Yeah, I’m betting on major revamp of an AI workstation or even rack server. Desktop/server CPU mainboard with LPCAMM2 or, even better, SOCAMM2 slots and several PCIE x16 for discrete GPUs.
Given all the memory craze we now experiencing, nothing technically prevents them from designing a mainboard with no soldered memory at all. You can buy minimal amount first and later expand it when (and if) it becomes cheaper. Since Framework is targeting geek audience if the first place, it makes total sense to sell computer hardware with no memory installed, thus mitigating the supply hell and yet delivering something new.
Also, local AI inference is really popular right now, I bet, Framework wasn’t expecting that level of demand for 128GB Desktop, so it’s quite natural to do it more. Free money basically.
All of this kinda matches their manifesto. All Matrix references are pointing to matrix multiplication which is the foundation algorithm for all modern AI/ML.
P.S.: Of course this is just my speculation. I would be more than happy to have a humble but oh so useful Framework printer ![]()
