In early June, around 2,500 decision-makers from Web3, stablecoins, and AI infrastructure discussed precisely these questions at the Proof of Talk event in the Louvre Palace in Paris: How do autonomous systems pay on the internet, and which solution will become the standard? The term "agentic payments" initially sounds like chatbots that, for example, shop online and pay directly for users. While this area is also a growth market, agentic payments refers to something different: agent-to-agent and machine-to-machine payments, in which servers, APIs, or AI agents autonomously process micro-amounts for data, compute, or content – without a human checkout.
Agentic Payments means machines that pay for machines
When a user uses ChatGPT to order something on Amazon, the industry refers to this as "agentic commerce"—Stripe and OpenAI also call it "instant checkout in chat." For infrastructure and payment architecture, this is a consumer use case with card rails (credit/debit) and mandate protocols (such as Google's AP2), but not the machine-based micropayment scenario described as agentic payment. The latter involves server-to-server requests: a crawler, for example, pays per item request, an AI agent per API call, a model per inference slot. The amounts are sometimes in the micro-cent range. Such micropayments are not economically viable for traditional card networks like MasterCard or Visa, nor for comparable existing payment solutions.
Case study: Data is becoming scarcer – and AI models need it in trillions.
Genuine, authentic content once again commands a market price because frontier models are based on massive training data. At its core, a Large Language Model is a high-dimensional mathematical function with billions to trillions of optimized weights; through backpropagation, it minimizes an error function, making input-output pairs generalizable. This requires enormous, high-quality, usable corpora – and these are becoming increasingly scarce: According to Epoch AI, high-quality human-generated text could be exhausted between 2026 and 2032. Simultaneously, restrictions are increasing on nearly a third of the top web sources for crawling, i.e., the automated searching by, for example, AI agents.
Publishers are responding to the crawling trend with pay-per-use models: Cloudflare is testing Pay Per Crawl – a fixed price per request a crawler makes. This price is also in the previously mentioned microcent range, and these payments are processed by Cloudflare via HTTP 402. Payments for content are a very specific use case for Agentic Payments. The Really Simple Licensing (RSL) specification defines machine-readable license types for this purpose. crawl, use and training in robots.txt and headers. RSL's co-founders include Reddit, Automattic (WordPress/Elementor), and Fastly. Soon, WordPress owners might be able to monetize their individual blog posts using such agentic payments—that is, every time an AI agent crawls the blog to use the content. Indirectly, AI would thus pay humans for the originality of their content. The fundamental principle for all agent- and machine-based systems of the future is: they need autonomous microtransactions. Not only with content payments, but also with other payment methods, the individual costs per request are often in the cent or sub-cent range. Existing payment systems are not designed for such payments and would incur more processing costs than the actual amount is worth.
Blockchain promises transparency, AI monopolizes computing power – micropayments as a bridge
The idea of linking AI and blockchain emerged early on: if data, computational results, and payments run on a traceable blockchain layer, origin and rights could be more clearly assigned – and licensing logic could be structured in a delegable way, instead of being negotiated manually each time. Research frameworks like IBis (Dataset Provenance, Bilateral Licenses) or Content ARCs (Authenticity → Rights → Compensation via Distributed Ledger Technology) aim precisely at this. They already demonstrate that this fusion of the two future technologies can work.
AI itself, however, is moving in a different direction. Frontier models run in centralized computing clusters and proprietary stacks; blockchain, on the other hand, relies on open protocols and verifiable states. In plain terms, this means that both technologies promise transparency, but scale according to opposing logics – monopolized interference on the one hand, decentralized settlement on the other.
This is precisely where x402 and stablecoin micropayments can bridge the gap: They enable HTTP-native settlement for agentic workloads and verifiable "on-chain" settlement—without requiring the training itself to run on-chain. In particular, x402 solves the fundamental problem with micropayments that also affects blockchains: Transaction costs are higher than the transaction amount itself.
x402 turns HTTP 402 into a payment channel – open to any payment model
The HTTP status code 402 Payment Required has been around longer than most web developers realize: it was introduced back in 1997 in RFC 2068 – an official standard document that defines how web servers and clients communicate with each other. Since then, the IETF (the body that standardizes internet protocols) has consistently listed the code as "reserved for future use" in RFC 2616, RFC 7231, and RFC 9110 – meaning reserved, but without an implemented micropayment standard. It was reserved by the HTTP/1.1 authors, including Roy Fielding and Tim Berners-Lee. Individual services later used 402 sporadically and proprietaryly, for example, for API quota notices; the status code was never standardized across the web.
Status codes are the three-digit numbers a server uses to classify each request: 200 means "successful, the data is here," 404 "resource not found." The 402 code was intended from the beginning as a paywall signal – "access only with payment.".
x402 builds on precisely this mechanism. Fundamentally, the web functions as a request/response – a question and answer process: The client (browser, app, or AI agent) sends an HTTP request to a server. GET requests retrieve data, while POST requests typically send data for processing. The server responds with a status code, headers (metadata about the request), and a body (the actual content).
If payment is required for a resource, the server responds with a 402 error and includes the following information in the header: PAYMENT REQUIRED The payment terms are provided as Base64-encoded JSON, a machine-readable data format in coded text form. The client signs a payment payload (the specific payment authorization) and resubmits the request with the header. PAYMENT SIGNATURE. Only after Verify (payment verification) and Settle (final processing) does 200 follow – along with the resource and the header. PAYMENT RESPONSE. The status code 402 is therefore only a signal "payment required". Which payment model is applied is contained in the payload.
To prevent the standard from becoming the exclusive domain of a single company, the Coinbase Developer Platform released the protocol as an open standard in May 2025 (whitepaper, Apache 2.0); launch partners include AWS, Anthropic, and Circle. In September 2025, Cloudflare and Coinbase announced a joint foundation; in April 2026, the Linux Foundation launched the x402 Foundation with governance provided by Coinbase, Cloudflare, and Stripe – with contributions from Google, Visa, Mastercard, and AWS.
Specifications and documentation are available at x402.org and docs.x402.org SDKs (Software Development Kits – pre-built code libraries that allow developers to integrate the protocol) facilitate integration. The reference code is available in the GitHub repository. x402-foundation/x402 – a publicly accessible code archive on GitHub. Version 2 supports CAIP-2 network IDs for the standardized naming of blockchains and a facilitator API with the endpoints /verify and /settle – i.e., the URLs under which an intermediary service checks and completes payments.
What x402 leaves open is the actual payment model. Within the protocol, these are scheme definitions in JSON – theoretically, any internet-enabled middleware can connect: dummy schemes for testing, map networks, or internal accounting systems in closed ecosystems. Conceptually, abstracted bank rails (SEPA, TARGET2, SWIFT-MT) would also be conceivable. In practice, stablecoin schemes on the EVM (Ethereum Virtual Machine – the common execution environment for smart contracts on Ethereum and compatible networks) dominate today because programmatic signatures and sub-cent settlement are already implemented there. In a closed ecosystem, custom x402 models can be defined as long as the server and client use the same scheme. The advantage of the open work of the Foundation: interoperability across vendor boundaries.
EVM-L2s and stablecoins gain traction through fees, not vision.
The growing relevance of stablecoins can be quantified. Chainalysis estimates the adjusted stablecoin volume in 2025 at $28 trillion of real economic activity; the BIS cites a market capitalization of over $300 billion, roughly 98 percent of which is USD-denominated. Both observe increasing use for treasury and cross-border payments – a trend also evident in emerging markets (see...). Stablecoin deployment in Africa).
For agent-based micropayments, a second point comes into play: settlement – the final transfer of funds – must be machine-verifiable. Publicly verifiable on-chain events (transaction records stored in the blockchain and accessible to everyone) are therefore a good fit for open-source protocols, open APIs (programmable interfaces through which software accesses services), and autonomous clients. They allow for an audit trail for each request without requiring a human to manually account for every cent.
Technically, Ethereum/EVM has long provided the compatibility layer – the common foundation upon which many blockchains and payment protocols are built. x402 v2 uses CAIP-2 IDs for network naming, for example. eip155:8453 for base or eip155:137 for polygon; the prefix eip155 Here, stands for Ethereum-compatible chains. For gasless USDC payments, i.e., without a transaction fee, EIP-3009 is used – an Ethereum Improvement Proposal that allows signed off-chain authorizations without the payer having to pay gas (the transaction fee on the blockchain). Permit2 complements this for any ERC-20 tokens, the standard for fungible cryptocurrencies on EVM.
Coinbase's CDP facilitator (a service for verifying and settling) operates on Base, Polygon, Arbitrum, World, and Solana; the whitepaper mentions sub-second settlement and gas fees below $0.0001. For comparison, credit card networks often charge fees of around $0.30 per transaction, making cent or sub-cent payments simply uneconomical.
Whoever achieves production volume, SDK distribution (i.e., ready-made developer tools for rapid integration), and enterprise connectivity first sets the de facto standard – currently with a lead over Base/EVM thanks to Coinbase infrastructure. A two-cent transaction shouldn't cost five cents in fees; therefore, the L2 economy – meaning inexpensive transactions on Layer 2 networks built on Ethereum – is a prerequisite, not a nice-to-have.
Conclusion
For agent-based micropayments, the EVM ecosystem—especially USDC on L2s like Base and Polygon—is currently the most pragmatic blockchain ecosystem: open specifications, low fees, and existing stablecoin liquidity. However, the next stage of maturity is not just about technology, but also governance: custody models, mandate and liability issues for autonomous agents, and interoperable custody. This will determine whether x402 remains a niche protocol or becomes infrastructure.
Much is being built from scratch – at the same time, the market is relying on existing layers: HTTP, EVM, stablecoins. TokenPay operates at precisely this interface – cross-border fiat and stablecoin flows – and builds partner infrastructure on existing rails instead of parallel, isolated solutions.
You can find more information about TokenPay infrastructure here. here
Sources
- RFC 9110 HTTP 402 Payment Required
- Coinbase Introducing x402
- x402 Whitepaper
- Linux Foundation x402 Foundation Launch
- Cloudflare Blog x402
- Stanford AI Index 2025
- Epoch AI Will we run out of data?
- Longpre et al. Consent in crisis
- Cloudflare Introducing Pay Per Crawl
- RSL Standard 1.0
- Tiger Research Payments 3.0: The AI Agent Payments Engine
- Google Cloud AP2 Agent Payments Protocol
- IMF Notes 2026/004
- Chainalysis Stablecoin utility & future of payments
- BIS Papers No. 170
- Coinbase CDP x402 Documentation
- Circle Agent Nanopayments
- Content ARCs (arXiv 2025) / IBis (arXiv 2024)
- Proof of Talk 2026
- RFC 2068 HTTP/1.1 (Jan. 1997)
- MDN HTTP 402 Payment Required
- OpenAI Buy it in ChatGPT
- docs.x402.org
- Fastly Control and Monetize Your Content with RSL
- RSL Collective Press Release
- UK Government Report on Copyright and Artificial Intelligence
- Automattic x402-Pay
- EIP-3009
- USDC.org
- Stripe
- Cloudflare Press Release
- Markets Insider
- IBis
- Content ARCs
- Square Credit Card Processing Fees Explained
- NanoTrans Agent Paywall
