There’s a bigger problem here which is that models themselves are general commodities and there’s just not enough difference between them for any one player to differentiate themselves. A company can get ahead of others by a few months, but then the rest quickly close the gap. It’s a really low margin business because you constantly have to burn a ton of money just to stay a bit ahead, and you have diminishing returns the longer you do it.

The only rational approach is to treat models as shared infrastructure akin to Linux because the money is going to be in customization niches. Companies will charge to tune models for specific use cases and charge support for that. There’s also going to be money at the bottom for hardware vendors making chips and memory. But the middle tier of generic LLMs is just relentless involution driving profits towards the bottom.