This is all just my opinion but I think the competitiveness of AI will force a lot of companies to continue to upgrade their data centres despite causing a lot of CAPEX. For sure, some companies will stick with their old GPUs for a while but eventually will have to upgrade. Just as an example, let's say Anthropic upgrades to Rubin chips but OpenAI is training and running on Blackwell. Anthropic's models would get smarter and quicker and OpenAI would need to upgrade even if doesn't really want to. Basically, in my view, the AI infrastructure arms race will make this annual GPU release schedule make sense. As for the excess inventory for Nvidia, I think they would control their production numbers quite intensely based on orders and any left over previous generation chips should have no problem being marked down a bit and sold to companies that don't need the absolute cutting edge technology. Of course, even the top companies won't upgrade every year but I believe this annual release cycle will fuel the AI arms race. Again, this is all just my opinion.
Comments