The hardware is pretty much all off-the-shelf and low cost. Just a matter of someone open sourcing or “leaking” the design.

Hopefully open source AI is more important to the Chinese government than just an economic attack that they are willing to stop given some economic incentives by the US. Like maybe it helps all the Chinese labs keep a higher AI capability baseline with the shared research and model weights. I suppose we’ll have to see. An agreement to stop open source probably wouldn’t be publicly mentioned because of how unpopular it would be with the tech industry, so we’ll only know if it happened if the Chinese labs stop releasing open weight models over the next several months.

“Non-state actors” = you and me.

Yes, the current latest generation of closed AI models like Mythos and GPT 5.5-Cyber are already being held back from public release. Only the government and a few corporations get to have them. Once there is an open source equivalent, most likely from China, the response from the US will be interesting. An outright ban is unenforceable, but banning US hosting providers from serving it is something i saw proposed. Doing that will have the effect of encouraging more people and companies to self-host the AI rather than renting it from the cloud though - they will end up spreading it more with their restrictions!

I’ve seen evidence that unified memory computers like Apple Studio and AMD Strix Halo can be clustered with RDMA over Ethernet to achieve usable tokens per second, like ~15 t/s with fairly large models. Of course these clusters are not cheap for most consumers especially at today’s hardware prices. I think once “the bubble pops” we will see a lot more interesting developments for locally hosting larger LLMs, like decommissioned servers ending up on eBay and just generally lower hardware costs for DIY AI clusters.