On May 26, Alipay launched two products that mark a significant step in how money actually moves inside the agentic economy: AI Wallet, a consumer-facing control layer for AI-executed transactions, and Token Pay, a B2B settlement product purpose-built for AI model providers.

The announcement came at the Alipay AI Payment Ecosystem Conference and follows a run of milestones that show just how fast this category is moving. In February 2026, Alipay AI Pay became the first AI-native payment product globally to exceed 100 million users. At that point, the platform was processing more than 120 million AI agent transactions per week. By the time of the May 26 launch, total transactions had crossed 300 million.

These are not prototype numbers. Alipay is processing AI agent payments at a scale most payment networks took years to reach with human-initiated transactions.

What AI Wallet and Token Pay Actually Do

The distinction between the two products is worth understanding clearly, because they address different sides of the same problem.

AI Wallet sits inside the Alipay app and gives individual users visibility and control over payments that AI agents make on their behalf. Before a transaction completes, during it, and after, the user can see what the agent did, why, and how much it cost. This matters because AI agents increasingly shop, book, and transact without direct human input at every step. Without a proper control layer, users have no practical way to understand, dispute, or audit those actions.

Token Pay is aimed at AI model companies. It provides a one-stop platform for managing subscription billing, token top-ups inside AI agents, and complex micro-transactions between agent platforms, model providers, and merchants. For any company building AI products where users pay per query, per action, or per output, Token Pay provides the payment infrastructure to make that commercially viable.

Together they address a gap that has been obvious to anyone deploying real AI agents: the payment rails underneath most agentic applications were designed for humans clicking checkout buttons, not software agents executing workflows at speed and scale.

The Agentic Commerce Trust Protocol

Alongside the product launches, Alipay formalised China’s first Agentic Commerce Trust Protocol in partnership with AI companies including MiniMax and Stepfun. The protocol establishes a common technical and commercial language for how AI agents, service platforms, and payment systems communicate and verify intent before money changes hands.

Trust is the hard problem in agentic commerce. When an AI agent authorises a payment on your behalf, how does the merchant know you sanctioned it? How does the user know the agent didn’t overpay or transact with a fraudulent party? The Trust Protocol attempts to create a verifiable chain of authorisation that makes these questions answerable at scale.

Cyril Han, CEO of Ant Group (Alipay’s parent), framed it directly: “While the essence of commerce remains unchanged in the age of AI, the emergence of AI agents is reshaping everything. Alipay is building a new generation of AI payment services to accelerate the growth of the agentic commerce ecosystem.”

Why This Matters Beyond China

Most of Alipay’s 300 million transactions are happening in China, where mini-programmes from major retailers like Luckin Coffee and hardware integrations with AI smart glasses have pushed agentic payments into everyday consumer life faster than anywhere else.

But the infrastructure pattern matters globally. Every business that deploys AI agents to handle bookings, procurement, customer service, or operational tasks eventually runs into the same friction: the agents can do the work, but the payment layer cannot keep up.

Token Pay’s model, a single platform handling subscriptions, per-token billing, and micro-transactions for AI workflows, is exactly what enterprise AI deployments need. Currently, teams cobbling together agentic workflows across tools like Anthropic Claude, OpenAI GPT-5.5, and custom models face a mess of billing integrations that were designed for software-as-a-service products, not agent-driven workloads.

Alipay’s move puts pressure on Western payment providers to offer equivalent infrastructure. Visa already announced its Agentic Commerce framework earlier this year. Stripe, Braintree, and other major processors have all quietly expanded their API capabilities to handle agent-initiated transactions. The market is standardising around the idea that payment rails need to understand agent intent, not just process transactions.

What This Means for Business

If you are building or deploying AI agents in your business, the payment question will arrive sooner than you expect.

For businesses using AI agents internally, the immediate concern is spend visibility. When an AI agent makes 50 API calls, triggers 12 service lookups, and sends a report, how do you see what that cost? AI Wallet’s model, giving users a clear view of what agents spent and why, is the right design pattern regardless of which payment provider ultimately delivers it.

For businesses building AI-powered products or services, Token Pay’s model is the one to watch. The ability to handle per-token billing, subscription management, and micro-transactions in a single API is a real constraint on how AI products can be commercially structured. A SaaS pricing model works poorly when your users are AI agents consuming variable amounts of compute on unpredictable schedules.

For decision-makers evaluating AI investments more broadly, this launch is a signal that the agentic economy is becoming infrastructure, not just a research concept. When payment systems are building native support for agent-initiated commerce, the transition from AI as a tool to AI as an autonomous operator is no longer theoretical.

---

The shift to agentic operations is picking up speed, and the commercial infrastructure to support it is arriving in real time. If your business is thinking through what AI agents could handle operationally, the conversation starts here. And if your team needs the skills to work effectively alongside AI agents as they take on more of the operational layer, EDNA Learn’s business programme is built for exactly that moment.