the efficiency point is simple. inference providers can use RAM to make money and so they are willing to pay more for it. they don't need to confiscate it. they can outbid consumers and reduce effective consumer demand, which is exactly what they are already doing.

the underlying economics are unescapable. when there was a glut of RAM because no profitable uses existed for it manufacturers shoved it into consumer products because "bigger number is good" but it produces very little utility sitting idle in consumer devices.