My guess is that the Tesla Autopilot Software team is working on language/compiler abstractions to reduce their dependency on semiconductor hardware manufacturers, particularly NVIDIA [1]. NVIDIA hardware has been Tesla cars for many years, and they have a dominating position in both embedded and data center hardware for self-driving AI, and they are not afraid to partner with Tesla's competitors [2]. This whole ecosystem is based on CUDA toolchain, which is built atop LLVM [3]. And Chris Lattner is the original developer of LLVM.
Considering the recent hire of Tesla's new Vice President of Autopilot Hardware Engineering team was Jim Keller[0], the lead architect for AMD's K8 (Athlon 64) and a very influential force behind both Apple's A* chips and AMD's Zen (Ryzen), I think you're right that they're looking to rely less on NVIDIA.
My guess would be that Tesla's ambitions are set on controlling the whole stack. Something along the lines of a custom software toolchain for tailor-built silicon.
Or perhaps they want to go down the road of FPGA / ASIC. There is some LLVM based FPGA tooling around today [1], but perhaps they want to uplevel the core programming language to a new programming language? Chris has done this already from Obj-C to Swift.
My guess would be that Tesla's ambitions are set on controlling the whole stack. Something along the lines of a custom software toolchain for tailor-built silicon.
[0] https://en.wikipedia.org/wiki/Jim_Keller_(engineer)