The big reason why pneumatics aren't used as much in robotics is that air is very compressible, which leads to all sorts of nastiness when you make pneumatic actuators.
Because air is compressible, compressing air is not very efficient compared to say hydraulics. This is bad for exoskeletons.
The other problem is that the compressibility of air limits the 'bandwidth,' or how fast these actuators can actuate and un-actuate controllably, achievable with these actuators. In addition, the bandwidth of pneumatic actuators is often below the frequency of human walking, making them impractical most exoskeleton applications.
And of course, there's more to it than compressibility at play here that could make these actuators impractical. Since these actuators use a rubber membrane, these actuators are subject to hysteresis, significant temperature effects, and creep.
Rubber, when stretched and unstretched quickly, heats up causing the rubber to change it's stiffness. In addition, the temperature of the rubber can also change due the air being pumped into them. Rubber, and other elastomers, also experiences a phenomenon known as creep, where it slowly stretches out with time. In fact, current industrial pneumatic muscles NEVER actuate exactly the same because of all this, and one has to use interesting control approaches.
I am also skeptical that this will be cheaper in practice than mass produced electric actuators. While the actuators themselves are cheap, the valves and other hardware necessary to control them are not. The actuators themselves will certainly wear out sooner than electric actuators(>10 years continuous operation for robot arm actuators) due to the creep mentioned above. With lower efficiency and increased maintenance costs the overall cost of using these could very well be higher.
In fact, the brushless motor in the compressor they show in the video probably can provide" 25% of the power of a normal human elbow," and the only reason it can't be used on a human elbow is that much of that power is at high-speed with low torque. If someone were to develop a compact and efficient gearbox for turning high speed- low torque into low speed-high torque, then one could mass produce it and skip all the pneumatic silliness.(Or just do what that company that robotics google just bought did and use watercooling/overclocking)