Because their need is growing rapidly, and the adoption of neuromorphic hardware will not keep pace. They don't want to be constrained.

Also, neuromorphic systems are less flexible, whereas computers and GPUs can simulate new AI architectures without changing the hardware.