[3rd3]

As others have noted, it is more about the maximum complexity increasing than mean or median. Simple structures keep existing as long as they have their niche, and a human's niche is not (yet) that of viruses.

This also reminds of Gall's law that complex systems evolve from simpler ones.

You can also see it in neural nets, where larger ones have a higher spatiotemporal resolution and can do more complex things.

More model capacity allows to model the environment and self more accurately which allows to outperform other structures in negentropy consumption often at the cost of the other structures (zero sum).

This exerts selection pressure toward increasing complexity.

That also largely explains group and country disparities.

I am not sure that non-evolving things really fit into the same pattern. A burning fire does not necessarily displace inert matter, nor did it arise from competition.

Physics and chemistry are more fractal-like possibly the result of enumeration of all computational structures (see Tegmark's mathematical universe hypothesis or Wolfram's ideas on the computational universe). Not fractal-like in terms of self-similarity (although there is some at different scales), but fractal-like in terms of chaotic complexity like a pseudorandom number generator but with more rule-like structures in between. Wolfram also classified such computational patterns.