Truth is, Accelerator by now is far from being a potential concurrent for other GPGPU libraries such as CUDA.
Accelerator now just provides access to the ParallelArrays lib which just provide basic operations on arrays, and delayed computation. For basic operations it's great. But when you want to work on advanced GPGPU projects, it's not enough... You don't have enough control on what you're actually programming. While using Accelerator you actually have no idea of what's going on the GPU. I hope the next version will provide more low-level control.