Are there aspects of AI that can be just as effectively done on a DIY swarm of GPUs as on A/H100 servers?

I'm curious if there are certain and specific things like AI image generation, database analysis, etc that can be done commercially and effectively as you can on something like an A100, except by small scale people with lots of parallel GPUs like RTX 3000/4000 series?

I was scaling up for crypto mining before that went downhill. Mostly ASICs, but I'm wondering if there would be some kind of commercial use for setups similar to GPU ETH mining rigs, except for …

a100 ai image analysis artificial database diy etc gpus h100 image image generation people scale servers small something