Comment by zozbot234

The most reasonable story you can tell for a nVidia moat is their know-how in designing datacenter-scale hardware and getting it fabbed and deployed. That's inherently hard to replicate. CUDA itself can be replicated in theory (it's basically just a compute API) but that turns out not to be worth it since the nVidia ecosystem really is higher quality for the cost.