Nothing released so far inherently "needs" a datacenter, it's just a matter of how much performance you require. Slow, high-latency inference will be a natural way to run "datacenter" models locally.
Yes it does. You will not be able to run models like DeepSeek v4 (>1.5 trillion parameters) on a regular workstation any time soon, unless by "slow" you mean "unusable". And those are the models that are ~6 months behind Opus 4.7.
What qualifies as "unusable" when I can just run a batch of inferences unattended/overnight and wake up to fresh results the next day? That's a kind of slow workflow that could even be adapted to uses like coding, given enough effort. Besides, you're kinda overstating how heavy DeepSeek V4 Pro really is, the 1.6T are total parameters. They're not all active simultaneously.