Because if you don't, someone who does is going to invade for energy and resources. Therefore, refusing to expand is a slow form of suicide, unless you can be somehow certain there's no entity like that in the potentially accessible space.

At some point growth in any star system is going to be finished - energy extraction near maximum, no free matter left to convert into computronium. After that, only two solutions remain: stagnation (death) or expansion. Assuming no instant communication (not just FTL - instant), the bandwidth argument only means it makes more sense to haul energy and computronium from the rest of the accessible space into one point.

We may very well already see it somewhere far away, thinking it's "natural" rather than a construct. I think under current knowledge the most likely prediction is something resembling a sphere (minimizing distances between points) that utilizes black hole power plants, feeding them matter from stars and nebulae.

From the outside, it would look like a giant void with nothing in the center. Maybe there's something like that in the center of the Boötes void?

I've heard that described as "where are the supernova defense systems?" Of course, I'm not aware of any explanations for why it's physically impossible to have a relatively small system sufficiently "hardened" to survive a supernova, or perhaps a way to defend oneself (including hiding) that doesn't require stasis.

why do it?

Everyone doesn't have to do it. Just one person out of billions does. Think of the strange and perverse collection of behaviors we observe in a population of a few hundred million people online and speaking English. Expand that to billions of hyperintelligent minds and somebody is going to want to colonize the universe.

But anyway back to the discussion:

>The mass of one planet being able to simulate billions of planets is only possible if the simulation is rather poor

The mass of one solar system! And you don't need to simulate the system to a planck length. In fact the book specifically addresses that scenario - different simulations have different planck lengths and planck times, depending on the needs of the users of that simulation. For most users (certainly your average modern baseline human) that length might be on the orders of fractions of a mm, and some users might not require a physically meaningful interface at all!

>Everyone doesn't have to do it. Just one person out of billions does.

That's assuming that the culture doesn't have significant bans on von-neumann style self-replicating machines. If they allowed such devices of course you'd expect an expanding light cone of computronium around every significantly advanced culture but any sane long-lived culture would recognize that such machines are just about the only existential threat that a culture with those capabilities would face. That might significantly limit the 'explorer' class to those entities willing to do the equivalent of taking an ice pick to the frontal cortex in modern humans in order to undertake a multi-multi-millennia solitary journey into the void. Some fraction of those trips would fail, and the successes would take a non-trivial amount of time to launch another. Maybe those growth rates are smaller than we would suppose and the universe has only had stars capable of producing the elements of life for a little while on a universal scale, maybe those light cones just haven't reached this corner of space yet.