Not really, but they are only supported in CUDA. For OpenCL you will have to use clBLAs and clFFT (written and maintained by AMD). This means they are not exactly tuned for other architectures.

I prefer to use OpenCL, but I think the second key factor why Nvidia dominates (the first is they integrated tooling) is that cuBLAS, cuFFT, cuDNN et.al work out of the box when you install CUDA Toolkit. clBLAS and clFFT? Good luck using them even on AMD hardware, yet alone on someone else's. And, of course, there is no cuDNN equivalent (that is beyond experimental)...