Considering the amd evo-x2 is half the price of the nvidia dgx spark and still runs llms faster most of the time I'd say you don't need nvidia. It may be better for training models, but for running them amd is fine. With the nvidia cards not being sold in China but amd cards are and their own chips, Microsoft, Google, IBM and more using their own custom chips, developers aren't relying on cuda so much anymore.