I think this all hinges on when or if these things can get produced/maintained for less than or close to the cost of a human soldier. Imagine if these got down to $10,000 to produce. Then it would be cheaper to arm a 1-2MM unit militia of humanoid robots than it would actual human marines.
Human soldiers are exceptionally expensive. You can see that in both traditional foot-soldiers and in other areas like ships and planes, where the costs skyrocket because of needing to design around / for the safety of people first.
Robot soldiers will be far less expensive than human soldiers in every respect 50 years from now. A robot soldier can be powered down, and stop consuming resources when not needed (which is most of the time).
If you could produce a robot soldier that did what humans do, with a 5 year shelf life, and you could sell it for $400k to $500k, they'd sell like crazy. Just match the 5 year cost of a human soldier, then by removing the human soldier casualty angle it becomes well worth it to a government.
The value of a combat robot is in avoiding casualties rather than cost savings. Casualties lose America wars, because it is a democracy, and it fights without an existential threat.
However, I think your numbers for the cost of humans are way low. The US army put the cost to train a soldier at $150'000. In addition, there is pay, and veterans benefits. The US will spend 1.5 trillion on veterans between now and 2022.
