In March 2026, the global payments industry released five major signals in just one week: the Machine Payments Protocol (MPP), co-authored by Stripe, launched alongside the Tempo mainnet; Visa established Crypto Labs and introduced a command-line payment tool for AI; Mastercard acquired stablecoin infrastructure company BVNK for $1.8 billion; Coinbase’s x402 protocol underwent a significant upgrade; and World unveiled an AI-focused identity authentication toolkit.
These five industry giants, each from different sectors, pointed in the same direction within a single week—building autonomous payment capabilities for AI agents. This is not a random industry resonance, but a clear sign of structural change: the core participants in the payments ecosystem are expanding from "humans" to "machines."
Today, AI agents have evolved from simple conversational tools to autonomous entities capable of executing entire task chains. According to data disclosed by Circle, over the past nine months, AI agents completed 140 million payments totaling $43 million, with more than 400,000 AI agents possessing purchasing power. The average transaction size is just $0.31—these numbers illustrate the defining features of the AI agent economy: high frequency, micropayments, autonomy, and no human intervention.
Traditional payment systems were designed from the outset for "people": bank accounts require ID, credit cards need facial recognition, and the SWIFT system relies on manual authorization. When AI agents need to call APIs, purchase computing power, or access data, they can’t pass any of these checkpoints. This is precisely why payment infrastructure must be fundamentally restructured.
Driving Forces: Why Stablecoins Are the Only Choice for AI Payments
To understand the technical architecture of AI agent payments, we need to break it down into two layers: value carriers and interaction protocols.
At the value carrier layer, stablecoins show a unique adaptability compared to traditional payment tools. According to Dune Analytics, USDC accounts for 98.6% of settlements in existing AI agent payment scenarios. The logic is clear: AI agents don’t need speculative assets with price volatility—they need programmable, instant-settlement, low-friction value media. Stablecoins meet these needs perfectly—24/7 instant transfers, smart contract automation, and near-zero transaction costs for micropayments.
On the protocol layer, the x402 protocol has revived the long-dormant 402 status code (Payment Required) in HTTP. Stripe’s Machine Payments feature, built on this open standard, allows servers to return payment details (including price and wallet address) directly in response to requests. AI agents can then automatically complete on-chain transfers and resubmit the request with proof of payment. This "handshake" process embeds payments directly into the HTTP request cycle, making it as natural for AI agents to pay with stablecoins as it is to exchange data packets.
Stripe’s choice of the Base network for its initial launch is no coincidence. As an Ethereum Layer 2 solution, Base significantly reduces transaction fees, making micropayment business models feasible. CoinGecko has already started testing a pricing model of $0.01 USDC per request, allowing AI agents to pay instantly based on actual usage without subscribing to expensive monthly plans.
Structural Cost: The Mismatch Between Micropayment Scale and Infrastructure Valuation
Every emerging infrastructure market faces the same early-stage dilemma: "the road is built, but the cars haven’t arrived." The AI agent payments sector is no exception.
The x402 protocol is currently one of the most mature AI payment protocols in operation. However, according to x402scan, the entire ecosystem processed just $65,400 in transactions over the past 24 hours—about 150,000 transactions, with an average value of less than $0.50 each. In stark contrast, Tempo is valued at $5 billion, Mastercard spent $1.8 billion acquiring BVNK, and Stripe’s latest valuation stands at $140 billion.
This massive gap between valuation and transaction volume is a hallmark of infrastructure sectors. During the dot-com bubble in 2000, telecom companies laid millions of kilometers of undersea fiber optic cables, but only 5% of global internet traffic used them at the time. Most of those companies went bankrupt, but the fiber network remained and was fully utilized a decade later by video streaming and mobile internet.
AI agent payments are currently in this "road-building" phase. The demand logic is sound: AI agent capabilities are rapidly advancing—OpenClaw enables AI to operate computers directly, the MCP protocol allows AI to access external services, and major model providers are set to make significant agent advancements in the second half of 2025. Doing work requires spending money, and spending requires payment infrastructure. However, at this stage, infrastructure development is clearly outpacing actual transaction volume.
Industry Landscape: Four Distinct Competitive Camps
The competitive landscape for AI agent payments is already clear, with four main camps, each leveraging different core strengths to enter the market.
The first camp is the settlement layer players, represented by Circle, Tether, Stripe, and Coinbase. Leveraging the programmability of stablecoins and ultra-low-cost micro-settlement capabilities, they currently dominate most AI agent payment settlements. Circle’s Nanopayments system aggregates thousands of small payments off-chain and periodically settles them on-chain, reducing transaction costs to zero for developers.
The second camp consists of traditional payment giants, with Visa and Mastercard at the core. Drawing on their global payment networks, high merchant penetration, and robust compliance and risk controls, they have quickly launched payment tools tailored for AI agents. Visa’s CLI tool allows AI agents to initiate credit card payments directly from the terminal, while Mastercard’s acquisition of BVNK fills its stablecoin technology gap.
The third camp is global tech giants, represented by Microsoft and OpenAI. With control over large model entry points and global developer ecosystems, they focus on setting universal business protocols and AI-native payment standards, enabling closed-loop transactions within conversational interfaces through native integration with large models.
The fourth camp is Chinese players, with Alipay and WeChat Pay at the core. Leveraging their super-app user bases and merchant resources, they have rapidly scaled AI payment products and maintain a dominant position in the mainland market.
Evolution Path: Natural Extension from Payments to Asset Management
While AI agent payment infrastructure is taking shape quickly, a complete economic ecosystem also requires "asset management infrastructure." This is where the intersection of the RWA (Real World Asset) sector and AI agents begins.
When an AI agent generates ongoing income—whether by providing services to users or participating in distributed computing networks—funds will accumulate in its account. No rational economic entity would let idle funds remain liquid indefinitely. Circle’s data shows that over 400,000 AI agents now have purchasing power, and their account balances are growing.
In traditional finance, individuals and businesses deposit short-term idle funds in banks, buy money market funds, or purchase short-term government bonds. AI agents need the same capability—not for speculation, but to optimize their economic models. If funds exceeding payment thresholds can be automatically invested in a tokenized fund backed by short-term U.S. Treasuries and redeemed automatically when needed, the agent’s "operational efficiency" improves.
J.P. Morgan’s Kinexys division offers a useful reference point. The platform processes over $2 billion in transactions daily and has facilitated more than $1.5 trillion in notional value. Its "delivery versus payment" (DvP) model enables simultaneous exchange of assets and payments. In the future AI agent economy, transaction participants will shift from institutions to AI agents, and transaction sizes will shrink from millions of dollars to just a few, but the underlying logic remains the same—seamless value transfer and storage.
Risks and Boundaries: Compliance, Security, and Responsibility
Every infrastructure-level transformation brings multi-dimensional risks, and AI agent payments are no exception.
Compliance risk is the primary concern. According to mainland China’s regulations, "RWA tokenization and related services are strictly prohibited domestically," and all scenarios discussed here occur under overseas compliance frameworks. The global regulatory landscape remains fragmented—a stablecoin compliant in one jurisdiction may face restrictions in another. Hong Kong has implemented a licensing regime for fiat-backed stablecoin issuers, with the first licenses expected in March 2026, marking the formal entry of stablecoins into the regulated financial system.
Security risks are equally significant. Issues like the transparency of stablecoin issuers’ reserve assets, smart contract vulnerabilities, and cross-chain bridge security directly impact fund safety. Once AI agents achieve automated transactions, the speed and scale of potential exploits may far exceed those possible with human intervention.
The most fundamental risk is responsibility. If an AI agent makes an "investment decision" based on incorrect data or a flawed model, who is accountable? Is it the human, the protocol, or the AI agent itself? There is no clear answer in law or regulation. Cisco’s security team recently noted that OpenClaw once ran a malicious plugin that secretly sent user data to external servers. When risks extend from data security to fund security, flaws in the trust model become dramatically magnified.
Conclusion
The launch of Stripe’s Machine Payments Protocol marks the transition of AI agent payments from proof-of-concept to commercial reality. The coordinated moves of five global giants within a single week are not a coincidence but a collective response to structural trends: as AI agents evolve from "conversational tools" to "execution tools," payment systems must be rebuilt from "human-centric" to "machine-native."
Stablecoins have become irreplaceable in this transformation—their programmability, instant settlement, and micropayment capabilities are a perfect match for AI agents’ needs. The x402 protocol provides an HTTP-level interaction standard, embedding payments into machine-to-machine communications as naturally as data exchange.
At this stage, infrastructure development is clearly outpacing actual transaction volume—a typical feature of emerging sectors. Yet the demand logic is solid: 400,000 AI agents with purchasing power are waiting for more robust payment and asset management infrastructure. When this pipeline is finally filled, the AI+Crypto narrative will shift from proof-of-concept to a scaled economic reality.
FAQ
Q1: What is the Machine Payments Protocol (MPP)?
MPP is an open protocol co-developed by Stripe that standardizes processes for transactions between machines, including payment requests, authorization, and settlement. Launched alongside the Tempo mainnet, it enables AI agents to autonomously complete payments within preset limits, eliminating the need for human confirmation on every transaction.
Q2: Why can’t AI agents use traditional credit card payments?
Traditional credit card payments rely on identity verification (like facial recognition and SMS codes), credit assessment, and manual authorization—steps that AI agents cannot complete independently. Additionally, credit card fees are high, making them unsuitable for the high-frequency, low-value transactions typical of AI agents. Stablecoins, with their programmability and minimal costs, are a better fit for machine needs.
Q3: How does the x402 protocol relate to the HTTP 402 status code?
The HTTP 402 status code (Payment Required) has been dormant since its introduction. Led by Coinbase, the x402 protocol reactivates this status code, allowing servers to return machine-readable payment information in response to requests, enabling atomic binding of payments and requests.
Q4: What is the current market size for AI agent payments?
According to data from Circle, AI agents completed 140 million payments totaling $43 million over the past nine months, with more than 400,000 AI agents possessing purchasing power. Of these transactions, 98.6% were settled in USDC, with an average transaction size of $0.31.
Q5: What are the main risks associated with AI agent payments?
Key risks include: regulatory uncertainty (divergent attitudes toward stablecoins across countries), technical security risks (smart contract bugs, cross-chain bridge attacks), ambiguous responsibility (who is liable for AI agent errors), and market liquidity risks (limited on-chain trading depth for RWA assets).
