Lianlian DigiTech Agent Wallet Bets on AI Agent Payments

What happened

Hangzhou-based payments company Lianlian DigiTech has put agentic commerce at the center of its growth story. In its first-half 2026 results filing in Hong Kong, it described an Agent Wallet, a payment platform built so that AI agents can carry out tasks and, in some cases, start transactions for the people or companies they work for1. The company says the wallet deals with the hard parts of letting software spend money: confirming who an agent is, managing what it has been authorized to do, setting transaction limits, controlling risk, and keeping a record of what happened12. Its first target customers are AI product companies, Web3 businesses and large corporates1.

The wallet did not come out of nowhere. It grows out of the "AI Native + Globalization" strategy Lianlian set out at its Investor Day in Hangzhou on April 17. There, founder and CEO Zhang Zhengyu said the company was moving from cross-border payments specialist to "AI-native global financial infrastructure provider"3. Over the summer, the plan picked up commercial partners. Lianlian signed agentic-payment agreements with Visa and UnionPay International and completed what it calls Greater China's first live B2B agentic transaction1218.

The numbers behind the pivot

The AI push sits on a payments business that is growing fastest outside China. First-half revenue rose 11.9% to RMB875.6 million (about $122 million)1. Global payments did most of the work: segment revenue rose 27.1% to RMB600.9 million, and total payment volume grew 25.9% to RMB249.9 billion12. The global segment carried a gross margin of 70.8%, against 51.9% for the company overall6.

At home the picture was very different. Domestic payment revenue fell 49.8% to RMB106 million after Lianlian deliberately cut back on transaction types that contributed less1. Value-added services rose 83.2% to RMB164.2 million, helped mostly by virtual bank cards1. The company also lists intelligent marketing and other embedded financial services in this segment6.

The mix explains the strategy. Lianlian's profits now come from cross-border flows and from services built on top of payments. Agentic commerce is a way to push both further.

Where the coverage diverges: profit

Coverage of the results split most clearly on profitability, and the headlines depend on which number a reader sees. The company's own release and the outlets that leaned on it led with adjusted operating profit, which rose 147.3% to RMB156 million once non-operating items were excluded7. One Hong Kong fintech outlet headlined that profit had doubled5. The CFO, Wen Yingse, also pointed to an 8.4-percentage-point drop in the combined expense ratio9.

Crowdfund Insider read the same filing and reported something else: first-half profit fell to RMB10.9 million from RMB1.51 billion a year earlier1. The outlet said the prior-year figure was inflated by large one-off gains tied to LianTong, Lianlian's venture with American Express1.

Both figures are accurate, and neither tells the whole story alone. The fairest reading is that the core business is becoming more efficient, while headline earnings are small next to last year's windfall. That matters for the agentic bet. Lianlian is funding new AI businesses, investing in AI-native startups and pursuing acquisitions13 while its bottom-line profit is thin, not flush.

How autonomous is "autonomous"?

The second split is about how much the agents are actually allowed to do. The April Investor Day material described the most ambitious version. It said the Agent Wallet would let AI agents complete transactions on their own, support machine-to-machine value exchange, and close fully automated payment loops in which agents initiate, approve and settle transactions without human intervention34.

The products that reached the market in July are more cautious. In the Visa transaction, Lianlian's LoopXPay agent spotted a purchasing need, recommended and compared suppliers, placed an order for a product sample and paid for it, all within pre-set spending controls and approval parameters18. Caixin's account of the same deal also stresses that the agent worked within preset spending limits16. The UnionPay deal is explicitly human-in-the-loop: the agent handles procurement tasks, but the user keeps final approval1317.

The August filing's language sits between these two positions. It talks about authorization management and transaction limits rather than unchecked autonomy1. The coverage also uses different verbs for the wallet's status. Crowdfund Insider and Small World describe it as still in development12. The company's results release says it "was launched" during the half6, and an August press release lists it among products already on the market8.

The likeliest explanation is that a first version exists for selected use cases, while the fully autonomous version shown to investors remains a goal. That gap is normal, and it is probably sensible for a regulated payments firm. Still, anyone judging the company should weigh the human-approved procurement pilots more heavily than the machine-to-machine vision.

Why the card networks matter

The partnerships are the strongest evidence that this is more than a slide-deck idea. LoopXPay is registered in Visa's Agentic Directory, a registry businesses can use to identify verified AI agents18. The directory supports Visa's Trusted Agent Protocol, which provides identity verification, transparency and controls for transactions driven by agents18. Caixin presents this as adding a "know your agent" layer to compliance16, which fits closely with the identity and authorization problems the Agent Wallet says it solves1.

Visa and Lianlian say they plan to extend agent use into procurement, digital advertising optimization and B2B platform payments18. With UnionPay, the first focus is global procurement and AI token top-ups, meaning payments to replenish AI service credits128. The UnionPay agreement also covers overseas merchant acceptance and joint technology work12. Both partners have described the procurement pilots as a testing ground, with standardization and wider adoption as the longer-term goal1317.

The fact that Lianlian is working with both a U.S. card network and China's domestic network matters. Its business is connecting Chinese supply chains with overseas buyers and sellers, and the company describes itself as a link between Eastern supply chains and global markets3. If AI agents start buying and selling across that corridor, Lianlian wants to be the trusted layer whichever network clears the payment.

Licenses as the moat

The least glamorous part of the story may be the most important. Lianlian held 68 payment licenses and related qualifications as of June1. A later company release put the figure at 698. Recent additions include a Canadian money services business license, obtained through an acquisition in April, which covers foreign exchange, remittances, virtual currency and payments1. A Category 3D payment license followed in Dubai in May2. The company says it also holds money transmitter licenses in several U.S. states and a virtual asset trading platform license that underpins its digital-currency and blockchain services6.

That licensing base, Lianlian argues, is what turns agentic payments from a demo into a business. Its chief architect, Zhou Jiangxiang, told investors that licenses, compliance history and cross-border data together create a defensible position built on trust and network effects3. In practice, an AI agent buying from a supplier in another country still triggers foreign-exchange, sanctions and anti-money-laundering obligations. A provider that already meets those rules in more than 200 countries and regions and settles in more than 140 currencies1 has a real advantage over an AI startup that only has a clever agent.

Lianlian is also using AI inside its own operations. It says its enterprise agent platform already handles cross-border compliance reviews, risk screening, treasury management, customs reviews and customer service1. That is the kind of back-office automation that could help explain the lower expense ratio9.

The verdict

Lianlian's move into agentic payments is a credible bet, not a finished product. In its favor, it has live transactions with Visa, a signed program with UnionPay, a global network of licenses, and a fast-growing high-margin cross-border business181216. Against it, Lianlian has not disclosed the size of its AI investments or the financial terms of its card-network deals1, reported profit is thin once last year's one-off gains are stripped out1, and actual deployments still keep a human in the approval loop13.

The right lens is therefore positioning. Lianlian is trying to build the identity, authorization and settlement layer for agent-to-agent trade between China and the rest of the world before the market fully exists. The next things to watch are whether Agent Wallet transaction volume shows up as its own line in future filings, and whether the procurement pilots move from samples to repeat, larger purchases.