Egregoros

@Goalkeeper @caekislove @cjd Inference demand is gonna feed NVDIA for a century. If you want to run open source, you still need GPUs with a ton of VRAM. Whenever someone wants AI anywhere (except China), NVDIA will make money.

I think what this reveals is labs are on the bottom of the infra stack, and treat themselves like the top of the value chain instead of the rock-solid foundation. Right now, they’re arguing that they need to be given a de facto duopoly on development until they can get big enough that their AGI does everything and then they can pay back their investors. In reality, Microsoft is fine using OA or Kimi for copilot, it barely makes a difference to them. Kimi is way cheaper btw, because they can run it on their own hardware without paying an API fee to OA.