The 80's joke was "GaAs, the material of the future"
"Always was and always will be."
Back then DARPA was running the "Very High Speed IC" Programme and targeting 32bit MIPS (for the low transistor count) processors at the unbelievable speed of 200Mhz.
GaAs was tough to mfg (IIRC the process was originally developed out of RSRE Malvern, now a bit of Quietq), fragile, expensive and dangerous (Arsenic vapourises readily at the operating temperature). As noted the NMOS process was power hungry. IIRC "holes" which are important in PMOS, that make up the other half of "complimentary", move slower in in GaAs than Silicon, something perhaps someone should have picked up on sooner.
Oh and the oxide layer is a pig. But hopefully InGaAs will be much better.
This is V 0.1 technology. No one will push this unless their backs are really against the wall.