Lianlian DigiTech Agent Wallet Bets on AI-Driven B2B Payments

What happened

Lianlian DigiTech, a Hangzhou-based payments company, is putting AI agents at the center of its pitch to investors. In its interim filing to the Hong Kong stock exchange, the company said it is building an "Agent Wallet" payment platform for what it calls agentic commerce. In that model, software agents carry out tasks and may start transactions for the people or businesses they work for.1 The company says the platform is meant to handle the hard parts of letting machines spend money: confirming an agent's identity, managing what it is authorized to do, setting transaction limits, controlling risk and keeping a traceable record of each transaction.12 Its target customers are AI product companies, Web3 firms and large enterprises.1

The news came with first-half results that look solid on some measures and weak on others. Revenue for the six months to June 30 rose 11.9% to RMB875.6 million, about $122 million. Global payments drove most of that growth, with revenue up 27.1% to RMB600.9 million.1 Total payment volume in the global business grew 25.9% to RMB249.9 billion.6 Domestic payment revenue fell by almost half to RMB106 million, which the company says was a deliberate pullback from lower-value transaction types. Value-added services, helped by a growing virtual bank card business, rose 83.2% to RMB164.2 million.1

A strategy months in the making

The Agent Wallet did not appear out of nowhere. At its Investor Day in Hangzhou on April 17, the company said it was moving from a cross-border payments specialist to an "AI-native global financial infrastructure provider."3 Founder and CEO Zhang Zhengyu presented the change under the theme of building trusted intelligent financial infrastructure. Chief architect Zhou Jiangxiang described the Agent Wallet as a way for agents to complete transactions on their own, including machine-to-machine exchanges.3 The company also said it wants to be a neutral infrastructure provider for both human-driven and machine-driven commerce. Its proposed business model includes transaction processing, clearing and settlement, credit services and enterprise subscriptions.5

The partnerships came over the summer. In late July, Visa and Lianlian announced what they described as Greater China's first live B2B agentic transaction. Lianlian's procurement agent, LoopXPay, found a purchasing need, recommended and compared suppliers, placed an order for a product sample and paid for it, all within preset spending and approval limits. LoopXPay is listed in Visa's Agentic Directory, a registry of verified agents that supports Visa's Trusted Agent Protocol. Caixin said the transaction showed that autonomous agents can work commercially and pointed to a compliance shift toward "know your agent" rules.25

About the same time, Lianlian signed a strategic agreement with UnionPay International on July 18. It covers global procurement and AI token top-ups, plus work on overseas merchant acceptance and joint research.26 The first product from that deal is a procurement tool in which the agent handles supplier matching, product selection and payment, while a human keeps final approval.2123 With Visa, the two companies plan to extend agent use to digital advertising optimization and B2B platform payments.9

Where the coverage diverges

Reports differ on one basic point: whether the Agent Wallet is finished. Trade outlets summarizing the exchange filing say the company is still developing the platform.12 Lianlian's own results release says it was "launched" during the half as infrastructure for agent-era payments.67 An August company release lists it as a commercial product alongside tools named Lianlian Zhishu, LoopAI and an Open API Integration Skill.9 Both versions may be true in a narrow sense. Something exists and is being sold, but the full identity, authorization and limits system described in the filing reads as a plan still being built.

The company also seems to have changed its view on autonomy. In April it said agents would "initiate, approve, and settle" transactions without any human involvement.3 The products that actually shipped are more cautious. The Visa transaction ran within preset limits and approval settings, and the UnionPay product keeps a human in the loop by design.21 This looks like a sensible retreat. Banks and card networks still have to answer who approved a payment, what the agent was allowed to buy and who is liable when something goes wrong.17 Until those questions are settled, supervised autonomy is the version regulators and counterparties are likely to accept.

The financial picture also depends on which outlet you read. One Hong Kong fintech site ran a headline saying profit had doubled4, based on the company's adjusted operating profit, which rose 147.3% to RMB156 million excluding non-operating items.6 Reported first-half profit actually fell to RMB10.9 million from RMB1.51 billion a year earlier. The prior-year figure was inflated by one-time gains from LianTong, Lianlian's venture with American Express.1 Both numbers are accurate. The adjusted figure is the better guide to how the core business is doing, but readers should know that a headline about profit doubling can hide a fall of more than 99% in reported profit.

Why it matters: China's agentic payments race

Lianlian is not the only Chinese company working on this. In May, Ant Group launched an AI Wallet and a product called Token Pay for Alipay. The wallet lets users set which merchants an agent can use, how much it can spend and what it can buy.13 Ant also said its AI Pay product passed 100 million users in February, which gives it a scale advantage no smaller payments company can match.13 In September, Ant International said it would work with Visa and Mastercard on shared "know your agent" standards so that an agent registered with one network would not need to register again with the others.1115 The companies cited estimates that agents could handle $3 trillion to $5 trillion in consumer commerce by 2030.11

In the U.S., Visa has built Intelligent Commerce, an OpenAI integration and InFlow agent cards. Stripe has its Agentic Commerce Suite, and Mastercard has Agent Pay and a machine-to-machine extension. All three moved within months of each other in 2026.14 Mastercard's Agent Pay for Machines targets continuous, small agent-to-agent payments17, the same machine-to-machine market Lianlian described in April.3

Given that competition, Lianlian's choice of market makes sense. It is not trying to compete with Alipay for consumers. It is aiming at cross-border B2B procurement, where settlement delays, currency risk, layered compliance checks and manual reconciliation have long caused problems.23 Lianlian's existing business is built for that: 68 payment licenses and related qualifications, coverage of more than 200 countries and regions, and settlement in more than 140 currencies.1 The company also says it is the only Chinese provider holding every U.S. state-level money transmitter license.6 This year it added a Canadian money services license through an acquisition and a Category 3D payment license in Dubai.2

The analysis: licenses are the real asset

The main point here is that Lianlian's advantage in agentic payments comes from licenses and compliance, not from AI. Agent frameworks are quickly becoming common technology. The card networks are writing the protocols and building the registries1415, and Lianlian works inside those systems, as LoopXPay's listing in Visa's directory shows. What is harder to copy is a regulated network that can legally settle an agent's purchase from a Chinese factory to a buyer in another jurisdiction. The company appears to know this. It has used its enterprise agent platform internally for compliance reviews, risk screening, treasury and customs checks before selling AI externally.1

The numbers support that view. The global payments business has a 70.8% gross margin6, and that business is what funds the AI push. Revenue from agentic products is not broken out, and the filing does not disclose the size of the AI startup investments or the financial terms of the Visa and UnionPay deals.1 For now, the Agent Wallet is best seen as a strategic option attached to a profitable cross-border payments business, not a revenue line in its own right.

The next test is whether the procurement pilots grow beyond single transactions and showcase deals. The UnionPay partners say they will adjust the product based on real use and push for industry standards over time.21 If Lianlian can show repeat, high-volume agent-driven B2B flows on its own network, its rebranding as AI infrastructure will be credible. If not, the Agent Wallet will mostly be a label on an already strong cross-border payments business.