"Important, important, very important."

This is the tone set by Alipay insiders on the strategic position of AI payment within the group. The word "important" is repeated three times, and its weight appears.

On May 26, Alipay disclosed a set of data: it has completed a total of 300 million AI agent payments and supports 95% of universal agent frameworks, becoming "the world's first large-scale commercial AI native payment infrastructure." "Token Pay" and "AI Wallet" released on the same day, together with the previous "AI Pay" and "AI Receive", constitute Alipay's "four-piece full stack" for the agent era.

Alipay is not the only company that is actively deploying AI payment. In February this year, JD.com launched "JD AI Payment"; in April, China UnionPay took the lead in releasing the "Agent Payment Open Protocol Framework" (APOP), attracting 19 domestic and foreign institutions to complete the first batch of production verification; in the same month, WeChat Pay quietly launched an AI-oriented payment access tool set. Even some payment companies are no longer satisfied with staying at the payment layer. For example, Lianlian Digital launched “Lianlian Smart Hub”, which aggregates more than 300 large models into a unified calling interface and moves upstream closer to agent decision-making.

Ant Group CEO Han Xinyi said: The nature of business has not changed in the AI era, but the new role of Agent restructures everything.

Alipay itself is also being restructured.

When a user says "order me a cup of coffee" to AI glasses or a dialog box, and the decision-making, ordering, payment, and reconciliation are all completed by the agent, Alipay's business model that previously relied on pulling users into the App and feeding the financial business with traffic from payment scenarios may be quietly dismantled.

Alipay does not deny this. In response to a question from Barron's Chinese website about "will it be channelized?" Li Jiajia, co-president of Ant Group's digital payment business group, said, "We must be confident today. If we think too much, we will not be able to do it. If we firmly do this well and expand the ecology, there will naturally be new opportunities. We are not worried about the issue of channelization at the moment. We may find that it will be like that in a year, but there is no way. This is our choice."

What exactly is AI payment

The so-called AI payment does not mean "let AI help you enter your password".

The real AI Agent payment is that the agent runs the entire transaction chain on behalf of the user under the user's authorization, including a series of processes such as intention understanding, order placement, pre-authorization, confirmation, refund, and reconciliation. For example, if you authorize a travel agent to plan a business trip next week, it will automatically compare prices, order flights and wine, book a car, and place an order for payment. You only need to glance at the bill in the AI ​​wallet.

In the latest case of Alipay, its “four-piece full stack” for the agent era has its own division of labor.

The floor-to-ceiling scene is spread to both soft and hard sides at the same time. On the software side, AI Fu has supported 95% of general agent frameworks such as Qianwen, JVS Claw, Claude Code, Hermes Agent, etc., and has settled on AI tool platforms such as TRAE SOLO, Qoder, and Kouzi. On the hardware side, it has teamed up with Qianwen AI glasses, Rokid, future smart AI headsets, and smart cockpits from Ideal, Chery, Geely, Dongfeng and others.

Li Jiajia gave an example, "In the past, when driving to pay, you had to roll down the window to scan the QR code. Now, with the help of the camera on the car, you can automatically scan the QR code and interact with the smart agent in the car to complete active inquiries and payments."

Judging from the number of payments, AI payments are expanding rapidly. A total of 300 million smart payments were made, with 200 million added in the past two months. Zhu Lin, general manager of AI payment at Ant Group, predicts that the number of transactions in a single month may exceed 1 billion by October this year.

It is worth noting that Alipay admitted that 90% of the current 300 million transactions came from Qianwen. From a certain perspective, vertical integration gives Alipay a speed advantage that no independent payment company has. However, it is undeniable that Alipay's current tone of "the world's first large-scale commercial AI native payment infrastructure" still mainly relies on Alibaba's own system.

AI payment requires full link reconstruction

Apart from scale and scenarios, the real difficulty of AI payment lies at the bottom level. During the interview, Zhu Lin mainly focused on several challenges.

The first is intention restoration. The original payment was limited by data boundaries and sensitive information. In fact, the payment company could not fully know what the user bought. The deduction was completed and the remaining responsibility fell to the merchant. However, for AI to allow the agent to accurately execute user instructions, it must penetrate this layer of data boundaries and truly restore the original intention of the user when entrusting it. What Alipay wants to build is a "large payment model" that can understand the intention of natural language.

The second is the ambiguity of the intention itself. For example, the user says to the Agent, "Buy me a breakfast every morning." Soy milk and fried dough sticks, Starbucks sandwiches, and set meals from the restaurant downstairs are all options. To determine whether a deduction is really in line with the user's original intention, the system must have sufficient knowledge of user habits and categories of things. It cannot rely on rules, but must rely on models.

The deeper difficulty lies in ecological synergy. Agent portals, developers, service providers, and payment acquirers all need to redesign the collaboration mechanism and trust model. Zhu Lin admitted that “we cannot rely on ourselves alone” and must “break through many existing trust models.” What payment companies could accomplish independently in the past has become a multi-party collaboration project in the Agent era.

The superposition of the three difficulties is essentially a paradigm shift: from KYC to KYA, that is, from "know your customer" to "know your agent". Traditional risk control determines whether the user is the person, but in the AI ​​era, the answer is whether the agent is performing within the scope authorized by the user.

New scenarios have also created new risks. In February this year, an incident occurred in the cryptocurrency circle: an AI Agent named Lobstar Wilde was deceived by a fabricated help post on social media and transferred $450,000 independently, while the user originally only wanted to give 4 SOL.

The response idea given by Alipay is to use blockchain to trace the entire process, and cooperate with compensation and subsequent accountability. Zhu Lin said that their design goal is to allow every problematic transaction to be restored "whether the user's intention is not clear, the model has hallucinations, or the merchant provides an error."

Alipay also stated that AI payment only assumes the role of order execution. The payment process is always authorized by the user and must be verified by face, fingerprint or password, and the multi-dimensional risk control system intercepts risks in real time, and promises "I dare to compensate you if you dare to pay."

Is Alipay worried about being channelized?

For Alipay, the long-term concern of AI payment is that its logic itself deconstructs the entry value of Alipay. After all, when AI agents become the middle layer between users and the business world, the action of "using Alipay" itself may be disappearing.

In this sense, is Alipay worried about being channelized?

Facing this question, Li Jiajia, co-president of Ant Group’s digital payment business group, said that he is “not worried” about the issue of channelization. But he also added, "Maybe it will turn out to be like that in a year, but there is no way. This is our choice."

But this is not the proposition of Alipay. Since the end of 2025, JD.com, China UnionPay, and WeChat Pay have entered the AI ​​payment track almost simultaneously.

JD.com launched ClawTip, an A2A micropayment infrastructure, targeting small-amount transactions between agents; China UnionPay released the "Agent Payment Open Protocol Framework" (APOP), trying to occupy the bottom layer of the industry with a standard protocol; WeChat Pay launched an AI Skill skill package and an AI-friendly API, allowing developers to generate payment codes in natural language, and launched merchant tools such as AI menu recognition.

Each company seems to have different product forms, but in fact they are competing for the same thing: the right to formulate rules for the next generation of business interactions. When an agent makes transaction decisions on behalf of humans, who will define the rules for identity verification, fund transfer, and authorization boundaries? Who can become the trust layer called by the Agent by default? Who can continue to occupy the payment ecological niche in the Agent world after the App disappears.

Li Jiajia traced the origins of Alipay to Barron’s Chinese: Back then, guaranteed transactions solved the trust problem of e-commerce, and QR codes solved the threshold problem of offline mobile payments. Neither of them had a clear profit model in the early days, but they eventually achieved commercial success.

Today, AI payment faces the same proposition. The more important direction is to first allow consumers to dare to hand over their wallets to agents, allow merchants to collect payments smoothly, and allow developers to be willing to build intelligent agents. Alipay has even launched an incentive plan for AI developers, including special token subsidies, zero payment rate discounts for individual developers, supply support, etc.

As for the future profit model and whether Alipay will be "channelized", we will have to wait until the agent economy really takes off.