It is said that $XPeng(XPEV.US)'s fourth-quarter new model (internal code F57) will not use any LiDAR, but will instead adopt a pure vision autonomous driving solution similar to Tesla's FSD.

In fact, XPeng has expressed similar ideas about gradually transitioning to a pure vision solution. However, in Tesla's pure vision solution, while the upfront hardware costs have indeed decreased, the large amount of information collected by cameras requires sufficient training and inference computing power on the backend. This shifts the core of autonomous driving competition back to the accumulation of computing power and algorithmic capabilities.

As a result, even after purchasing NVIDIA's GPUs, Tesla still needs to develop its own DOJO computing power. Domestic players face a clear ceiling in stacking computing power, with most autonomous driving and automotive companies relying on cloud computing providers. It remains questionable whether smaller companies like XPeng can financially sustain imitating the practices of big players like Tesla.