XPENG has revealed new details about X World, its AI-driven simulation platform designed to train and validate next-generation intelligent driving systems. The company also introduced X Cache, an acceleration technique that XPENG says can boost AI simulation efficiency by up to 2.7x. Together, these technologies support the development of XPENG’s VLA 2.0 architecture and the company’s broader initiative in what it calls Physical AI.
A digital driving simulator for AI
At its core, X World is an advanced, highly realistic driving simulator built for artificial intelligence. Much like pilots train in flight simulators, XPENG uses X World to recreate complex traffic situations digitally, enabling intelligent driving systems to learn, adapt and improve before encountering similar scenarios on real roads.
The platform can simulate millions of different driving situations in a virtual environment, including dense urban traffic, sudden lane changes, heavy rain, unexpected pedestrian behavior and other scenarios that traditionally require extensive real-world testing.
This approach reduces the need for costly test miles while improving development speed, scalability and safety validation.
XPENG gives examples such as a cyclist suddenly appearing in central Amsterdam, heavy rain on a German autobahn, or complex urban traffic in Paris. Scenarios that would normally require millions of kilometers of real-world driving can now be recreated digitally and tested at scale.
X Cache speeds up AI simulations
XPENG also unveiled X Cache, a new acceleration technique designed for AI-based world models.
One major challenge in large-scale driving simulation is the massive compute needed to continuously generate realistic traffic environments in real time. XPENG says X Cache addresses this by identifying reusable visual information across video frames, reducing redundant computation and increasing inference speed by up to 2.7x.
This improvement is achieved without retraining the underlying model.
In practice, that means X World can run faster, more efficiently and at larger scale, bringing real-time AI simulation closer to practical use in autonomous and intelligent vehicle development.
Part of XPENG’s Physical AI strategy
X World and X Cache together form part of XPENG’s broader Physical AI strategy, which integrates artificial intelligence not only in digital interfaces but also into real physical products.
For XPENG, this means developing AI technologies for vehicles, robotics and future mobility solutions.
The technologies are already used internally across XPENG’s autonomous driving development processes for:
• Closed-loop simulation testing
• AI training and reinforcement learning
• Synthetic data generation
• Safety and performance validation
Simulation capacity is expanding rapidly
XPENG reports rapid growth in its simulation capacity over the past year.
According to the company, the number of simulated scenarios has increased from around 30,000 to over 500,000. Daily virtual testing now equates to roughly 30 million kilometers of driving.
XPENG argues that world models will play an increasingly important role in the future of intelligent driving by enabling faster AI training, large-scale validation, and the creation of safer AI systems for diverse global traffic environments.
This development also highlights how the automotive industry is increasingly relying on advanced AI simulation to lower development costs, shorten testing cycles and enhance safety before technologies reach public roads.
With X World and X Cache, XPENG aims to strengthen its position in the race for AI-driven mobility, where simulation platforms, world models and Physical AI are expected to be central to the next generation of intelligent vehicles.
Technical reports and research
XPENG has published technical reports and research materials about these technologies through arXiv and the company’s official research channels.
Among the shared materials are the X World official site, a paper titled “X-Cache: Cross-Chunk Block Caching for Few-Step Autoregressive World Models Inference,” and an arXiv entry documenting the research.
These publications illustrate how AI simulation is quickly becoming an integral part of autonomous and intelligent vehicle development, where digital training in virtual environments can complement—and in some cases rival—physical road testing.
About XPENG
XPENG is a leading technology company that designs, develops, manufactures and markets smart mobility solutions. Its mission is to accelerate the shift to smart electric vehicles through innovative technology and to shape the future mobility experience.
To optimize the user experience, the company has built a full-stack Advanced Driver Assistance System (XPILOT) and a smart operating system (Xmart OS) to enhance in-car functionality. XPENG has also developed core vehicle systems to improve driving dynamics, including electric drivetrains and advanced electronic architecture.
XPENG is publicly listed in New York and Hong Kong and is headquartered in Guangzhou, China, with regional offices in Beijing, Shanghai, Silicon Valley, San Diego and Amsterdam. The company’s smart electric vehicles are primarily manufactured at its facilities in Zhaoqing and Guangzhou in Guangdong Province.