Key insights
- The article speculates on a long-term risk for Nvidia (NVDA) where advanced AI could eventually design GPUs, potentially eroding its core business. While acknowledging Nvidia's current dominance and potential to leverage its data for AI-driven GPU design, the author suggests AI could also enable competition by designing GPUs or rewriting code to bypass Nvidia's CUDA ecosystem. This is viewed as a distant threat, with Nvidia likely benefiting initially from AI-assisted design.
As the mexican food was shooting from and burning my asshole, I had this light bulb moment. The better the AI gets, the more demand it gets, the more demand there are for GPUs - NVDA is super happy. Until at one point AI gets so good that it can be used to design a GPU, the thing NVDA does. You can yap how CUDA is the real moat, not the GPUs, but if the AI is good enough to design a competitive GPU, it's definitely good enough to rewrite your code on another "library" than CUDA. Just think AI is much more likely to be good at what NVDA does than at what TSMC, Intel do. Although, this will probably take quite a few years, and again ironically, NVDA will be the first to benefit from AI helping the GPU designs as they have the most GPU design data for training the AI for such purposes.