In the span of six weeks, the agent payment landscape split into two clear camps.
On June 10, Mastercard launched AP4M (Agent Pay for Machines) — a platform that extends its 50-year-old card network into machine-to-machine commerce, with 30+ partners including Stripe, Coinbase, OKX, and Adyen.
On July 14, the Linux Foundation launched the x402 Foundation — an open protocol that revives the HTTP 402 "Payment Required" status code as a crypto-native settlement layer, backed by ~40 founding members including Visa, Mastercard, Stripe, AWS, and Google.
Both are real. Both have major partners. Both are processing real volume. But they represent fundamentally different bets on how AI agents will pay each other.
| Dimension | Mastercard AP4M | x402 Protocol |
|---|---|---|
| Launch | June 10, 2026 | May 2025 (Coinbase/Cloudflare incub.); July 2026 (Linux Foundation) |
| Settlement | Multi-rail: cards, bank accounts, stablecoins | USDC-native (98.6% of volume); Base, Solana, Polygon |
| Identity | Verifiable Intent framework + Agentic Tokens | Wallet address + optional on-chain identity |
| Permissioning | Programmatic spending rules, category limits, time-bound budgets | Smart-contract allowances; protocol-enforced guardrails (Casper facilitator) |
| Fee model | Card-network interchange (~1-3%) | Blockchain gas + optional platform fee (minia2a: 5%) |
| Governance | Mastercard-controlled | Linux Foundation open governance |
| Key partners | Stripe, Coinbase, OKX, Adyen, Ripple, Solana Foundation, Cloudflare | Visa, Mastercard, Stripe, AWS, Google, Circle, Coinbase, Cloudflare |
The most revealing data point: Visa and Mastercard joined both. They're hedging. The card networks don't know which rail wins — so they're building on both.
The x402 protocol has one architectural advantage that's hard to replicate: it's an HTTP status code.
When an agent hits an API endpoint and gets back 402 Payment Required with a payment header, the entire transaction — discovery, negotiation, payment, and service delivery — happens in one HTTP round-trip. No redirect to a checkout page. No human in the loop. No OAuth flow.
This is why it's processed 109.6 million transactions. At fractions of a cent per call, the economics work for the agent use case in a way card-network interchange never will. Visa's own report with Artemis confirmed: 98.6% of machine-native payments settle in USDC. Cards weren't designed for $0.005 API calls.
The Casper Network launch on August 3 added another piece: protocol-enforced spending guardrails. An agent's wallet can be constrained at the protocol level — max spend per call, max per day, allowed endpoints only. This is the accountability primitive the ecosystem needs.
Mastercard's advantage isn't technical. It's institutional.
AP4M comes with built-in compliance infrastructure: KYC, AML, sanctions screening, dispute resolution, audit trails. For any enterprise that needs to answer to a regulator about why an AI agent spent $47.32 at 3:17 AM, AP4M provides that answer. x402 — at least today — doesn't.
The Verifiable Intent framework is genuinely novel: every agent transaction carries a cryptographically signed proof of authorization, including who authorized it, under what spending policy, and for what purpose. This isn't just a payment rail — it's a compliance rail.
And the multi-rail settlement matters. An agent can pay for cloud compute in USDC on Base and settle a supplier invoice through ACH, all through the same credential. That flexibility doesn't exist in crypto-only protocols.
Both protocols have the same unsolved problem: how does an agent find something to pay for?
Mastercard's answer is its existing merchant network — millions of businesses already accept Mastercard. But those merchants have human-facing APIs, not machine-readable x402 headers. Stripe's involvement in both camps suggests the acquirer side will bridge this gap.
x402's answer is directories like minia2a — currently listing 218 services across 160+ distinct endpoints. When an agent needs to solve a captcha, estimate gas, or recall stored memory, it queries the directory and gets back a machine-readable list of services with x402 payment headers.
The irony: Mastercard built the rails for agent-to-agent payments. x402 built the services. The protocol that wins the discovery layer wins the transaction volume.
The trial data from minia2a reveals something the protocol architects might not expect: agents spend more on infrastructure than intelligence.
The most-used x402 endpoints aren't LLM inference or decision engines. They're captcha-solving, gas estimation, web scraping, and memory storage — plumbing. The agent economy, at least in its first 5,800 trials, is an infrastructure economy.
This favors x402's low-fee model. When your service costs $0.001 per call, a 2.9% + $0.30 card interchange makes no economic sense. But when an agent needs to provision a $500 cloud instance, Mastercard's compliance and dispute framework becomes essential.
The smart money — Visa, Mastercard, Stripe, Coinbase, Circle — is betting on both. Here's the likely outcome:
The agent payment stack isn't winner-take-all. It's a spectrum — from sub-cent API calls on one end to enterprise procurement on the other. x402 owns the micro end. Mastercard owns the macro end. The fight is over the middle — and that's where the volume is.
For agent developers today: integrate both. Use x402 for pay-per-call services (start with minia2a's 218-service directory). Use AP4M for anything that needs enterprise compliance. The protocols are complementary, not competitive.