Anyone have insight as to whether these will be useful for ML tasks - mostly on pre-trained models?

I have bulk jobs that I like to do for things like transcription and I'd like to add summarization as well (Whisper + Llama basically). My GTX 1070 is fine for proof-of-concept, but I was planning to build a 4090 box for this stuff.

Nvidia is notorious for nerfing the RTX cards for ML tasks on RAM specifically. Is there any universe in which it makes sense to put my 4090 box budget ($4000+ all in) towards an updated MacBook? I can get a laptop with 128Gb RAM for $4600 and the convenience of not having to run things remote and deal with another OS would be a big win.