At some point I believe we will start to bump up against the limits we saw in the first three computer revolutions (tubes, transistors, integrated circuits). This will cause the pendulum to swing from commodification to optimization. What I mean is, you won't be able to snap your fingers and scale up or out, thus optimization will begin again in earnest. Clearly this isn't always the case: GPU shaders, crypto ASICs, video processing... there are plenty of places where optimization is crucial for high performance loop tuning. But optimization hasn't been required across the board like there was just before the three big innovations I described hit.