pull down to refresh
My usecase would likely be a combination of archival of resources (to preserve them against censorship and to include nonstandard info) and LLMs (to allow for private controllable use of that info). I also might like to be able to generate children from the parent LLM to be able to be used in a disconnected environment. I think it's also a fine balancing act. I think current AI is provably a bubble. That means that competing hardware may decrease in price the future, however the state also likely sees AI as a huge opportunity to consolidate compute of existing archival and collection of data via bulk data collection processes and systems. An interesting juxtaposition for the future price of compute and hardware broadly. Do you consider NVRAM in your considerations? As in, may RAM remain expensive, but storage not or vice versa?
Depends on your usecase. Storage? AI inference? coding, video editing, streaming?
If its anything other than running local LLMs, you don't have to worry so much. IF there are restrictions on compute, it would be mainly to prevent people from running powerful models at home. All other usecases need much fewer resources. They're not going to stop you from buying a raspberry pi. They might want to make it harder for you to buy 100gb+ of RAM, or top end graphics card.
I would say build one strong machine for your usescase with the best specs you can afford today. If its for AI, its already getting prohibitively expensive for most, and it won't get any better. Prices are likely to increase for everything compute related for a long time to come. That might be the biggest blocker, if not arbitrary laws.