AI agents are acquiring financial infrastructure. In the span of 90 days — from February to May 2026 — Coinbase, MoonPay, Tether-backed Oobit, and Crossmint each shipped production-grade payment rails designed not for humans, but for autonomous software. Coinbase's x402 protocol has processed 165...
"Very soon there are going to be more AI agents than humans making transactions. They can't open a bank account, but they can own a crypto wallet." — Brian Armstrong, CEO, Coinbase
AI agents are acquiring financial infrastructure. In the span of 90 days — from February to May 2026 — Coinbase, MoonPay, Tether-backed Oobit, and Crossmint each shipped production-grade payment rails designed not for humans, but for autonomous software. Coinbase's x402 protocol has processed 165 million transactions across 480,000 transacting agents. MoonPay launched a Mastercard-linked virtual card that lets agents spend stablecoins at 150 million merchants. Oobit released Visa-networked Agent Cards funded from corporate USDT treasuries. Crossmint's GOAT SDK crossed 100,000 downloads, supporting agent interactions across 30+ chains.
The market is no longer theoretical. McKinsey's midpoint scenario projects AI-powered agents could generate $2.9 trillion in U.S. economic value per year by 2030. Gartner forecasts 40% of enterprise applications will include task-specific AI agents by end of 2026. The question facing blockchain infrastructure is whether crypto rails — stablecoins, programmable wallets, on-chain settlement — become the default financial backbone for non-human economic actors.
Four distinct infrastructure stacks reached production between February and May 2026, each targeting a different layer of the agent payment problem:
| Provider | Product | Network | Launch | Key Metric | |----------|---------|---------|--------|------------| | Coinbase | Agentic Wallets + x402 | Base (primarily) | Feb 2026 | 165M txns, 480K agents | | MoonPay | MoonAgents Card | Mastercard via Monavate | May 1, 2026 | 150M+ merchant acceptance | | Oobit (Tether-backed) | Agent Cards | Visa | Apr 30, 2026 | 150M merchants, 100+ countries | | Crossmint | GOAT SDK | 30+ chains | Ongoing | 100K+ downloads, 200+ tools |
The approaches differ structurally. Coinbase built a native on-chain protocol (x402) that settles machine-to-machine payments in stablecoins without touching legacy card rails. MoonPay and Oobit took the opposite approach: bridging on-chain stablecoin balances to existing Visa and Mastercard networks so agents can transact at traditional merchants. Crossmint built the middleware — a wallet-agnostic SDK that lets developers connect agents to any wallet provider and any chain.
These are not announcements. Each product is live and processing transactions.
Coinbase open-sourced x402 in May 2025, naming it after the HTTP 402 "Payment Required" status code that has sat unused in the HTTP specification since 1997. The protocol enables machine-to-machine payments at the HTTP layer — an agent hits a paywalled API endpoint, receives a 402 response, and automatically executes a stablecoin payment to gain access.
As of late April 2026, according to Coinbase disclosures, x402 has processed:
The average transaction size implied by these figures is approximately $0.30 — consistent with micropayment and API-access use cases rather than large-value transfers.
On April 2, 2026, Coinbase donated x402 to the x402 Foundation at the Linux Foundation. The foundation's launch member list spans 22 organizations including Adyen, American Express, AWS, Circle, Google, Mastercard, Microsoft, Shopify, Solana Foundation, Stripe, and Visa. This breadth of institutional backing for an agent payment standard is without precedent.
On April 21, 2026, Coinbase launched Agentic.Market — a directory of x402-paywalled services functioning as an app store for AI agents, where autonomous software discovers and pays for digital services without human intervention.
However, a March 2026 CoinDesk report noted that despite high transaction counts, "demand is just not there yet" for x402 micropayments, suggesting much of the volume may stem from testing and developer experimentation rather than organic commercial demand.
Two companies shipped card products for AI agents in the same week. The architectural choices reveal different bets on how the agent economy will operate.
MoonPay's MoonAgents Card (May 1, 2026): A virtual Mastercard debit card issued through Monavate, a regulated payments provider that is a principal member of Visa, Mastercard, and Discover networks. The card links a self-custodial wallet to Mastercard rails. Custody never leaves the user — the wallet retains keys, and approvals are revocable. Available via MoonPay CLI in the UK and LATAM at launch. Exodus Movement is the launch wallet partner. Identity verification (KYC) is required before issuance.
Oobit's Agent Cards (April 30, 2026): Virtual Visa cards funded directly from a company's USDT stablecoin treasury. Spend policies are configured at setup and enforced server-side with no override path. Every transaction generates a real-time, human-readable log. Compatible with AI agent frameworks from OpenAI, Claude, AutoGen, and LangChain. Onboarding requires KYB (Know Your Business) verification. Oobit, backed by Tether, covers 150 million merchants across 100+ countries through its existing infrastructure.
The distinction matters: MoonPay's product is consumer-facing, linking individual self-custodial wallets to card rails. Oobit's is enterprise-facing, connecting corporate stablecoin treasuries to employee-equivalent spending by software agents.
Both require traditional identity verification. Neither enables fully autonomous, permissionless agent spending. The agents gain purchasing power at traditional merchants, but within human-defined policy guardrails.
Tether's positioning in the agent economy extends beyond its $189 billion USDT stablecoin. The company is building a vertically integrated stack:
Ardoino's public statements frame the scale explicitly: "When AI agents will be everywhere, there will be the need of trillions of payments per day." He has projected "1 trillion AI agents within 15 years" and stated that these agents "will not have accounts with PayPal or Wells Fargo. They will use stablecoins because programmability is what AI requires."
Tether indicated it may advance integration between WDK and QVAC to support Agent Cards-related features. The strategic logic: if QVAC-powered agents run locally on user devices and transact via WDK-powered wallets funded with USDT, Tether captures value at multiple layers of the agent payment stack — issuance, wallet infrastructure, and AI runtime.
This vertical integration mirrors Tether's existing strategy of expanding beyond stablecoin issuance into infrastructure (data centers, telecom, mining), but applied to the agent economy.
The middleware layer — how developers actually connect agents to financial rails — is consolidating around a small number of frameworks.
Crossmint GOAT SDK: MIT-licensed, open-source, wallet-agnostic. Supports 30+ chains, 5 major agent frameworks (Vercel AI, LangChain, LlamaIndex, MCP, OpenAI), TypeScript and Python, and 200+ on-chain tools. Over 100,000 downloads. Uniquely offers both stablecoin rails and card network rails (Visa, Mastercard) through a single integration.
Coinbase Agentic Wallets: Non-custodial wallets secured in Trusted Execution Environments (TEEs). Programmable spending limits including session caps and transaction limits. Tightly integrated with x402 and the Base network.
Cobo: Enterprise-focused agentic wallet platform positioned for active traders with institutional-grade custody and policy engines.
In a 14-week beta program running from October 2025 through January 2026, over 1,000 participants created more than 9,500 agents that executed 187,000 autonomous crypto transactions — early but measurable proof that developers are building with these tools.
Investment capital reflects the trend. According to industry data, investment in agent-focused projects surged from 5% to 36% of all crypto AI deals between H2 2023 and H1 2025. Haun Ventures raised $1 billion in 2026 with agentic economy infrastructure explicitly in its investment thesis.
Published forecasts vary widely, making precision difficult. What follows are the ranges from identifiable sources:
Agent Infrastructure Market:
Enterprise Adoption:
Crypto-Specific Metrics:
The gap between $50 million in actual x402 volume and $30 trillion in projected machine purchasing power spans three orders of magnitude. The infrastructure is being built, but organic commercial demand remains nascent.
From an economic value perspective, the agent payment stack introduces new fee extraction points:
Stablecoin issuance layer: Tether/Circle earn yield on reserves (~5% on U.S. Treasuries backing USDT/USDC). This revenue accrues regardless of whether agents or humans transact.
Wallet/SDK layer: Coinbase (via CDP fees), Crossmint (via API pricing), and others charge developers for wallet provisioning and transaction processing.
Card network layer: Visa and Mastercard extract their standard interchange fees (typically 1.5–3%) when agents spend at traditional merchants via MoonPay or Oobit cards.
Card issuance layer: Monavate (MoonPay) and Oobit capture issuer-side fees and markup.
Protocol/chain layer: Base, Solana, and other L1/L2 networks collect gas fees on settlement transactions.
Application layer: Agent.market and similar directories may introduce listing fees or revenue shares.
The stacking of these fees means an agent paying for a $10 API service via an Oobit Agent Card on Visa rails could face cumulative fees from USDT conversion, Visa interchange, Oobit issuance margin, and merchant processing — potentially 3–5% total. By contrast, the same payment settled natively via x402 on Base incurs only Base gas fees (sub-cent). This cost differential will likely determine which settlement path scales.
The parallel emergence of four competing agent payment infrastructures in a single quarter marks a structural shift in how blockchain networks are being positioned. The bet is no longer on humans using crypto — it is on software using crypto because it has no alternative. Agents cannot satisfy KYC requirements, cannot open bank accounts, and cannot sign traditional financial agreements. Programmable stablecoins and permissionless wallets solve these constraints by design.
The economic question is whether agent transaction volumes will reach the scale necessary to justify the infrastructure investment. At $50 million in cumulative x402 volume against a $30 trillion projected machine economy, the ratio of infrastructure-to-demand remains heavily tilted toward supply. The companies shipping these products are making a timing bet: that building agent financial rails now creates defensible positioning for a market that may take 3–5 years to materialize at meaningful scale.
The fee economics will likely determine the winning architecture. Native on-chain settlement via x402 or equivalent protocols offers sub-cent transaction costs. Card-bridged solutions via MoonPay and Oobit add legacy interchange fees but gain immediate merchant acceptance. Which path dominates depends on whether agents primarily transact with other agents (favoring on-chain settlement) or with human merchants (favoring card bridges). The answer is probably both — but the revenue distribution across the stack will differ substantially.