I'm not sure this information is grounded, but I remember to have read somewhere the inference is indeed profitable. My personal experience is similar. Running 2x3090s draw 500-600W and you can locally run amazing models with a similar setup.
> inference is indeed profitable