Maybe for the competitive consumer graphics market, but I don't see them disrupting Nvidia's hold in the HPC world. With Nvidia's NVLink, you can achieve high bandwidth data transfer between graphics cards without having to pass through the PCIe to CPU. If you have large computations that require syncing data among multiple GPU's, NVLink is your most performant option by a large margin. That's not to mention how much further ahead CUDA is from Sycl, OpenCL, ROCm, etc. I certainly welcome the competition (and hopefully an open standard). In my experience, CUDA is ahead on developer tooling, performance and productivity features (Thrust, Unified Memory, C++17 support, etc).