I dislike this kind of news coverage very much. It is designed to tease rather than inform. Although, I have to admit I was intrigued less by the claim that the game procedurally generates every atom, and more by the question it begs of how you'd actually generate every atom.

Usually, of course, atoms don't matter. The ideal gas law, for example, is essentially a "rule of thumb" which gives you useful ways to predict the behavior of large ensembles of atoms; e.g. to average over the movements of statistically significant (avagadro's number or so) individual particles - indeed, one of the most amazing things in physics is the connection between Newtonian physics and thermodynamics via statistical mechanics.

In any event, my naive answer to the question of what should we simulate would be "don't simulate anything you don't have to" which means that unless you have scanning electron microscopes in the game you don't simulate atoms at all. You mostly use approximations. In games, collisions are important so surfaces (and their properties) tend to be important. And so those simple geometries defines the data structure you use to define the world. In essence mass is defined in a computer program to be a volume that behaves a certain way in the presence of acceleration. But there is no need to describe materials as a lattice of much smaller particles. It's almost never relevant to the simulation.

So, yeah, I don't think it's reasonable to expect a game to simulate a world such that ad hoc chemical reactions can take place, etc. But it's not unreasonable to expect in-game scanning electron microscopes to be able to realistically resolve the details of any material.