The table below compares the Web3 infrastructure tokens in the Agent Infrastructure thesis across the full stack — settlement, execution, compute, intelligence, memory, data, verification, identity, connectivity, and agent networks. Some are broad foundational layers; others are purpose-built for a single function. Together they form the rails on which scalable, trust-minimized agent activity is built.
| Token | Layer | Role in the Agent Economy | Key Strengths | Risks / Considerations | Thesis Fit | Article |
|---|---|---|---|---|---|---|
| Ethereum ETH |
Settlement | The base settlement layer — where ownership, smart contracts, AVSs, and agent agreements are recorded and made final. | Highest security, deepest liquidity, largest developer ecosystem; programmable settlement at scale. | Higher base-layer fees (mitigated by L2s); competition from other L1s. | Core | Read → |
| Solana SOL |
Execution | The high-performance shared-chain execution environment — fast, low-cost transactions for agent activity and DeFi. | Speed and low cost; deep DeFi and on-chain agent activity; large ecosystem. | Historical network outages; centralization concerns. | Core | Read → |
| Avalanche AVAX |
Execution (Sovereign Chains) | Sovereign, custom L1s with sub-second finality — the DeFAI angle: a chain tuned per agent workload rather than one shared chain. | Sub-second finality, EVM compatibility, cheap post-Etna L1 deployment, institutional RWA traction. | Crowded fast-L1 field; subnet fragmentation can dilute AVAX value capture; narrative-differentiation risk. | Complementary | Read → |
| NEAR NEAR |
Intents | The intent layer — agents express desired outcomes and the network resolves how to fulfill them across chains. | Intent-centric design and chain abstraction; usability focus; active AI orientation. | Competitive intent/abstraction field; adoption must translate into durable usage. | High | Read → |
| Bittensor TAO |
Intelligence | A decentralized marketplace for machine intelligence — subnets compete to produce inference, training, and agent capabilities. | Proven intelligence market; rich subnet ecosystem; strong incentive design. | High FDV vs. circulating supply; subnet competition intensity. | Core (Intelligence) | Read → |
| Allora ALLO |
Collaborative Intelligence | Synthesis layer — combines many models into a single accuracy-weighted prediction agents can act on. | Accuracy-weighted, context-aware predictions; verifiable contribution model. | Must prove the combined output beats single models; adoption and value-capture risk. | Medium | Read → |
| Sentient SENT |
Open Reasoning | Open development of composable, verifiable reasoning components — the building blocks of intelligence, built in the open. | Open, composable reasoning ethos; community curation (GRID). | Earliest-stage and most speculative; "AGI" framing is ambition, not deliverable. | Medium (Speculative) | Read → |
| Chainlink LINK |
Data | The data layer — oracles deliver verified external information agents need before they can act. | Dominant oracle network; deep integrations; data-integrity focus. | Oracle competition; data-integrity and dependency risk. | High | Read → |
| Render RENDER |
Compute (GPU) | Decentralized GPU network for AI inference and rendering workloads. | Mature GPU infrastructure; established AI-workload support. | Historically rendering-focused; compute-market competition. | Medium-High | Read → |
| Akash AKT |
Compute (Open Cloud) | An open, decentralized cloud marketplace for on-demand compute and GPU. | Cost-effective compute marketplace; open-cloud positioning. | Competition from larger GPU networks; utilization must recur. | Medium-High | Read → |
| Aethir ATH |
Compute (Enterprise GPU) | Distributed, enterprise-grade GPU supply for high-performance AI and agent workloads. | Enterprise, high-performance focus; treasury-backed compute model. | DePIN operational/hardware risk; commoditization and margin pressure. | Medium | Read → |
| Internet Computer ICP |
Compute (On-Chain) | On-chain compute and agent hosting — applications and agents run directly on the network. | Full on-chain execution and hosting; distinct from compute marketplaces. | Complexity; adoption and competitive positioning. | Medium-High | Read → |
| Phala PHA |
Confidential Compute | TEE-based confidential compute — private, verifiable AI inference and agent execution that never exposes data or models. | Hardware-backed privacy + attestation; real inference volume; enterprise compliance. | Hardware/TEE dependency; Ethereum-L2 migration and evolving PHA→vPHA mechanics (verify on docs); zk competition. | High | Read → |
| 0G 0G |
Agent-Native L1 (Full-Stack) | A purpose-built chain for agents — verifiable compute, storage, data availability, and an EVM L1 integrated as one stack. | Full-stack, agent-native design; well-funded; mainnet live with real ecosystem traction. | Execution risk on full-stack delivery; crowded AI-L1 field; value capture depends on real agent usage. | Very High | Read → |
| Filecoin FIL |
Memory (Storage) | Decentralized storage — the persistent memory layer where agent data and state are kept and retrieved. | Large storage network; provider collateral model; retrievable agent memory. | Storage-market competition; usage must reflect real demand. | High | Read → |
| Arweave AR |
Memory (Permanent) | Permanent storage — a one-time-payment, permanent record layer for agent data that must never be lost. | Permanence model; distinct from rentable storage; durable records. | Permanence-economics durability; differentiation from FIL. | Medium-High | Read → |
| EigenCloud EIGEN |
Verifiable Trust | Cryptoeconomic security and verifiable computation via restaking — lets agents trust records and computations, not just assert them. | Horizontal trust layer; restaking-based security; verifiable compute/inference + forkability. | Slashing and systemic restaking risk; verifiable-compute adoption; zk competition. | Core (Verification) | Read → |
| World WLD |
Human Identity | Proof-of-personhood — lets an agent prove a unique, verified human stands behind it (World ID + AgentKit). | Large verified-human base; shipped agent toolkit; high-profile backing and integrations. | Central biometric/regulatory risk; WLD is governance/utility, not the credential itself; indirect value capture. | Satellite | Read → |
| Helium HNT |
Physical Connectivity | Decentralized wireless — the bridge that lets agents reach real-world devices, sensors, and the physical edge. | Established DePIN network; real coverage; distinct physical-edge role. | Peripheral layer — relevance depends on edge-agent growth; usage-vs-coverage and competition risk. | Peripheral | Read → |
| Venice VVV |
Private Inference | Private, permissionless inference access — capacity delivered through staking rather than per-query billing. | Capacity-backed token model; privacy positioning; established and liquid. | Niche demand risk; large-platform competition; regulatory exposure on permissionless content. | Medium | Read → |
| Artificial Superintelligence Alliance FET / ASI |
Agent Networks (Coordination) | Autonomous agent coordination and multi-agent systems — infrastructure for agents to find and work with each other. | Direct agent narrative; merged established teams; coordination focus. | Post-merger execution risk; evolving naming/ticker (verify ASI/FET). | High | Read → |
| Virtuals Protocol VIRTUAL |
Agent Networks (Commerce) | Agent creation, tokenization, and commerce — rails for building, owning, and monetizing agents (ACP). | Strong monetization rails; real traction; agent-economy primitives. | Competition among agent platforms; durability of agent revenue models. | Very High | Read → |