Ask HN: Best path forward for cross-platform GPU compute code in 2024?

I quite like writing code for the GPU but I want to branch out and consider options that would run on both Mac and Windows. It seems the most wide-spread or powerful GPU technologies are quite platform-specific. Both CUDA and Metal are really enjoyable but each is locked to a single OS. There are efforts like SYCL but they have limited support or maturity for all platforms.

What direction should I focus my efforts? What alternatives am I not considering?