Three competing architectural models are vying to become the default payment infrastructure for AI agents: x402, a stablecoin micropayment layer embedded in HTTP; AP2, an authorization and trust protocol; and card delegation via Mastercard Agent Pay and Visa Agent Scoring. As of October 2026, the...
"AI agents and automated systems are becoming active participants in the global economy, yet they have lacked a native, secure way to transact." — Jim Zemlin, CEO, Linux Foundation (x402 Foundation operational launch, July 14, 2026)
Three competing architectural models are vying to become the default payment infrastructure for AI agents: x402, a stablecoin micropayment layer embedded in HTTP; AP2, an authorization and trust protocol; and card delegation via Mastercard Agent Pay and Visa Agent Scoring. As of October 2026, the combined weekly settlement value across x402 rails totals approximately $368,000 — a figure that is orders of magnitude below the trillions processed annually by card networks. The market is structurally nascent despite projections from Juniper Research of $8 billion in global agentic spend in 2026, scaling to $1.5 trillion by 2030.
Each protocol reflects fundamentally different assumptions about who controls spending authority, where value accrues, and which economic rails should carry machine commerce. x402 disintermediates traditional payment processors by settling directly on-chain with near-zero fees. AP2 is rail-agnostic and concerned exclusively with authorization — it proves a human approved a transaction without moving money. Card delegation preserves existing interchange economics for consumer-facing spend while selectively adopting stablecoin rails for machine-to-machine micropayments. The three models are partially complementary and partially in tension, and the outcome will determine how an estimated $3 trillion to $5 trillion in agentic retail spend (per McKinsey, October 2025) is ultimately settled.
Multiple research firms have published agentic commerce forecasts in 2025–2026. The figures vary widely depending on how "agentic spend" is defined:
| Source | Metric | 2026 | 2030 | |---|---|---|---| | Juniper Research (Apr 2026) | Global agentic spend | $8B | $1.5T | | McKinsey (Oct 2025) | Agentic-orchestrated global retail spend | — | $3T–$5T | | Grand View Research | Agentic commerce market | $7.7B | $65.5B (2033) | | Gartner | Broader agentic AI spending | >$200B | — |
These projections span different scopes. Juniper and Grand View measure direct agentic transaction value; McKinsey measures total retail spend influenced or orchestrated by agents; Gartner captures all agentic AI spending including infrastructure and software, not just payments.
The observable reality is substantially smaller. Coinbase reported 165 million cumulative x402 transactions, 69,000 active agents, and $50 million in total volume as of April 21, 2026. Weekly x402 stablecoin transfers hit 8.7 million in the week of August 17, 2026, but total settlement value was approximately $367,950. This represents a 93% decline in settlement volume from the late-2025 peak of $10.017 million per week, according to analysis by Jamie Coutts at Helios Analytics. Much of the 2025 spike is attributed to developer testing rather than sustained commercial activity.
| Dimension | x402 | AP2 | Card Delegation | |---|---|---|---| | Primary function | Money movement | Authorization/trust | Payment orchestration | | Who moves funds | Blockchain (Base, Solana) | Existing rails (agnostic) | Cards, bank accounts, or stablecoins | | Settlement asset | USDC (98.6% share) | Rail-dependent | Card networks or USDC on Base | | Protocol fee | Zero (gas only) | Zero | Interchange on card-settled txns | | Avg. ticket size | $0.31–$0.48 | N/A | Not disclosed | | Governance | x402 Foundation (Linux Foundation) | FIDO Alliance | Proprietary (Mastercard/Visa) | | Lead org | Coinbase | Google | Mastercard, Visa | | Status (Oct 2026) | 165M+ cumulative txns | v0.2.0 shipped, reference impls | Early developer access (AP4M) |
x402 revives the HTTP 402 "Payment Required" status code, defined in 1991 but never implemented. The protocol flow is mechanical: a server returns a 402 status with payment instructions; the client retries the request with a cryptographically signed stablecoin payment embedded in the HTTP header. Settlement occurs on-chain, primarily in USDC on Coinbase's Base L2 or on Solana.
Transaction economics. Gas costs are approximately $0.00025 per transaction on Solana and a fraction of a cent on Base. There is no protocol-level fee. According to Coinbase data, 87% of x402 payments are valued between $0.0001 and $0.10, confirming the protocol's orientation toward sub-dollar micropayments.
Adoption velocity. x402 crossed 100 million transactions on Base within three quarters, per Chainalysis. Stripe launched x402 integration on February 11, 2026, generating approximately 156,000 weekly transactions in its early phase. Time from wallet funding to first x402 payment fell from a median of 73 days in late 2025 to 18 minutes by Q3 2026. The share of wallets spending within one hour of funding rose from 4% (Q3 2025) to 67% (Q3 2026), suggesting declining friction in onboarding.
Institutional backing. The x402 Foundation launched under the Linux Foundation on July 14, 2026, with 40 member organizations. The 17 premier members include Visa, Mastercard, Stripe, Google, AWS, Cloudflare, Coinbase, American Express, and the Solana Foundation — a coalition that spans card networks, cloud providers, and blockchain infrastructure.
Volume decline. Despite growing transaction counts, settlement value has dropped sharply. The 93% year-to-date decline in weekly settlement volume suggests that much of the activity consists of low-value test transactions rather than commercially meaningful spend.
Google announced the Agent Payments Protocol (AP2) on September 16, 2025, with more than 60 partners including PayPal, Mastercard, American Express, Adyen, Coinbase, Salesforce, Worldpay, JCB, UnionPay, Etsy, Revolut, MetaMask, and the Ethereum Foundation. The consortium grew past 100 organizations by October 2025.
AP2 does not move money. It defines what an agent is authorized to spend, on what categories, up to what limits, and with what audit trail. The protocol uses cryptographically signed "mandates" to record and verify user authorization. It is explicitly rail-agnostic: settlement can occur over card networks, stablecoins, bank transfers, or any other payment method.
Version 0.2.0 shipped in April 2026, adding "Human Not Present" payment capabilities — enabling agents to transact when the authorizing human is offline. The protocol was donated to the FIDO Alliance for community governance. Reference implementations exist in Python (the majority codebase), TypeScript, Kotlin, and Go.
AP2's functional scope is complementary to x402. In a combined deployment, AP2 would prove that a human authorized a specific transaction; x402 would execute the settlement. This layered model addresses a gap in the x402 specification, which handles money movement but does not natively encode spending permissions or human oversight.
Mastercard launched AP4M on June 10, 2026, with more than 30 partners. According to Jorn Lambert, Mastercard's Chief Product Officer, "Agent Pay for Machines will create the conditions for a superbloom of AI business models." Lambert stated that "machine payments can make it possible for services to be bought and sold among agents at fundamentally different scales than payments today — very high volumes, very small values, very fast and at extremely low latency."
On October 20, 2026, at Cloudflare Connect, Mastercard opened early developer access — making buyer- and seller-side APIs, SDKs, and a sandbox environment available to third parties. A live demonstration showed an AI agent executing nine payments across two merchants for a marketing campaign, operating within a pre-approved budget.
The significant architectural detail: AP4M uses the x402 protocol for settlement on Base using USDC. For machine-to-machine digital services, Mastercard is effectively bypassing its own interchange-based card rails in favor of stablecoin settlement. Lambert told Fortune in October 2026 that the product is unlikely to be a "huge revenue driver next year" but acknowledged that "we're rapidly getting to something that will see real adoption."
At Visa Payments Forum 2026, Visa announced Agent Scoring, an Agentic Registry, and a Large Transaction Model. The Agentic Directory verifies legitimate participants in agentic commerce. Agent Score, built with New Generation, measures a merchant's AI-agent readiness — whether AI agents can navigate, understand, and complete tasks on a given site.
Visa's stablecoin settlement reached an annualized run rate of $7 billion as of March 2026, with more than 160 stablecoin-linked card programs live or in development. Visa integrated into OpenAI's agentic platform and co-launched the Open USD stablecoin initiative with Stripe, Mastercard, BlackRock, and Coinbase.
Launched February 11, 2026, Coinbase's Agentic Wallets provide MPC-secured wallets with programmable session caps, per-transaction limits, gasless settlement on Base, and native x402 support. The wallets are compatible with Claude, Codex, and Gemini via MCP server integration, positioning them as the default wallet layer for major AI model providers.
The three protocols route economic value to fundamentally different beneficiaries:
x402 — Zero protocol fees eliminate intermediary rent. Value flows to three parties: L2 sequencers (gas fees, fractions of a cent), Circle as USDC issuer (float income on reserves), and the merchant (full transaction amount minus gas). No interchange. No processor margin. This is the most capital-efficient model for the merchant but provides no recurring revenue stream to a network operator.
AP2 — The protocol itself captures no value. Because AP2 is rail-agnostic, value accrues to whichever settlement layer is used — card interchange if settled on Visa/Mastercard, gas fees if settled via x402, or bank transfer fees if settled via ACH/wire. AP2's value proposition to Google is strategic positioning as a standards body, not direct payment revenue.
Card delegation — Traditional interchange economics apply to card-settled consumer transactions, preserving Mastercard and Visa's existing revenue model. However, AP4M's use of x402/USDC for machine-to-machine digital services explicitly routes around interchange for that transaction class. This creates an internal tension: card networks are adopting stablecoin rails that structurally reduce their per-transaction revenue.
The card networks' participation as premier members of the x402 Foundation reflects a strategic hedge. If machine commerce scales to the trillions projected by McKinsey, card networks risk losing that volume entirely to stablecoin-native rails unless they participate in the protocol's governance and integration. Current weekly x402 settlement of $367,950 is negligible against card networks' annual volume. But the networks are evidently positioning for a future in which that ratio shifts.
The three protocols are not strictly competing — they address different layers of the payment stack. A plausible production architecture uses all three: AP2 encodes the human's spending authorization; x402 settles the payment on-chain; and card delegation (via AP4M or Visa's Agentic Registry) provides the orchestration layer, merchant verification, and regulatory compliance framework.
Mastercard's AP4M already implements this layered approach by using x402 for settlement. Coinbase's Agentic Wallets are compatible with AP2's mandate structure. Visa's Agentic Directory could serve as a trust registry for any settlement rail.
The convergence risk is governance fragmentation. x402 sits under the Linux Foundation; AP2 under the FIDO Alliance; and card delegation products remain proprietary to Mastercard and Visa. Interoperability between these governance structures is not guaranteed. As of October 2026, no single reference implementation integrates all three layers.
The three-protocol landscape reflects a rational division of labor more than a winner-take-all competition. x402 moves money. AP2 proves authorization. Card delegation provides orchestration and compliance. The question is not which protocol wins, but how they compose — and who governs the integration points.
The economic stakes are significant in projection and trivial in present reality. Weekly x402 settlement of $368,000 would need to grow by a factor of approximately 10,000 to approach the Juniper 2030 forecast. Whether that growth materializes depends less on protocol design — all three architectures are technically functional — and more on whether AI agents generate enough autonomous commercial demand to justify dedicated payment infrastructure. As of October 2026, the infrastructure is being built. The demand has not yet arrived at scale.