without fail, every single company that tried to forecast their power consumption has got it wrong (usually at least 20-30% underestimating), so its always best to try it after the chips come back from the fab to see what the actual power consumption is.
also, as you say, most power we measure is at the wall... because there are significant losses due to dc/dc power conversion, and ac/dc power conversion, and cooling systems, and host computers, etc... so its best to compare like for like and compare them all at the wall.