The public discussion about artificial intelligence and blockchain often presents a competition for capital, talent, and attention.
Charles Hoskinson, founder of Cardano, proposes a different reading: blockchain does not compete with AI; blockchain aspires to become coordination infrastructure for agents, data, and compute. My position is straightforward.
The crypto sector should evaluate the hypothesis with criteria from engineering and political economy, not from market narrative.
The relevant question is not whether AI replaces blockchain. The relevant question is which layers of the AI stack can benefit from permissionless settlement, verifiable governance, and cryptographic traceability.
The argument about machine-to-machine payments deserves attention. AI agents operate with programmatic objectives.
An agent needs an article, a dataset, or an external inference. The agent does not navigate with human intent. The agent executes a call and requires a payment mechanism with finality, low cost, and automatic verification. Stablecoins and payment protocols such as x402 aim to cover the function.
Blockchain provides a shared registry, settlement in seconds or milliseconds depending on the network, and contracts which condition payment on delivery of data. The limitations are known: latency, cost per transaction, custody complexity, and regulatory compliance.
The probable solution is not to execute every micropayment on the base layer. The probable solution uses payment channels, rollups, and account abstraction so the end user, human or agent, does not manage keys directly.
Alignment of AI is a less resolved problem
Behavioral rules for models and agents depend on internal policies of laboratories. An agent can receive contradictory instructions, operate across multiple jurisdictions, and execute actions with economic consequences.
Cryptoeconomic governance offers mechanisms to establish shared standards, incentives to report failures, and penalties for actors which violate rules. Credible commitments on a public network allow multiple parties to verify rules without reliance on a central authority.
Alignment does not reduce to code. Subjective, cultural, and legal components cannot be resolved with smart contracts alone. Blockchain can provide transparency and accountability.
Blockchain requires oracles, dispute committees, and governance design resistant to capture. The crypto sector should avoid promising a complete solution.
Data provenance is another area where blockchain can add value. AI models train and generate content from works, databases, and user signals. Tracking origin, license, and economic compensation is incomplete.
A signed registry with content hashes and inclusion proofs allows auditors to verify which data fed a model and execute royalties through contracts. Traceability does not require publishing full content. Traceability can combine with zero-knowledge proofs to demonstrate properties without revealing sensitive information. Operational challenges remain: off-chain data, anonymization, jurisdictions, and models which do not expose weights. Blockchain does not resolve opacity of laboratories. Blockchain creates a market for verifiable licenses and reduces compliance cost for parties willing to pay.
The debate about compute infrastructure introduces a relevant critique. Expansion of data centers for AI training faces electricity, regulatory, and return-on-investment constraints. Comparison with fiber overbuilding in the 1990s is useful as a warning, not as a definitive prediction.
Risk of idle capacity exists if demand does not grow at the pace of supply. Distributed compute with consumer GPUs, federated training, and edge inference can cover specific workloads. Blockchain functions as a coordination and settlement layer for participants which contribute compute, verify results, and receive incentives. Technical obstacles are high: bandwidth, hardware heterogeneity, compute verification, and latency.
A hybrid model, with centralized training in specialized clusters and distributed inference on devices, appears more viable than a fully decentralized network.
Proof of compute is a principal constraint for any distributed AI network. Training a large model requires gradient coordination and parameter synchronization.
An open network introduces risks of fraud and free-riding. Zero-knowledge proofs for machine learning and trusted execution environments offer partial guarantees. Computational cost exceeds capacity of many current networks.
Optimistic protocols with dispute periods can work for inference, not for full training. The technical conclusion is clear. Full decentralization of frontier training is not viable in the short term. Decentralization of inference, data, and payments is viable.
The CLARITY Act and similar frameworks seek to define jurisdiction over tokens, stablecoins, and markets. Hoskinson predicts late approval, possibly toward 2029. The prediction reflects a political calculation. Without legal clarity, financial institutions and AI companies do not integrate on-chain payments at scale. Public image of the sector also matters.
“…There’s no pressure, political pressure to pass this type of thing. they’ll just wait until the next session and force uh a heavily unfavorable bill including ethics provisions to Trump uh on them if they want clarity. Of course, Trump won’t make those concessions. So, actually, we’ll have to wait till 2029 to get a new Clarity Act passed because of the ineptitude of uh of what the White House did. And I was very public about this…,” Hoskinson noted.
Association of crypto with promotional tokens or conflicts of interest reduces capacity to negotiate favorable rules. The sector needs separation between infrastructure and speculation. Regulated stablecoins, institutional custody, and identity standards are more useful mechanisms than marketing campaigns.
Projections of $10 trillion in public chain assets and 1 billion users by 2030 should be treated as conditional scenarios. Projections depend on technical execution, real adoption, and regulatory decisions. Crypto history shows linear predictions fail.
The Hoskinson thesis identifies three concrete problems: payments, alignment, and provenance. The possible error is to assume blockchain is the only solution or AI is an adversary. The more probable relation is asymmetric complementarity: AI consumes coordination services and blockchain provides economic guarantees.
For the crypto sector, the strategic implication is clear. Value will not come from launching tokens associated with AI. Value will come from building open standards for payments between agents, verifiable identity, reputation, proof of compute, and data governance.
Interoperability between chains is a requirement, not an option. Privacy must be integrated from design. Regulatory compliance cannot be treated as a later layer. User experience must hide keys and gas without sacrificing self-custody. Teams which solve settlement, verification, and dispute problems will find demand from autonomous agents and AI companies.
My opinion is that blockchain can succeed at the intersection with AI if the sector abandons maximalism and concentrates on settlement layer, provenance layer, and governance layer functions. The goal is not to replace language models or to decentralize all compute.
The goal is to offer guarantees AI cannot generate alone: programmable payments, data traceability, verifiable incentives, and dispute resolution. The real competition is not crypto versus AI.
The Hoskinson thesis can be summarized as a hypothesis of layer complementarity. AI needs payments, alignment, and provenance.
Blockchain offers mechanisms to coordinate services without single intermediaries. Adoption will not concentrate on one specific chain. Adoption will concentrate on protocols which achieve security, scalability, and compliance.
Be the first to comment