This presentation explores a method for safe GPU programming in Rust by leveraging Rust's compiler guarantees, such as ownership and alias analysis, to improve runtime performance for HPC and machine learning applications. The research focuses on how these safety features can be used to optimize code for GPUs, aiming to provide both high performance and memory safety without the typical pitfalls of GPU programming. This work is relevant for developers looking to use Rust for high-performance computing on GPU hardware within the LLVM ecosystem.