logoalt Hacker News

IAmGraydon • today at 7:50 PM • 1 reply • view on HN

I think that Anthropic and OpenAI are going to have to come to terms with the fact that they add value in the model training arena, and that's their product. Deploying infrastructure for the public to use that model is going to be a losing game for them. The problem is the first one appears difficult to monetize, but I think the answer to this is going to be licensing their models to companies like AMD to print to the Taalas architecture. To me, that is the logical future this is all leading to - hardcoded models licensed to be printed direct to silicon for ultra fast inference.


Replies

danw1979 • today at 7:57 PM

Why can’t a company own infrastructure and rent it out directly to end users if we’re already assuming other companies could make money doing the same ?

Or is your point that vendors of hardwired inference kit won’t sell it to one of the most valuable companies on the planet ?