XPeng, a leading player in the electric vehicle market, according to Li Liyun, XPeng’s head of autonomous driving, the real competition is no longer confined to the vehicles themselves but is increasingly taking place in the cloud.
XPeng’s commitment to cloud-based infrastructure for autonomous driving is evident in the numbers. The company boasts a staggering 2.51 EFLOPS of computing power dedicated to training its AI models. This puts XPeng at the forefront of the industry in terms of raw computational capability.
At the heart of XPeng’s strategy lies a massive dataset comprising 20 million data clips. This vast repository of information serves as the foundation for the company’s machine learning algorithms, enabling more refined and accurate autonomous driving capabilities.
XPeng’s approach to development is characterized by speed and efficiency. The company iterates its models every two days, a pace that allows for rapid improvements and adaptations. Perhaps most impressively, XPeng reports a tenfold increase in performance every 18 months, showcasing the power of their cloud-based infrastructure.
Li Liyun emphasizes that while end-to-end AI driving technology is crucial, it’s merely the starting point. The true differentiation comes from the cloud-based infrastructure that supports these systems.
The process of improving autonomous driving capabilities is cyclical. Data collected from vehicles undergoes extensive cloud-based training using large AI models. This training is enhanced by world models, simulations, and knowledge distillation before being redeployed to the vehicles. This continuous loop of improvement relies heavily on advanced AI architecture and robust cloud infrastructure.
To achieve its cloud computing goals, XPeng has partnered with Alibaba Cloud since 2022. Together, they’ve built a large AI-native data center, achieving high levels of computing acceleration and stable performance.
While XPeng’s numbers are impressive, it’s important to note the challenges in making direct comparisons with other companies. The definition of a “data clip” can vary between organizations, making cross-company comparisons less meaningful. Similarly, raw computing power (measured in EFLOPS) doesn’t tell the whole story; efficiency in utilizing that power is equally crucial.
XPeng’s strength lies in its holistic approach to autonomous driving technology. The company excels in algorithms, data collection and processing, and efficient use of computing resources. While these strengths may not always be easily quantified, they manifest in the user experience of XPeng’s products.
Related Post
XPeng Tech Day Showcase: Next-Gen Humanoid Robot and Flying Car Horizons
Xpeng Autonomous Driving Vision: Plans to Launch a Highly Competitive Robotaxi by 2026
XPeng XNGP: Unleashing Nationwide Autonomous Driving with a Bold Blacklist Approach