The structural case for specialized AI inference providers

I've been thinking about why specialized inference providers keep appearing instead of everything being vertically integrated into model companies. The logic is pretty straightforward once you see it.