The Oxide approach, I think, would be to pick an open interconnect standard and work with one or more NVIDIA competitors to make hardware for it. E.g. AMD and a few of the specialized AI hardware companies. Essentially provide an open TPU standard.