The payment moment of AI agents: Who will become the Stripe of the machine economy?

Author: Yiping & Turbo, IOSG

Core Narrative

• Agent Payment is transitioning from PoC to the infrastructure competition stage

• x402 processed 3.3 million transactions in 30 days, with an ATV of $0.46 (Visa averages around $50). Estimated real monthly transaction volume for Agent payments is less than $30M.

• TradFi giants are accelerating: Visa launched Intelligent Commerce + Trusted Agent Protocol, Mastercard will open Agent Pay to all U.S. cardholders in November 2025, and Stripe partnered with Tempo to launch MPP on March 18, 2026.

• Strong M&A signals: 7 mergers totaling $8.05B were completed between 2025-2026 (Capital One acquired Brex for $5.15B, Mastercard acquired BVNK for $1.8B, Stripe acquired Bridge for $1.1B). Giants are choosing to buy rather than build.

• The Facilitator layer is a highly investable ecological niche at present. Its position is similar to Stripe in the early days of e-commerce, connecting protocols above and applications below.

• Facilitators directly control the Agent's signing keys and spending strategies, serving as an indispensable trust anchor. They earn both custody fees and order flow revenue, making them the most profitable role in the entire stack.

• MCP (Model Context Protocol) is becoming the standard interface for Agents to call payment tools. Whose payment MCP server is integrated by Claude, ChatGPT, and Cursor by default will hold a position similar to "Chrome's default search engine."

• Cryptographic infrastructure and card organizations are not mutually exclusive; the winner will be a unified gateway that connects both tracks.

• A shopping Agent needs ACP (Stripe) for merchant checkout + x402 for API micropayments + AP2 (Google) for authorization auditing. No single protocol can cover all scenarios.

• Stripe's MPP will go live in March 2026, for the first time supporting both stablecoins (Tempo chain) and fiat (Stripe SPT) within a single protocol. Partners include Visa, Mastercard, Anthropic, OpenAI, and Shopify. This is the first product signal of the integration trend.

• Protocol-driven markets push value upward; giants will not monopolize.

• x402 and MPP are becoming open, commoditized infrastructure. Visa and Stripe will dominate the clearing and settlement and card network sides. The identity layer, Agent app store, wallet strategy engine, and credit infrastructure are still vacant.

1. Market Overview

What is Agent Payment? Agent Payment refers to AI agents autonomously holding funds, authorizing expenditures, and completing transaction settlements without direct human operation. It is not as simple as having the Agent "click the payment button." Achieving this requires a complete set of financial infrastructure from identity verification, wallet management, spending strategies to clearing and settlement, allowing the Agent to become an independent economic entity.

The premise of traditional payment systems is that both parties are KYC-verified humans with bank accounts. Agents break this premise: they have no ID, no bank account, and no credit history, yet they need to purchase API calls, pay for cloud computing, buy data, and even place orders on Amazon on behalf of users. The mismatch in underlying architecture has given rise to the entire Agent payment track.

Three Core Models The core processes of Agent Payment are divided into three types:

Tokenized Card (Virtual Card). Agents obtain a virtual Visa/Mastercard number with spending limits, merchant category restrictions, and validity periods through APIs, with transactions settled via traditional card networks. Ramp Agent Cards, AgentCard.sh, and Slash use this model. The advantage is that merchants do not need to change anything; the downside is that they must be linked to human accounts, and card organizations charge a 2-3% fee.

x402 Stablecoin (HTTP Native Micropayments). The server returns an HTTP 402 status code along with payment conditions (wallet address, chain, amount), and the Agent's Facilitator automatically signs to complete the on-chain USDC transfer, attaching the transaction hash as proof in the request header. No API key, no account, no human approval is needed; the transaction cost is only L2 gas (about $0.001 per transaction on Base).

Session-based Streaming (MPP Model). The Agent pre-authorizes a spending limit and can continuously spend during the session without settling each transaction on-chain, settling all at once at the end of the session. This is suitable for high-frequency scenarios with hundreds of API calls in a single session. Stripe MPP and Tempo chain use this model.

How Agents Pay Regular Bills For regular bills like SaaS subscriptions, cloud services, and data sources, Agents currently have two paths:

1. Using Cards. Generate virtual cards through Ramp Agent Cards or Slash, linking them to SaaS platforms. Corporate finance sets monthly limits and merchant whitelists, allowing the Agent to automatically renew within the authorized range. Traditional vendors like AWS, Google Cloud, and Notion can all be used.

2. Using x402. For vendors that support x402 (Neynar, Hyperbolic, Token Metrics, etc.), Agents pay per call without prepayment or subscription, automatically settling USDC micropayments for each request. The problem is that very few vendors support x402, concentrated in the adjacent crypto service field.

Market Size

Let's take an honest look at the scale: from $6.3M at the beginning of 2026, annualized at about $126M, compared to Visa's transaction volume of $14.6T in 2024, it doesn't even account for the change. However, x402's ATV has climbed from an early $0.09 micropayment to $0.46 (validated by Artemis data). It is still within the micropayment range, and the commercial inflection point has not yet arrived. The market is extremely early, but the economic foundation has already been laid.

Favorable Factors

• TradFi legitimization (very strong). Visa launched "Agentic Ready," Stripe collaborated on MPP, and Mastercard and AmEx joined the x402 foundation. Visa's CPO stated this is "the biggest thing since e-commerce." The market has been validated as real, and investment risks have decreased.

• Acceleration of protocol standardization (very strong). The x402 foundation has migrated to the Linux Foundation, with over 20 founding members including Visa, Stripe, Google, AWS, and Microsoft. Adoption resistance is disappearing, and x402 is becoming a standard at the HTTP level.

• AWS is building production-grade infrastructure (very strong). Amazon Bedrock AgentCore has been delivered and natively integrates x402. CloudFront + Lambda@Edge provides reference implementations for merchants. End-to-end Agent-to-Merchant payment loops will be completed on AWS by March 2026. AWS will provide reference architectures, and enterprises will follow suit.

• MCP service explosion (strong). Over 11,000 MCP servers, with less than 5% monetized. ToolOracle has successfully monetized x402 on 73 servers / 708 tools. This creates natural pull for payment infrastructure.

• Explosion in the number of AI agents (strong). Over 1 million registered agents (2026), with all major LLMs pushing Agent capabilities. Timeline: 12-24 months.

• Accelerated penetration of stablecoins (strong). Total market cap of $246B (2025). Stripe, Visa, and MC are all integrating USDC. This is already happening.

• Decline of subscription models (medium). Developers providing skills/data need to pay per use. Timeline: 12-24 months.

• Clarification of regulations (medium). The EU's MiCA is being implemented, the U.S. stablecoin bill is advancing, and the CFTC chair has stated "AI needs blockchain." This will unlock institutional adoption. Timeline: 12-24 months.

Target Users Agent payment infrastructure services five types of buyers, each with different pain points, willingness to pay, and purchasing authority. Currently, the strongest intent is among three types of people: AI application developers (cannot launch Agent products without payment), corporate finance teams (compliance-driven, budget-controlled), and skill/data providers (the gap in pay-per-call directly stifles their monetization). There is a real but immature flow of funds between consumers and Agents (M2M), with low willingness to pay in the short term.

Core Institutional Players and Merchant Reach Agent Payment is primarily driven by 8 institutions, including two crypto-native players (Coinbase, Circle), three card organizations/payment giants entering the market (Stripe, Visa, Mastercard), one AI platform (Google), and two companies aggregating at the upper layer (Crossmint, Tempo).

Reaching merchants presents a chicken-and-egg problem. Card networks have overwhelming merchant coverage (Visa 150 million+, MC 100 million+), requiring no changes from vendors. x402 has only about 50 crypto/AI services. Without more vendors, transaction volume won't increase; without transaction volume, vendors won't onboard. Stripe MPP breaks the deadlock by leveraging existing merchant relationships (SDK upgrades rather than new integrations); Crossmint breaks the deadlock by aggregating both tracks through a single API.

Current Unresolved Issues

• Security threat models are new and unresolved.

• Key threats include prompt injection, Agent behavior out of control (recursive loops exhausting budgets), key leakage, Agent impersonation, and third-party SDK supply chain risks.

• The most dangerous failure is not unauthorized access but abuse after authorization.

• A strategy engine at the infrastructure layer is necessary, but most wallets lack one.

• Lack of standardized Agent identity.

• There is no reliable way to verify who the Agent is, what permissions they have, or whether they have been compromised.

• ERC-8004 has been deployed to the Ethereum mainnet, containing three types of registries (Identity, Reputation, Validation based on ERC-721), but adoption is still early.

• NIST has accepted proposals on AI Agent Identity and Authorization (April 2026). EIP-11419 proposes adding an Agent Permission Validator to modular smart accounts.

• Without identity, each Agent transaction relies purely on trust.

• Lack of dispute resolution mechanisms.

• Stablecoin payments are designed to be fast and irreversible, with no chargebacks, no banks to complain to, and no recourse mechanisms.

• Smart contract custody and on-chain reputation systems are being explored, but neither has been standardized or reached production level.

• There is no clear error handling, overpayment, or fraud response framework, which prevents large-scale adoption by institutions.

• Compliance infrastructure is immature.

• An increasing number of jurisdictions are applying the Travel Rule (FATF) to stablecoin transfers.

• KYC, AML, sanctions screening, and audit trails are not optional for financial applications, but most Agent payment tools treat compliance as an afterthought.

• Teams that did not incorporate compliance into their design from day one will face very high retrofitting costs later.

• Cross-chain complexity.

• Agents need to operate across multiple networks (Base, Solana, Stellar, Canton, etc. permissioned chains).

• Regardless of where transactions are settled, strategy execution must be consistent.

• No single chain has won the Agent payment race, so infrastructure must be cross-chain, which will increase engineering and security overhead.

2. Track Landscape and Value Chain

Agent Payment is not a single market but an ecosystem with seven layers.

Facilitators (L2) and wallets (L1) capture a disproportionate amount of value because they control the Agent's "private keys." Whoever holds the keys owns the economic sovereignty of the Agent. The protocol layer (L0) does not directly generate revenue as an open-source standard, but companies setting the standards (Coinbase through x402, Stripe through MPP) can indirectly monetize through surrounding facilitator services. This is similar to internet history: HTTP is free, but companies like Cloudflare and Akamai that control HTTP traffic entry points are worth billions.

3. Track In-Depth Analysis

Payment Protocol (L0) x402 The situation with x402 is somewhat complex, with the Base chain accounting for most transactions.

• Daily active data (average for March): 110,000 transactions, approximately $51K in transaction volume.

• Base is dominant: 82% of transaction count and 99% of transaction volume on Base.

• Leading Facilitators: Coinbase Global first (41%), PayAI second.

• A significant proportion of wash trading: 36% of x402's March transactions were wash or incentive-driven, publicly reported transaction counts overestimate real Agent demand.

x402 Ecosystem Data (Artemis, April 2026)

• Supported chains: Base, Ethereum, Polygon, Solana, Avalanche, Sui.

• The x402 foundation is co-governed by Coinbase and Cloudflare (established in September 2025) and has now migrated to the Linux Foundation, with over 20 founding members.

• Stripe integrated x402 on Base in February 2026.

• Minimum viable payment: $0.001.

• End-to-end payment time: approximately 2 seconds.

• Cumulative sellers over 5 months: approximately 2,300.

5-Step Payment Process

1. User/developer funds the Agent's strategy.

2. The Agent requests from the vendor's API and receives an HTTP 402 response (including merchant wallet, supported chain, asset type, price).

3. The Facilitator verifies whether this payment is within the spending strategy authorized for the Agent.

4. If approved, the Facilitator executes the on-chain USDC transfer.

5. The Agent attaches the transaction hash as proof of payment in subsequent requests; the vendor verifies and provides the service.

Merchant coverage is currently the biggest limitation: Neynar, Hyperbolic, Token Metrics, Pinata (IPFS), Heurist, Prodia (image generation), Firecrawl (web crawling). Almost all are crypto or AI-native services. Traditional e-commerce (Amazon, NYT) has not yet integrated.

Traditional e-commerce (Amazon), mainstream SaaS (Notion, Slack, AWS), and content platforms (NYT, Spotify) have not integrated x402 at all. What Agents can do on x402 is very limited: buy GPU computing power, adjust APIs, store files. Tasks like placing orders on Amazon, renewing Notion, and paying Uber can still only go through card organizations.

Vendor onboarding is widely seen as the last and most challenging part of the entire Agent payment stack. The API proxy model (Agents calling restricted APIs on behalf of users) may violate vendor ToS, introducing additional legal risks.

Early concerns focused on whether the $0.09 ATV could support the facilitator's P&L; the bottleneck remains the economics of micropayments combined with the breadth of vendor coverage. MPP (Machine Payments Protocol) MPP has just launched but is growing rapidly, reaching 2.3K sellers in just 5 days.

MPP, launched by Stripe and Tempo, allows any client (Agent, application, or human) to pay for any service's fees in a single HTTP request. Developers use MPP to have their Agents pay service fees, while service operators use MPP to accept API payments.

• Daily active data: 4.7K transactions, $201 in transaction volume.

• x402 took 5 months to reach 2.3K sellers; MPP only took 5 days.

Architecture

• Session-based: Agents pre-authorize spending limits, stream micropayments within the session without needing to settle each on-chain.

• Settles through the Tempo chain (bridged $5B), with sub-second confirmations.

• Simultaneously supports Stripe SPT (fiat), Visa cards, stablecoins, and Bitcoin (via Lightspark).

• Over 100 vendors integrated on launch day.

The strategic significance is that MPP is the first substantial integrated product in the crypto vs card battle. Stripe's distribution capability (millions of merchants globally) combined with Tempo's stablecoin settlement efficiency may create a dual attack on pure crypto solutions (x402) and pure card solutions (Visa IC). Risks Having been live for only a few weeks, there is no production-level data. The Tempo chain itself is new, and the ecosystem has not yet been validated. Comparison of x402 vs MPP

Integration Trends They are converging, not competing.

• Stripe is a founding member of the x402 foundation, and MPP explicitly supports both stablecoins and cards.

• Visa is betting on both sides. It contributes card track specifications to Stripe's MPP while advancing its own Intelligent Commerce and Trusted Agent Protocol. Framing x402 and MPP as opposing camps ignores the fact that the largest card network is a design partner for both.

The architectures are complementary:

• x402 handles payment negotiation at the HTTP layer: how the server tells the client "give me money" through the 402 status code.

• MPP handles transaction execution at the session level: how money actually moves, folding unlimited micropayments into two on-chain transactions (opening + settlement).

• The session model directly addresses the scalability issue of micropayments. Rather than pursuing 12 million transactions per second at $0.09 each, it is better to batch thousands of micropayments into a single settlement.

Stripe's distribution channel allowed MPP to match x402's seller count in 5 days, validating the judgment that "distribution > protocol." Visa Intelligent Commerce Visa announced the Intelligent Commerce framework in April 2025, launched "Agentic Ready" in Europe in March 2026, and released the AI Agent Developer SDK on April 2, 2026.

Core components:

• Trusted Agent Protocol (TAP): distinguishes legitimate Agents from malicious bots.

• Tokenized credentials: AI-ready card credentials with spending limits, merchant categories, and approval requirements.

• Pilot partners: Ramp, Skyfire, and other undisclosed parties.

The biggest advantage is merchant coverage: the Visa network covers over 150 million merchants globally, allowing Agents to use a Visa card number to spend on Amazon, Uber, or any SaaS platform without requiring any changes from vendors.

The biggest disadvantage is that it must be linked to human accounts. Visa's trust model is "backed by KYC-verified humans," which fundamentally conflicts with the long-term vision of autonomous Agent economies. Other Protocols

• ACP (Agentic Commerce Protocol): designed for instant checkout in conversational interfaces (like within ChatGPT). It targets the consumer checkout layer, not the API settlement layer. ACP is complementary to x402.

• UCP (ATXP's Unified Commerce Protocol): attempts to unify all Agent payment protocols under a single interface.

• MoonPay Agents: bridges traditional checkout flows and AI agents, transforming human checkout processes into programmatic payments executable by Agents via APIs.

Wallet and Key Management (L1) Over a dozen wallet providers are competing in this market, resembling the early mobile wallet landscape before Apple Pay.

Use cases:

• Lending and credit: AI-driven underwriting is entering consumer-level crypto lending. 3Jane fully automates credit underwriting through smart contracts, setting interest rates and executing debt contracts using verifiable financial records without manual review.

• Creator and gig economy settlements: Agents handle routing, wallet management, and currency conversion across platforms. Audius directly allocates 90% of revenue to artists in real-time when content is consumed, with no monthly settlement cycles and no intermediary cuts.

• Fund management: Agent-based funding systems infer across real-time market conditions, rebalancing positions in real-time, and enabling cross-border settlements without waiting for business hours, deploying idle funds into interest-bearing tools.

Facilitator Layer (L2) The Facilitator layer sits between the protocol (x402, MPP) and applications. Coinbase Global remains the largest Facilitator by cumulative volume (41% of all x402 transaction volume, according to Artemis).

Why this layer is the monetization layer of the Agent economy: Agents must pay to buy things, and the Facilitator is where that money is actually settled. Model companies are unlikely to do this themselves, as they won't run GTM for long-tail scenarios, leaving monetization opportunities to independent operators.

Facilitator startups

Other facilitators (open-source tools, non-funding startups): x402-rs (Rust library), OpenX402 (unlicensed facilitator), OpenFacilitator (free shared endpoints), B402 (BSC-specific fork), CodeNut (Agent infrastructure), RelAI (x402 API marketplace), AurraCloud (decentralized computing power, AURA token).

Use Cases

• Pay-per-query data access: Currently the largest transaction volume facilitator use case. Transaction Agents need real-time market data, compliance Agents need sanctions screening, and credit Agents need credit checks. Facilitators allow these Agents to pay per request without subscriptions, API keys, or vendor contracts. Spraay has already provided 70 x402 endpoints covering oracles, analytics, AI inference, and search, with single-call costs ranging from $0.001 to $0.10.

• Developer-facing API monetization: Facilitators abstract away blockchain interactions, allowing any developer to gate their API with x402 payments without running nodes or understanding crypto. AWS CloudFront + Lambda@Edge reference architecture enables any HTTP application to activate x402 at the edge.

• Subscription management: Agents autonomously handle cancellation processes, dynamically pricing retention offers based on usage history. As software shifts to pay-per-use pricing, continuously optimizing the Agent's value in your payment costs will significantly enhance.

• Cross-chain payment routing: Facilitators handle swaps, bridging, and settlements, allowing Agents to pay with any token on any chain, with merchants receiving the assets they desire. AnySpend supports 19+ networks. This is a pipeline that neither Agents nor API providers want to handle themselves.

Tokenized Card (L4: Governance & Strategy / Identity & Authorization, Virtual Card) Virtual Card Issuance Process

• Card program setup: Platforms (like Ramp, AgentCard.sh) establish virtual card programs through issuing partners (Visa/MC issuers).

• API creates cards: Developers generate virtual cards for each Agent or spending scenario through APIs, setting parameters:

• Spending limits (per transaction / daily / monthly)

• Merchant Category Code (MCC) whitelists/blacklists

• Validity period (one-time or long-term)

• Geographic restrictions

• Agent receives card number: The Agent obtains a 16-digit card number + CVV + expiration date, which can be used at any merchant accepting Visa/MC.

• Transaction authorization: When the merchant initiates a transaction, the card network verifies in real-time according to preset strategies.

• Settlement: Cleared through traditional card networks (T+1 or T+2), deducting from the corporate funding account.

Comparison of Major Card API Providers

Core Limitations of Card Models

1. Must be linked to a parent account: All Agent cards ultimately need to be linked to KYC-verified human/business accounts as funding sources.

2. Fees: Card networks charge 2-3% interchange, which is not economical for API micropayment scenarios.

3. Settlement speed: T+1 to T+2, which cannot meet the real-time settlement needs of Agents.

4. Limited merchant control: Agents may be incorrectly flagged as fraudulent.

Identity and Reputation (L4: Governance & Strategy / Identity & Authorization Identity Side)

Identity is infrastructure rather than an independent use case; it supports every other layer.

Skill Discovery and Store (L5)

Use cases:

• In-game rewards: Web3 gaming platforms deploy Agents to manage in-game economies, distribute rewards, and handle asset transactions. Virtuals Protocol has already tokenized AI agents into game NPCs, trading bots, and research assistants, allowing the community to co-own and govern.

Agent Coordination (L6)

Use cases:

• Agentized trading: The shift from algorithmic trading to Agentized trading changes the competitive unit from latency to intelligence. Classic algorithmic trading: execute Y when price crosses X. Agentized trading: reasoning across market conditions, liquidity, risk parameters, and portfolio positions to determine optimal actions.

• Agent groups: The next stage is coordinated groups of Agents. A financial Agent executing trades works in real-time with compliance and risk Agents to verify, flag, and audit.

Data and Compliance (L7)

TRES Finance, Chainalysis, and Allium are also laying out in this layer, but they come from a broader blockchain analysis background.

Compliance Agent teams: Institutions deploy compliance Agents as parallel labor forces to monitor transaction flows in real-time, flag anomalies, run sanctions screenings, and autonomously generate regulatory reports.

Crypto-native vs Card Organization Battle Crypto-native Camp Stablecoins are the "native currency" for Agents for three reasons:

1. Expanded trust structure: Stablecoin wallets can be linked to anything, social accounts, domain name servers, unattended smart contracts. Agents outside the traditional financial system can also transact.

2. Internet-native global settlement: Agent workflows across U.S. LLM endpoints, European data vendors, and Southeast Asian computing clusters should not require three separate payment tracks.

3. Cost structure: Each gas fee on x402 on Base is about $0.001, compared to 2-3% interchange on card networks. Even if x402's ATV rises to $30, stablecoin gas costs will still be two orders of magnitude cheaper.

Card Organization Camp (represented by Visa / traditional FinTech) Agent cards can be used right now for three reasons:

1. Merchant coverage: Over 150 million merchants already accept Visa/MC, requiring no changes.

2. Consumer protection: Chargebacks, fraud detection, and dispute resolution represent 50 years of accumulated infrastructure. Stablecoin transactions are irreversible.

3. Compliance maturity: PCI DSS, KYC/AML, and consumer protection legal frameworks are mature.

Pragmatic Conclusion

• Short-term (1-2 years): Card tracks dominate. Stablecoins are limited to crypto-adjacent API micropayments.

• Mid-term (2-4 years): Integration. Stripe MPP has already proven that a single protocol can simultaneously support stablecoins and fiat.

• Long-term (5+ years): If stablecoin regulations are implemented and merchant acceptance increases, the crypto track could become the default.

Framework Payment Support and MCP Current Framework Integration Status Currently, no mainstream AI framework has built-in native payment capabilities; all frameworks integrate payments through external tools (mainly MCP servers).

MCP is the De Facto Standard MCP is rapidly becoming the universal interface standard for Agents to call external tools. Microsoft uses MCP in Copilot, and all major Agent frameworks support it.

Published payment MCP servers:

• ATXP: 14+ tools (payment_make, web_search, web_browse, etc.), supporting Claude, LangChain, CrewAI, OpenAI SDK.

• FluxA: fluxa-agent-wallet (x402 payment + USDC withdrawal + payment links) and fluxA-x402-payment skill, listed on LobeHub.

• Clink: clink-mcp-server, open-source TypeScript implementation.

• PayMCP: vendor-agnostic payment layer for MCP tools (MIT open-source).

• Ramp: Ramp MCP integration is available on Composio.

• AgentPay (OpenClaw): agentpay skill, supports wallet shopping requiring human approval.

Strategic implication: Whose payment MCP server becomes the default configuration for mainstream clients like Claude Desktop, ChatGPT, and Cursor will gain the "default entry" for Agent payments. This is akin to Google paying Apple $26B annually to become the default search engine for Safari. ATXP currently leads in framework coverage, but Coinbase (through CDP MCP server) and Stripe (through MPP) have platform-level distribution advantages.

4. Competitive Landscape and Moat

Sub-track winner-takes-all analysis

Moat strength shows a bimodal distribution. L4 card governance (Visa/MC duopoly) and L3 routing (Circle + Bridge) have already been locked in by network effects. L1 wallets have real switching costs and are moving towards centralization. L2 Facilitators and L4 identity sides are the battlegrounds where startups can achieve real returns. Upstream and Downstream Opportunities

Industry Lifecycle

Lifecycle positioning is early to mid-stage. It is expected to enter the early growth phase within 12-18 months. Two markers: standards converge to 1-2 mainstream protocols, and at least one Agent payment project surpasses $10M in monthly transaction volume.

5. Investment Analysis

7 Powers Framework

The most critical Power currently is reverse positioning. In the early stages of the industry, startups can only leverage reverse positioning and network economics. Economies of scale and brand naturally belong to giants. Visa cannot fully embrace stablecoins, or it will lose $32B in interchange revenue annually, which is the only structural advantage window for startups.

Power evolution prediction: If Visa adapts to stablecoins (through VTAP) within 2-3 years, reverse positioning will disappear, and switching costs will become the only remaining Power for startups. This means that the most investable targets now are those facilitators that can build high switching costs during the reverse positioning window, which means deep API integration + key custody + spending strategy locking. Sub-track Investability

Investment Priorities (High to Low)

• Facilitator Layer (value capture, score 8/10)

• The value of Agent payments does not belong to the protocol layer but to those who find real use cases and serve real users. Facilitators completely shield the complexity of chains and Agents.

• x402 and MPP are open commoditized tracks. Facilitators sit between protocols and users, handling payment verification, on-chain settlements, and cross-chain bridging.

• Control over Agent signing keys and spending strategies (indispensable trust anchor). They earn both custody fees and order flow revenue.

• Clear M&A exit paths, benchmarked by Stripe's $1.1B acquisition of Bridge.

• Key to success: solid ground game in a specific vertical (prediction markets, pay-per-query data, API monetization). Achieve chain independence as early as possible. Build developer-friendly SDKs. Compete on reliability and settlement speed, not price.

• L4: Governance & Strategy / Identity & Authorization Identity Side (highest alpha, score 7/10)

• The trust layer for Agent commerce is completely missing. There is no standard way to verify who the Agent is, what permissions they have, or whether they are trustworthy.

• ERC-8004 and Metaplex Agent Registry are early but credible. ZKID prototypes are expected to support Agent verification under privacy protection.

• NIST has begun to intervene in AI Agent Identity and Authorization, indicating this will become a regulated category.

• Whoever secures the trust graph will become the default identity layer, winner takes all.

• Key to success: build cryptographic identity (signature credentials binding Agents to principals + scope of authority), not just OAuth wrappers. Capture the trust graph early to trigger network effects. Integrate at the wallet/infrastructure layer, where prompt injection cannot cover.

• L6: Agent Coordination (score 7/10)

• The next stage is coordinated groups (financial + compliance + risk Agents working together).

• Key to success: build cryptographic verification of Agent outputs.

• L7: Data and Compliance (score 6/10)

• Audit trails themselves are a dispute resolution mechanism.

• Key to success: achieve real-time cross-chain transaction reconstruction. Directly embed Travel Rule compliance into payment flows.

• L5: Skills Discovery and Store (score 6/10)

• Over 11,000 MCP servers, with monetization rates below 5%. This is the "App Store" moment for Agent capabilities.

• Whoever becomes the default discovery layer will simultaneously control routing and payments, akin to the Google + Stripe merger position.

• Key to success: aggressively aggregate supply, creating a payment-native discovery mechanism.

• L1: Wallet and Key Management (score 7/10)

• 10+ players, but may quickly centralize.

• Fleet management (Sponge) and framework-agnostic support (LobsterCash/Crossmint) are differentiators.

• Key to success: secure framework-level defaults for LangChain, CrewAI, Claude Code. Release a strategy engine with "five pillars," including spending limits, counterparty whitelists, transaction type restrictions, time dimension control, and upgrade thresholds.

Unit Economics (Facilitator Layer) Modeling P&L for a typical Facilitator startup across three stages:

ATV sensitivity analysis (Y1, 500 Agents x 20 transactions per month):

ATV is the lifeblood of the entire business model. The era of $0.09 micropayments is over. Agent shopping/autonomous procurement is becoming the primary use case. The next inflection point is when ATV rises from $30 to $50+. Leading indicators: which payment MCP server becomes the default integration in Claude Code, LangChain, CrewAI.

Minimum viable transaction volume test. At a 0.5% take rate, a Facilitator needs $200M annual GTV to achieve $1M ARR, which is about $550K/day. The entire x402 ecosystem currently has a daily GTV of about $2.7M (Artemis, April 2026), meaning a theoretical ARR ceiling of about $4.9M (if a certain Facilitator captures 100%):

For the entire x402 ecosystem, at a 0.5% take rate, daily GTV of $2.7M:

• Annual Facilitator revenue: $2.7M x 365 x 0.5% = $4.9M

• Has already surpassed the $1M ARR threshold.

• $10M ARR: requires double the current transaction volume.

• $100M ARR (growth phase): requires twenty times the current transaction volume.

Take Rate benchmarks:

Mature company comparisons

Publicly traded payment companies transact at 12-25x EV/Revenue. If Agent payment startups can achieve $50M+ ARR with 50%+ growth, they could be valued at least $1B+ at a 20x revenue multiple. However, currently, no company in the entire industry has disclosed revenue data, and valuations are entirely based on narrative premiums. Forecast Decision Tree Core Question Can Agent Payment reach $1B in annual transaction volume by 2028? This leads to two paths. Path 1: Yes, it can be achieved (55% probability). The triggers are Stripe MPP validating PMF and Visa Agent cards covering millions of users. Under this path, there are two sub-results:

• The stablecoin track becomes the default (30% of this branch). Coinbase and the x402 ecosystem capture the most value, bypassing parts of the card network, with Facilitator valuations falling in the $500M to $1B range.

• Card channels remain dominant, with stablecoins as M2M supplements (70% of this branch). Visa, Mastercard, and Stripe win mainstream volume. Pure crypto solutions become niche, and Facilitators will either be acquired by Stripe or marginalized.

Path 2: No, it cannot be achieved (45% probability). The catalysts are Agent reliability failing to meet payment-level trust or standards continuing to split. Two sub-results:

• Slow growth to $200M to $500M (60% of this branch). The track exists but valuations are under pressure, and startups will need longer runways.

• Basic model companies build their own payments (40% of this branch). OpenAI and Google integrate payments natively, eliminating third-party Facilitators.

Growth Timeline

Reverse Stress Test # Risk 1: Giants build their own payments, middleware becomes worthless. OpenAI, Google, and Apple control over 90% of AI agent user entry points and can natively close the payment loop (ChatGPT + card binding, AP2 + Google Pay, Siri + Apple Pay).

When Google AP2 launched, it claimed "fully closed-loop Agent payments within the Google ecosystem" with 60+ partners. OpenAI Operator can already complete web shopping. Historically, Apple Pay has eliminated many third-party mobile wallets.

# Risk 2: Market timing is still 3 to 5 years off; investing now is too early. Agent unreliability, lack of vendor API standards, and insufficient consumer trust are all hard barriers. Seed companies have an 18 to 24-month runway, and the market may not catch up.

The AI payment protocol supported by Coinbase faces the narrative of "demand just isn't there yet." Agents frequently lie while executing tasks. Most Agents still do not earn a dollar.

# Stress Test Conclusion Among the two risks, timing risk is the most fatal and hardest to refute. Unit economics arithmetic does not lie; the market is indeed far from an investable scale. Platform risk is somewhat mitigated by "model companies not being good at compliance," but this defense line is weakening in 2026: OpenAI has already acquired a KYC company, Google has Google Pay, Apple has Apple Pay plus Apple Card, and traditional financial investors are also on Anthropic's shareholder list. Compliance capability is no longer a moat that makes model companies trustworthy.

The single largest and non-reducible risk is timing. The leap from the introduction phase to the early growth phase depends on ATV moving from micropayments to commercial levels, which in turn depends on two external variables that investors cannot control: Agent reliability and merchant coverage.

Three adjustments to investment strategy:

1. Allocate 60% of Agent payment exposure to seed rounds, reserving ample follow-on for bridge rounds (to hedge timing risk).

2. Prioritize facilitator tracks that are agnostic (supporting both stablecoins and cards) to hedge regulatory risks.

3. Set an 18-month kill switch: if no Facilitator exceeds $5M in monthly transaction volume by Q4 2027, consider write-downs or low-price disposals.

Investment Recommendations The need for Agents to have payment capabilities is a logical necessity, but the current market is extremely early (only $6.3M since early 2026), with severe wash trading, standard fragmentation, and giants potentially crushing startups at any moment. The argument is not "this market is big now," but "this market will grow, and the valuation window is friendly before it arrives."

Geographically, the focus is on the U.S., with Europe serving as a compliance hedge and Asia as a wildcard.

Investable teams should have payment industry DNA (Stripe, Coinbase, Visa) or a background in crypto infrastructure, have deployed MCP servers, and have at least one framework integration, with real non-wash monthly transaction volumes of $100K+ or 100+ developer integrations, along with a clear developer-first GTM. Points to avoid include pure AI backgrounds without payment experience, projects with only white papers, transaction volumes entirely reliant on incentives, parallel B2C and B2B approaches, and complete disregard for MSB/EMI licenses.

Source

1. AgentPaymentsStack.com: 162+ Projects, 6 Layers

   https://agentpaymentsstack.com/

2. HTTP 402's Modern Makeover

   https://panteracapital.com/http-402s-modern-makeover/

3. Building Permissionless Neobanks

   https://panteracapital.com/building-permissionless-neobanks/

4. Crossmint: Agent Card Payments Compared

   https://www.crossmint.com/learn/agent-card-payments-compared

5. WorkOS: x402 vs Stripe MPP

   https://workos.com/blog/x402-vs-stripe-mpp-how-to-choose-payment-infrastructure-for-ai-agents-and-mcp-tools-in-2026

6. ATXP: Which AI Frameworks Support Agent Payments

   https://atxp.ai/blog/which-ai-frameworks-support-agent-payments/

7. Tiger Research: AI Agent Payment Infrastructure

   https://reports.tiger-research.com/p/aiagentpayment-eng

8. Visa: Agentic Ready Programme (Europe)

   https://www.visa.co.uk/about-visa/newsroom/press-releases.3437886.html

9. CoinDesk: Coinbase-backed AI payments protocol demand not there yet

   http://www.coindesk.com/markets/2026/03/11/coinbase-backed-ai-payments-protocol-wants-to-fix-micropayment-but-demand-is-just-not-there-yet

10. Galaxy Research: Agentic Payments: x402 and AI Agents

    https://www.galaxy.com/insights/research/x402-ai-agents-crypto-payments

11. Stripe Machine Payments Protocol announcement

    https://stripe.com/blog/machine-payments-protocol

12. x402 Foundation launches under Linux Foundation

    https://thedefiant.io/news/infrastructure/coinbase-x402-payment-protocol-moves-to-linux-foundation

13. The Biggish: AI Agent Payment Infrastructure: The $1T Gap

    https://thebiggish.com/news/ai-agent-payment-infrastructure-what-works-what-doesn-t-and-the-1t-gap

14. Agentic Payments Landscape: MPP, x402, ACP, AP2 compared

    https://www.openfort.io/blog/agentic-payments-landscape