moving up the tech stack does not really change the settlement/execution cost structure.

  • If L3 reaches its scaling limit, without further vertical scaling ("L4"), they will likely spin up another L3 (connected via native bridging) based on the same L2. The conclusion is that L3 will likely scale horizontally, not vertically.

2. Ecosystem Impact

(1) L3 will become another preferred direction for on-chain builders, which may lead to a situation where a few L2 "hubs" own millions of L3 "servers".

  • For on-chain developers, L3s represent a potential paradigm shift as they break down underlying barriers and lower the barrier to entry for developing mainstream-scale on-chain applications, potentially spawning an “app store” moment with millions of L3s.

  • L3 provides a platform for builders to experiment, ideal for high throughput/low cost applications - which can then leverage the underlying L2 hubs to support liquidity/onboarding/distribution.

  • The likely result is that there are tens to hundreds of L2 hubs, and potentially millions of L3 hubs.

(2) From a cost perspective, L3 has the potential to bring about an “AWS” moment.

  • One observation is that L2s are becoming their own on-chain hubs. Due to their proximity to L1s, L2s are often expensive to operate, and can cost anywhere from 7 to 8 figures USD per year.

  • On the other hand, the operating cost of L3 is much lower, with the annual cost of operating L3 ranging from US$25,000 to US$50,000.

(3) L3 developers will promote the popularity of more frameworks besides Solidity/Vyper, thus forming a multi-virtual machine environment.

  • There are a number of projects attempting to deploy alternative frameworks on Ethereum (e.g. MoveVM, SolanaVM, Arbitrum Stylus). The goal is to expand the range of developer products while leveraging existing network effects, liquidity, and Ethereum deposit methods.

  • This may first occur at the L2 level - but we can foresee these frameworks being deployed as L3 in order to leverage L2 hubs like Base.

  • The end result is that L2 can attract a broad range of developers at the L3 level while maintaining its own chain on the EVM (rather than trying to integrate multi-VM directly into L2).

(4) L3’s value stream will depend on the application layer

  • The KPIs for a single L3 are users, transactions, and token utility, not sorter fees. The average value created by a single L3 may be small — but as the number of L3s increases, this will create a network effect.

  • The growth of L3s generally benefits value creation on the software side (e.g., development tools, RaaS) and value creation on the protocol side (data availability, chain abstraction), but can only scale with a large number of L3s.

  • We can foresee that a single issuer/project may launch multiple L3s, thus forming its own L3 ecosystem. For example, an on-chain gaming ecosystem may provide one L3 for each game, and the emerging ecosystem will provide the remaining value to be shared with other stakeholders.

(5) L3 needs smoother interoperability and chain abstraction to succeed

  • If the intended purpose of L3 is to leverage the user experience of L2 users, and we expect there to be more and more L3s per application use case, then the interaction with these L3s needs to become seamless at the user level.

  • Similar to L2, L3 bridging can be achieved in two ways: native settlement of L3 to L2, or through a third-party vendor. Due to the experimental nature of the L3 technology stack, third-party vendors are more suitable for L3, which may result in a bridging layer that is neither unified nor flexible.

  • At the same time, L3 may only prioritize interoperability with the canonical L2 settlement chain, rather than aiming for full interoperability with all other chains. Therefore, they will focus on enhancing the functions and features of the bridge, such as reducing latency and providing one-stop liquidity, to improve the overall user experience.

  • There is ongoing protocol development around how to introduce native concepts at the sequencer level.

3. Future prospects

In summary, the L2 ecosystem will witness growth in L3 builders who want to create isolated on-chain application experiences while leveraging the underlying L2 hubs.