Two thought experiments I like to consider are: - if a company, say, AMD, found a way to produce GPUs at a fraction (say, x=10%) of the price of NVDA, would th

Two thought experiments I like to consider are: - if a company, say, AMD, found a way to produce GPUs at a fraction (say, x=10%) of the price of NVDA, would that increase aggregate demand, or keep it about the same (substitution for nVidia)? Would the price difference be enough to incentivize creation of a CUDA- alternative ecosystem? If not, what does the fraction x need to be? - Very reductively, GPUs seem to be universally needed because they are better at manipulating matrices than alternatives (eg, inverting a matrix, finding dot product, cosine similarity, etc). Are there alternative approaches that could come to market in the next 2-3 years that could be better, or better-per-$, than the current approach of just building bigger GPUs?