Vitalik Buterin, one of Ethereum’s founders, together with Davide Crapis, has introduced a novel research proposal exploring how blockchain technology can be applied to the billing of artificial intelligence-based services. Rather than advocating for running widely-used language models directly on-chain, their study underscores the importance of leveraging blockchain infrastructure as a privacy-preserving payment and settlement layer for digital services.
Reimagining AI and API Fee Models
Buterin’s framework centers on building an infrastructure that enables multiple API calls to be paid for via blockchain, without disclosing a user’s identity or compiling detailed usage histories. At the heart of this system is a model where users deposit stablecoins in advance, and each API request is authenticated using zero-knowledge proofs, thus ensuring both security and anonymity.
ZK API Usage Credits and the RLN Mechanism
Central to their proposal is the concept of “ZK API usage credits,” which operate through a zero-knowledge-based tool known as RLN. Traditionally used on blockchain as an anti-spam measure that protects user anonymity, RLN in this context allows users to prepay and then make multiple API requests without revealing their identity.
Within this system, users can deposit, for instance, 100 USDC to perform 500 language model queries, or 10 USDC for 10,000 Ethereum RPC transactions. Each API request confirms on-chain that the relevant credits remain available and ensures that transaction rights are not being double-spent. If a user attempts to reuse a right, the RLN mechanism automatically applies a financial penalty by deducting part of the user’s staked amount.
The system employs a “multi-call per deposit” logic for scaling. Rather than linking blockchain transaction volume to total queries, it relies on account verification and settlement intervals. For actions with variable costs, users can receive refunds for unused credits via rebate tickets, enhancing both flexibility and efficiency.
Infrastructure Ready, but Hurdles Remain
With the stablecoin market exceeding $307.6 billion in circulating supply, there is ample liquidity to support this kind of deposit-based payment model for high-frequency services. Ethereum’s layer-2 scaling solutions now enable thousands of transactions per second, while the average transaction fee has fallen to around $0.21—a technical backdrop that suggests feasibility for Buterin’s proposed architecture.
Nonetheless, the promise of perfect privacy is not fully realized by the proposal. Critics of the research have pointed out that even with cryptographically assured anonymity, certain metadata—such as timing, request types, or data structures—could still be correlated to de-anonymize users. The RLN-based solution, while promising, currently remains inactive in ongoing development. Full-scale integration will require additional software updates before the system is ready for real-world deployment.
For this concept to take hold, major API and artificial intelligence service providers would need to adopt on-chain payment infrastructures and begin accepting stablecoin transactions. However, the conventional sector already relies on firmly established, identity-based payment and compliance frameworks, which could make the transition to blockchain-based solutions a gradual one.
In repositioning Ethereum’s role within the world of artificial intelligence, Buterin’s approach moves away from executing applications directly on-chain. Instead, he envisions Ethereum as a neutral, programmable hub for payment and settlement. Should the model see traction, it could further stimulate activity in the stablecoin arena, while driving up transactional volumes on layer-2 networks.
