Alipay Expands AI and Mobility Ecosystem at 2026 Bund Conference with NFC Transit Alliance, In-Car Services, and Agentic Commerce Push

The alliance will combine transit services, terminal devices and payment capabilities, using Shanghai and Beijing as pilot cities to accelerate NFC 'tap-to-ride' across more devices and travel scenarios. The service allows users to tap a phone with the screen off, without opening an app, to pass through transit gates. It also supports ride-now-pay-later and pay-as-you-go models, so users do not need to manually recharge or worry about insufficient balances. Alipay said the value extends beyond faster entry: real-time metro information, card-face malls, real-time buses and metro network maps can be connected around one tap. In November 2025, Alipay and Shanghai Transport Card launched a 'ride first, pay later' service in Shanghai. In early 2026, they introduced the 'Shanghai NFC Travel Savings Card,' which combines transit, shared bikes, ride-hailing and cultural tourism tickets. At the conference, Shanghai Transport Card, Huawei and Alipay launched an upgraded Huawei NFC 'tap-to-ride' service and exclusive virtual card faces. An Alipay mobility executive said the company will continue to open its technology and ecosystem resources to cover more cities, terminals and travel scenarios.

In the smart cockpit sector, Alipay released three new in-car AI capabilities: Highway Mobile+, in-car movie ticket booking, and one-click refueling. The services rely on Alipay's AHA multi-agent cross-device protocol, which connects services across devices and brands, and on Ant Abao, a super service agent. Alipay said Ant Abao's mobile phone is its native core service screen, while the smart cockpit is becoming a second important service terminal. More than 10,000 AI-adapted life services in the Ant Abao ecosystem can be reused from phone screens to car screens. Merchants that adapt a service once can distribute it to phones, car machines and AI glasses. The services now cover 17 mainstream automakers and more than 16 million smart cars, with more than 60 additional designated automakers. Users can say 'help me pay the parking fee' or 'help me buy a movie ticket' in the car, and Ant Abao will invoke the service. The new movie ticket service lets users select a film, locate a cinema, choose a showtime and pay by voice. The one-click refueling service matches a nearby station and completes verification and payment without the driver getting out. In highway scenarios, users can open and bind the service in the cockpit and pay highway tolls by voice without opening the window, taking a card or using a phone. Cross-device continuity allows unfinished tasks to be completed on a phone after leaving the car.

At the 'New Possibilities of Agentic Commerce' panel on Sept 11, Fu Qiang, Kimi's growth and commercialization technology lead; Tristan, founder of Natural Selection; Zhao Penglan, senior partner at BAI Capital; and Lin Zhengmao, AI payment director at Ant Group, discussed how payment changes when AI agents conduct commerce. Lin said that in the AI era, a user can express an intent or goal and authorize an agent, after which the agent can search for goods, place orders and complete payment. Fu said that in Kimi's membership subscription scenario, users can speak a few words through a local model or other means to continue using the service. Zhao said he hopes payment will one day automatically buy an iPhone 18 Pro without being told. He argued that a bigger increment comes from 'silicon-based life payments,' such as an agent performing investment research that calls APIs 500 times in a minute, each costing a few cents. A human credit card would be blocked for such frequency. Tristan shared a case in which a user complained about expensive flights from Shanghai to Tokyo during Qingming Festival, and another ticket-selling user's agent proactively recommended a ticket; the user had not actively said 'I want to buy a ticket.' Tristan said the fundamental change is that human work becomes AI work, with humans only confirming. Zhao said payment has strong network effects, and if one person has 10 or even 100 agents, there could be 100 billion to 500 billion agents, creating new commercial value. Tristan said network effects may be 'epically strengthened' because agents have deeper context and can connect at higher dimensions; he offered a rough formula: internet network effects equal the square of nodes, while AI network effects equal the square of nodes times the depth of context. Traffic will remain, but the entry may become the user's own agent. Lin said merchants will need to gain the trust of agents so that agents will recommend and invoke them. Payment is moving from the last step to the whole chain of intent, authorization and agent execution, requiring agent identity, payment execution and transaction proof to be included in a trusted transaction chain. Fu said he hopes agents can independently earn money and send it to his Alipay account; Zhao expects AI-native payment to find product-market fit; Lin said agent-initiated payments exceeding human payments will appear soon.

Ant Lingbo Technology held a forum titled 'Beyond the Foundation: Scenario Breakthroughs and Collaborative Evolution of Embodied Intelligence' on Sept 11. Shen Yujun, chief scientist at Ant Lingbo, said the debate over VLA or WAM should return to first principles: what capabilities robots need. He said a general brain needs to extract effective representations from raw signals, fuse and align vision and touch, and generate executable actions. Embodied models cannot simply copy digital-world information processing. On evaluation, Shen said two measures matter: how much cost partners pay to deploy a model in a specific scenario and whether that cost continues to fall with model upgrades; and whether a robot can stay in an open venue with people and the environment for a whole day without problems. Xue Tianfan, assistant professor at the Chinese University of Hong Kong, said different bodies have different sensors and modalities, requiring joint iteration of models, data and evaluation. Li Yonglu, associate professor at Shanghai Jiao Tong University, said a single robot demo conveys less and less information, and evaluation should first serve research and development. Huang Yongtao, chief data scientist at Ant Lingbo, said the data flywheel has not truly turned because too few robots enter real production and service scenarios; many remain in research, data collection or display, so failures and feedback do not flow back. He divided data quality into an engineering baseline and a training-objective layer, and said the key is to turn current model problems into quantifiable data filters. Zhang Jianwei, vice president of Guanglun Intelligence, placed evaluation in the loop: use simulation evaluation to find problems at scale, collect targeted data, verify through real deployment, and bring new failure cases back to training. Li Maoqing, vice president of Mifeng Technology, advocated tiered use of data. In the CEO panel, Zhu Xing, CEO of Ant Lingbo, said robots need to work even in their 'childhood,' but scenario selection should start with safer, more controllable, shorter-task-chain areas. He said the difficult problem has shifted from 'one' to 'many': a robot may work in one store, but the next store's shelf layout, product positions and task details change. Chang Lin, co-founder and CEO of Leju Robotics, said robots have already started 'real work' in production, including loading and unloading parts and carton depalletizing. Gu Jie, founder and CEO of Fourier, said robots entered hospitals and elder-care institutions before the embodied intelligence concept rose; hospitals and patients care about efficacy, safety and return on investment, not whether a model or traditional program completes the task. Niu Teng'ao, founder and CEO of Qingxinyi, said general robots capable of multiple tasks are still far from homes; the near-term path is interaction and companionship, building relationship understanding, scenario understanding, role construction and growth. Wang Cong, CEO of Digo Robotics, said embodied intelligence currently solves generalization in single scenarios, single tasks or even single end-effectors, and he focuses on whether a 'small foundation model' covering multiple manipulation tasks can emerge first. Wang Xiaogang, chairman of Daxia Robotics and co-founder of SenseTime, said a sustainable business model begins when a scenario has thousands of robots and customers break even. Ant Lingbo displayed a smart pharmacy with China National Pharmaceutical Group: a robot moves precisely and avoids obstacles in 80-centimeter-wide aisles, completes order taking, drug identification, picking and delivery without modifying shelves or store layouts. The solution has been deployed in China National Pharmaceutical Group retail stores. Ant Lingbo also launched an embodied large model challenge, inviting universities, research institutions, developers and industry partners to train and conduct real-machine practice based on LingBot-VLA 2.0.