Because tsmc isn’t charging them enough money to justify a massively increased price.
Nvidia of course is designing the chips, that’s extemely hard, basically the only companies in that side of the business are Nvidia and amd and to a lesser extent Intel. But it also writes a huge amount of supporting software and research to support using Nvidia chips, and that’s the real hard part. Once you depend on Nvidia hardware for software compatibilityyour options to change become very expensive.
Other companies are trying to get into the gpu/accelerator business, google, Amazon, a bunch of Chinese companies etc. But the ones that respect intellectual property are going to be stuck making hardware that isn’t CUDA compatible, and since a lot of the software depends on CUDA or other Nvidia libraries, Nvidia retains an advantage.
On the manufacturing side, while it’s true that the leading edge node for manufacturing is a tough place to be, that’s the corvette or Cadillac portion of GM or the Lamborghini or porsche part of Volkswagen group. Most of the business is lower value lower margin stuff that’s easier to make and more competitive. And tsmc depends on their own suppliers who can sell to competitors (Samsung, Intel, And to a lesser extent say global foundries, IBM, a few others), so tsmc might not be able to retain a node advantage for long, and then Nvidia could have their chips made by other suppliers or multiple suppliers.
Latest Answers