> one might think of pumping it “up”, but really the only hard work is producing the pressure difference P_feed - P_permeate or so — water is buoyant to an extent that almost exactly negates its weight

If you package the permeate in a balloon (hopefully a very strong one!) and let the balloon rise to the surface, buoyancy is very relevant.

If you instead pipe it -- to simplify the analysis, let's say it's going straight up a vertical chimney to the surface -- it doesn't seem like buoyancy is relevant.

Take a vertical water-filled pipe sealed at the bottom and open to the air on top. The water N meters from the top of the pipe will be the same pressure as N meters inside the ocean -- even if the pipe's nowhere near an external body of water! A water column self-pressurizes due to the potential gradient of Earth's gravity.

Now put the bottom of the pipe at the bottom of the ocean, you can unseal it and stick a pump on it.

You put 1 kg of water into the pipe N meters below the surface. You take 1 kg of water out at the surface. And repeat in a cycle. Some part of the system has to be doing enough work to lift 1 kg of water N meters per cycle. That work has to come from the pump -- where else could it come from?

I'm skeptical of any notion that water "floats" to the surface of the pipe "for free"!