TLDR;
- Xpeng unveiled VLA 2.0, a 720-billion-parameter AI model powering vehicles, robots, and flying systems.
- The company will open-source the model to accelerate “Physical AI” innovation across industries.
- Volkswagen becomes the first commercial customer, signaling growing cooperation in AI mobility.
- Open SDKs with AutoNavi aim to build a wider ecosystem for camera-only autonomous systems.
At its 2025 AI Day in Guangzhou, Chinese electric vehicle maker Xpeng introduced what it calls the next chapter in “Physical AI”, an era where artificial intelligence bridges the digital and physical worlds through vehicles, robots, and aerial systems.
The highlight of the event was the unveiling of VLA 2.0, a large-scale AI model trained for visual-to-action reasoning, designed to process real-world images and execute driving or motion tasks directly. Unlike typical perception models, VLA 2.0 is built for real-time, autonomous decision-making across platforms, from electric cars to robotaxis and even flying vehicles.
With 720 billion parameters, the model represents one of the most ambitious AI architectures in the mobility space. Xpeng’s engineers say the system can process complex environmental data at speeds that allow vehicles and machines to make dynamic decisions without relying heavily on cloud inference.
Open-Source Model to Drive Collaboration
In a major strategic move, Xpeng announced that VLA 2.0 will be open-sourced for business partners and developers. The company also revealed plans to release an open SDK in collaboration with AutoNavi, a major Chinese digital mapping firm, to empower startups and enterprises to build self-driving tools, software integrations, and safety validation systems.
The open-source model is licensed under permissive terms, allowing partners to adapt it for various applications , from industrial robots to next-generation autonomous fleets. Xpeng believes this collaborative model will fast-track innovation in edge AI, a field where computation happens directly in vehicles rather than in distant servers.
“Physical AI is not a closed garden,” said an Xpeng spokesperson. “Our goal is to build an ecosystem where intelligence flows across devices, industries, and partners.”
Volkswagen Taps In as Launch Partner
Among the early adopters of VLA 2.0 is Volkswagen, marking a deepening of Sino-European collaboration in AI mobility.
The German automaker will integrate VLA 2.0 into its next-generation smart driving systems, expanding on its existing partnership with Xpeng that began in 2023.
Analysts view Volkswagen’s participation as a strategic endorsement of Xpeng’s technology stack. With projected B2B licensing revenues between RMB 2–3 billion annually by 2027, the deal positions Xpeng as both a technology vendor and innovation leader beyond its core EV business.
Robotaxis and Flying Systems Next in Line
Xpeng’s ambitions go far beyond code. At the same event, the company previewed its 2026 Robotaxi lineup, which will rely solely on camera vision, echoing Tesla’s approach but scaled for the dense complexity of Chinese roads.
Three Robotaxi models are planned for release in 2026, alongside a pilot program whose locations remain undisclosed. The vehicles will be powered by Xpeng’s in-house AI stack and Turing AI chips, which deliver up to 3,000 tera operations per second (TOPS), enabling high-speed decision-making directly on the vehicle.
Xpeng also introduced Next-Gen IRON, a robotic platform for manufacturing and logistics, and two flying systems from ARIDGE, marking the company’s steady advance into aerial mobility.


