| No dont think that's their plan. One of the key issues of cloud gaming that has failed a few times now is that the datacenter servers need to be damned close to get that sub 100ms. Google didnt have enough datacenters to do it and Stadia died. In a way it's the same problem, they need to get the gpus within probably like 50-100kms of the users. That's plausible to do, but that's doubling or tripling the number of operated datacenters. If you look at chatgpt's stargate. They essentially will have 2 of them worldwide? Anthropic/claude is beholden to AWS compute which is us-east-1 and us-west-2 only. I think AWS plans to open in signapore soon? Elon's Grok is in memphis only; and there's drama around their generators and such? Far too centralized for all the major players and I dont see any even medium term chances of getting out there enough. It's too much power, you pretty much have to park it near major generation. What I think they are still doing is training/research, processing current load. Working on energy efficiency. That tops/wattage ratio but also figuring out caching and other super compute limits and constraints. That was the brilliance of MOE; but a major player ought to have a mixture of mixture of experts. Next big thing for AI is cheaper cost picture and video generation. This is Veo4 or Sora or comfyui. I'm also expecting a better google home or echo type device. Something that's AI and not hard coded skills; and ideally all local and no built in government tap. |