AMD recently announced that its ROCm platform has adopted SPIR-V as a portable intermediate representation for GPU code. This represents a significant transition from ROCm's prior model of compiling GPU code ahead of time (AOT) for each target architecture. The new approach moves to a just-in-time (JIT), "compile-once, specialize-on-device" model, aligning with modern CPU software compilation strategies that address the increasing matrix of target hardware and workloads. According to AMD, this shift enhances portability and flexibility for developers. AMD demonstrated the new compilation model and its trade-offs with a single-kernel HIP example, showing measurements on MI350X (CDNA4) and RX 9070 XT (RDNA4) hardware. AMD stated that "With its adoption of SPIR-V, ROCm now brings the compile-once, specialize-on-device model to AMD GPUs."