That is only when running at the *same* switching frequency for transistors... keep in mind some of these technologies are not entirely transistor based so the power usage metrics that would apply for CPUs don't entirely apply to memory... memory is inherently mostly dark anyway though.
Stacked memory like the HBM in AMD's new GPUs being a fine example, or stacked nand... can allow the same data to be transferred at lower frequencies due to physically smaller bus lengths and wider buses. Which leads to lower power per work done.
Think of a physical hard drive... the only active area is directly under the read/write head.
Also, regarding the human brain... its mostly a combined memory/processing unit... interesting really to think about! Moving data from one place to another wastes energy.. and it just so happens that our brains process and store data in the same place and it would also seem that the different types of data it stores (muscle memory etc,thoughts and ideas, are stored in the regions those things are processed...)
If you can fit 400Gb on a micro sd... why not just put it on a the die with the cpu at the very least in the same package! You could literally install static game texture data into your GPU... and never access it from the main system disk after installation again rather loading it strait out of a bandwith optimisted nand or 3d xpoint memory...