I think they're still drinking the cool aid, but between the development of LLM specialist chips and the delay in GPT5 (because Open AI has got nothing) they're adopting a wait and see approach.
Once it's clear we're reaching the limitation of LLM scaling (and we absolutely are already there.) There will be a correction, either a crash entirely as the bubble collapses or, more likely because number must go up, a move back to the Generalist approach of Nvidia as hybrid and neo-neurosymbolic models take over the forefront of hype.
I think they're still drinking the cool aid, but between the development of LLM specialist chips and the delay in GPT5 (because Open AI has got nothing) they're adopting a wait and see approach.
Once it's clear we're reaching the limitation of LLM scaling (and we absolutely are already there.) There will be a correction, either a crash entirely as the bubble collapses or, more likely because number must go up, a move back to the Generalist approach of Nvidia as hybrid and neo-neurosymbolic models take over the forefront of hype.