Everyone talks about agi timelines, hardly anyone talks about compute gatekeeping

It feels like most of the discourse here assumes the only thing that matters is algorithmic progress or when exactly agi drops. but if compute access stays concentrated in a tiny set of players like open ai and microsoft, isnt that bottleneck just as important as the model breakthroughs themselves? How do you think about centralization of compute vs centralization of ideas. if a few labs own all the h100s, it doesnt really matter if the "ideas" are open source or not. we might end up with a futu