Discussion about this post

User's avatar
Robots and Chips's avatar

The switching cost argument is absolutly critical and underappreciated in most NVIDIA coverage. Once training infrastructure and inference pipelines are built on CUDA, rewriting everything for competing architectures becomes a multi-million dollar multi-year project. Google's GEO pivot is similarly smart because they control both the search index and the AI layer, which gives them structural advantages that pure AI startups can't replicate.

Expand full comment
1 more comment...

No posts