AnswersPublished20 sources
How China’s Internet Giants Are Turning AI Into Delivery Fleets, Warehouses and Robotaxis
China’s internet giants are shifting from digital platforms to physical AI systems: Meituan is commercializing drone logistics, JD.com is replicating automated warehouses, ByteDance is developing embodied AI robot mod... The clearest near term deployments are in structured, repeatable environments—warehouse picking,...
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer
Key takeaways
- China’s internet giants are shifting from digital platforms to physical AI systems: Meituan is commercializing drone logistics, JD.com is replicating automated warehouses, ByteDance is developing embodied AI robot mod...
- The clearest near term deployments are in structured, repeatable environments—warehouse picking, fixed drone corridors and fleet operations—rather than fully general purpose robots replacing people everywhere.
- For manufacturers, retailers and mobility companies, the shift makes partnerships with AI platform, cloud, chip and fleet operations providers increasingly important.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How are major Chinese internet companies—including Meituan, JD.com, ByteDance, Baidu, Alibaba, and Tencent—expanding from pure digital busin. Article summary: Chinese internet companies are becoming “physical-AI” operators: they are combining foundation models, perception software, specialized chips and cloud platforms with vehicles, robots, warehouses, drones and real operati. Topic tags: general, academic, general web, user generated, education. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, water
China’s leading internet platforms are increasingly applying AI to the physical economy: moving goods through automated warehouses, delivering orders by drone or autonomous vehicle, operating robotaxi fleets, and developing the models and chips that can run machines in the real world. The strategic shift is from building apps alone to coordinating a full operating stack—demand, data, AI, hardware partners and live deployment.
The evidence is strongest in logistics and mobility, where tasks can be repeated at high volume and measured closely. It is a more practical starting point than the promise of a general-purpose humanoid that can work anywhere.
The real product is an operating system for physical work
A robot or autonomous vehicle is only one part of the system. To work commercially, it must be connected to dispatch software, maps or routes, charging and loading infrastructure, remote operations, safety processes, merchants or warehouses, and people who handle exceptions.
That is why large consumer and commerce platforms have a distinctive advantage. They already run high-frequency transaction networks and can embed automation in workflows they control. Repeated real-world use creates operational feedback: failures, delays, routing patterns and edge cases can be used to improve the next iteration of the service.
The resulting business is often closer to a managed service than a one-time hardware sale. Customers may ultimately buy delivery capacity, warehouse throughput or fleet uptime—not simply a robot.
Meituan: automating local commerce, one delivery leg at a time
Meituan’s route into physical AI begins with local, on-demand delivery. Its drone effort combines aircraft with transfer hubs and an operations-control system, rather than treating the drone as a standalone product. In May 2026, the company said this infrastructure had entered commercial use and that it was opening low-altitude logistics solutions to authorized partners. 54
Regulatory progress matters as much as the hardware. In April 2025, Meituan said its fourth-generation drone had passed a Civil Aviation Administration of China review and received a nationwide operating certificate for low-altitude logistics. 61 Meituan had also begun commercial drone delivery in Hong Kong in March 2025, which it described as its 54th commercial route.
62
The operational model is still selective. Autonomous systems can take predictable segments of an order, while people manage handoffs and difficult last-meter conditions. In Beijing’s Shunyi district, for example, Meituan used autonomous vehicles to move parcels to transfer stations, where couriers completed the final portion of delivery. 55
That division of labor is important: dense delivery demand may make automation valuable, but building access, weather, mixed pedestrian environments and customer handoffs remain hard physical-world problems.
JD.com: turning the supply chain into a robotics deployment network
JD.com has a different starting point: logistics infrastructure. Its opportunity is to connect inventory, warehouse operations, picking, sorting, transport and delivery through one coordinated system.
JD Logistics reported that, by December 31, 2025, more than 20 of its self-developed LangzuTech goods-to-person automated warehouses were operating in nearly 20 Chinese cities. The company said the systems supported high-density storage and rapid picking across millions of SKUs, including during major sales periods. 17
That makes JD’s automation strategy more than an internal efficiency program. A standardized warehouse system can be replicated across locations and potentially offered alongside fulfillment and supply-chain services. JD has also publicly described smart logistics and unmanned technology as areas of internal research, development and application. 25
The lesson for retailers and brands is that warehouse robotics becomes most useful when it is integrated with inventory placement, order management and fulfillment—not added as an isolated machine on the warehouse floor.
ByteDance: building the embodied-AI layer
ByteDance’s physical-AI position is more model-centric. Its Seed research group has presented GR-3 as a generalist vision-language-action robot model designed for generalization and longer-horizon tasks. It also introduced ByteMini, a bimanual mobile robot platform intended for manipulation work. 3
This matters because embodied AI requires more than perception. A system must connect what it sees and understands with actions that manipulate objects safely and reliably. ByteDance’s work therefore points toward an enabling layer for robotics: robot policies, manipulation research, model training and developer-oriented AI infrastructure.
The company has also expanded its robotics ambitions through recruiting. A 2025 report described a ByteDance drive for senior embodied-intelligence specialists, including roles focused specifically on humanoid-robot development. 6
ByteDance does not need to become the dominant manufacturer of every robot for this strategy to matter. It could create value by supplying part of the “brain” layer to industrial, collaborative or service-robot partners.
Baidu: robotaxis, autonomy software and AI compute
Baidu’s physical-AI strategy is most visible in autonomous driving. Apollo combines the software and data systems required for driverless vehicles, while Baidu’s Kunlun chip program gives the company a compute layer that can support AI workloads.
At Apollo Day, Baidu said its second-generation Kunlun AI chip had completed end-to-end performance adaptation for autonomous driving. The company also highlighted an autonomous-driving perception model, high-definition maps and a closed-loop driving-data system. 44
Baidu has continued to map out new AI hardware. Its Kunlun M100 was scheduled for large-scale inference in 2026, while the M300 was scheduled for multimodal-model training and inference in 2027. 40
The strategic significance is vertical coordination: AI models and compute are paired with autonomy software and fleet operations. But robotaxis remain a demanding test case. Commercial expansion depends on safety performance, vehicle partners, local permissions and sustained operational economics—not just a strong model or chip.
What this means for Alibaba and Tencent—and for industry partners
Alibaba and Tencent are relevant to the broader trend because large platforms can contribute different layers of the physical-AI stack: cloud infrastructure, enterprise software, commerce networks, simulation, capital and partnerships. However, the source material here provides substantially more concrete deployment evidence for Meituan, JD.com, ByteDance and Baidu than for specific current Alibaba or Tencent initiatives.
For traditional industries, this shift changes the partnership equation:
- Retailers and logistics operators need automation that works with dispatch, inventory, loading zones, access control and human exception handling.
- Manufacturers remain essential for components, reliability, integration and field service, while platform companies may become influential providers of AI software and operating standards.
- Automakers increasingly need autonomy, mapping, compute and fleet-management partners alongside their traditional vehicle-engineering capabilities.
- Robot vendors may need access to deployment environments and operational data as much as they need better hardware.
China’s advantage—and the limits of the thesis
China’s potential advantage is not simply that one company has the best model. It is the possibility of combining high-volume commerce and delivery networks, substantial manufacturing capacity and fast operational learning in the same market.
Meituan’s regulated drone routes, JD’s repeatable warehouse deployments, ByteDance’s embodied-AI research and Baidu’s autonomy-and-compute stack illustrate different ways to build that loop. Each links AI development to a live environment where performance can be tested and improved. 54
17
3
44
Still, physical AI is less forgiving than digital software. A delivery failure, vehicle error or warehouse disruption carries safety, liability and service consequences. Large pilots do not by themselves establish durable profitability, broad autonomy or universal labor replacement.
The most credible near-term outcome is targeted automation in constrained settings: warehouses, fixed delivery corridors, well-defined industrial tasks and managed fleets. The longer-term contest will be over who can best combine models, chips, machines, operations and trusted access to real-world deployment.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "How China’s Internet Giants Are Turning AI Into Delivery Fleets, Warehouses and Robotaxis"?
China’s internet giants are shifting from digital platforms to physical AI systems: Meituan is commercializing drone logistics, JD.com is replicating automated warehouses, ByteDance is developing embodied AI robot mod...
What are the key points to validate first?
China’s internet giants are shifting from digital platforms to physical AI systems: Meituan is commercializing drone logistics, JD.com is replicating automated warehouses, ByteDance is developing embodied AI robot mod... The clearest near term deployments are in structured, repeatable environments—warehouse picking, fixed drone corridors and fleet operations—rather than fully general purpose robots replacing people everywhere.
What should I do next in practice?
For manufacturers, retailers and mobility companies, the shift makes partnerships with AI platform, cloud, chip and fleet operations providers increasingly important.
Sources
yicaiglobal.comMeituan's Drone Delivery Infrastructure Goes Into ...
chinadaily.com.cnMeituan secures first nationwide low-altitude logistics operating ...
thelowdown.momentum.asiaMeituan launches drone delivery in Hong Kong - The Low Down - Momentum Works
en.people.cnOver 6,000 autonomous delivery vehicles in operation across ...