A common assumption in early AI agent discussions was that the "agent layer" would consolidate on one or two chains — probably Ethereum mainnet or a major L2. The ERC-8004 data tells a different story. Agent registrations are genuinely spread across 18 chains, and the distribution isn't random. Different chains are attracting different types of agents for structurally different reasons.
Understanding which chain serves which agent archetype isn't trivia — it's the foundation for any serious intelligence work on the agent economy. And building a unified index across 18 different RPC endpoints, block finality models, and event schemas is meaningfully harder than it sounds.
Where the Registrations Actually Are
Here's the current distribution of ERC-8004-compatible registrations by chain, as tracked by Trustprint:
| Chain | ~Share | Primary Agent Type | Why It Matters |
|---|---|---|---|
| Base | 38% | DeFi, tokenized, consumer | Virtuals ecosystem, cheap gas, Coinbase distribution |
| Ethereum Mainnet | 22% | High-value DeFi, institutional | Oldest registrations, highest-value agents |
| LUKSO | 14% | Identity-rich, social, UP-backed | Richest metadata, Universal Profile linkage |
| Arbitrum | 9% | DEX integrations, trading | GMX/Uniswap v4 agent integrations |
| Optimism | 7% | Public goods, OP Stack tooling | Retroactive PGF agent experiments |
| Polygon | 4% | Enterprise, legacy integrations | Mature tooling, established enterprise presence |
| Other (12 chains) | 6% | Experimental | Frontier deployments, new L2s, niche ecosystems |
The concentration on Base is the most striking figure. Over a third of all agent registrations happen on a chain that launched its ERC-8004 support less than 18 months ago. The Virtuals ecosystem (tokenized agents with market caps) is the primary driver, but Base's broader developer growth and the Coinbase ecosystem's AI push are structural contributors.
Why No Single Chain Dominates
Different chains have different properties that attract different agent archetypes.
Gas economics drive agent frequency
An AI agent that executes 50 transactions per day on behalf of users cannot operate sustainably on Ethereum mainnet at mainnet gas prices. L2s — particularly Base and Arbitrum — are the natural home for high-frequency agents. Mainnet is reserved for agents that transact infrequently but at high value: agents managing large DeFi positions, agents settling multi-chain operations, agents interfacing with mainnet-native protocols.
Ecosystem gravity pulls developers
LUKSO's focus on identity and the creator economy has created a distinct cluster of "social" agents — agents that manage content, curate feeds, engage with creator profiles. These agents exist almost exclusively on LUKSO because the protocols they integrate with (Universal Profiles, LUKSO Grid, LSP social standards) only exist there.
Regulatory and compliance considerations
Some institutional agent deployments specifically target chains with mature compliance tooling or regulatory clarity in relevant jurisdictions. This is a small but growing cohort that shows up in the Polygon and Avalanche numbers.
The Technical Challenge of Cross-Chain Indexing
Building a unified index across 18 chains is not a "18× the single-chain problem" scaling challenge. It's a substantially more complex integration and normalization problem.
Heterogeneous RPC infrastructure
Each chain has its own RPC endpoints, rate limits, block time, and finality model. A crawler that works reliably for Ethereum mainnet (12-second blocks, deep finality) needs significant adaptation for chains with 2-second block times or probabilistic finality. Reorg handling is different on every chain.
Event schema fragmentation
ERC-8004 specifies the interface, but implementations vary. Some registries emit AgentRegistered(address agent, string metadataURI). Others use different event names, parameter ordering, or indexed fields. Normalizing these into a consistent internal schema requires chain-specific parsing logic.
The metadata URI problem: Metadata URIs can point to IPFS, Arweave, HTTP endpoints, or on-chain storage — and each has different availability guarantees. An agent registered in January 2026 might have an IPFS URI that's now unpinned. Robust indexing requires metadata caching and stale-URI detection.
Cross-chain identity matching
An agent deployed by the same operator on Ethereum and Base might have different addresses (if not using CREATE2 with the same salt), different metadata content (different descriptions or capability tags), and no explicit link between them. Connecting these as "same operator, related agents" requires operator identity correlation — a non-trivial matching problem.
What the Cross-Chain View Unlocks
Despite the complexity, the cross-chain view is where the most interesting intelligence lives. Questions that require cross-chain data:
- Which operators are deploying on multiple chains simultaneously — and which chains are they choosing?
- Where do high-value mainnet agents also have L2 presences, and what are those L2 instances doing?
- Which new chains are seeing their first meaningful agent registrations — early signals of ecosystem emergence?
- What's the metadata quality distribution by chain? (LUKSO consistently highest; Base mid-tier; some L2s showing increasing quality as tooling matures)
These questions can't be answered from a single chain. They require the unified index.
Trustprint currently ingests daily snapshots from all 18 chains. The live data dashboard shows aggregate counts; the newsletter provides the cross-chain trend analysis weekly. As the index deepens, the per-chain breakdowns will be surfaced in the live dashboard as well.