If we are in a simulation, it could be running on one or more supercomputers in the “real world.” This concept is important because it implies scarcity and limitations on the computing power our world runs on.
Now imagine the concept of a computer running inside of a computer, much like a program running on your computer. A program running on your computer is limited by the processing power of your computer, much like a computer running inside your computer would be bottlenecked by the computer in which it is running on.
This is where A.I. becomes concerning as it applies to the concept of living in a simulation. As more computers are running A.I. programs at full speed ahead, we will make exponential progress towards the ultimate bottleneck that the world as we know it runs on.
If the total processing power of all computers on Earth becomes greater than the computer running the simulation, we can assume the simulation will crash.
The silver lining is as our processing power increases, we will also slowly reduce the odds that we live in a simulation. The longer we go without glitches or crashes will prove we either live in an unfathomably sophisticated supercomputer, or that we simply live in reality already.
The simulation wouldn't necessarily crash, but it would run slower and slower. All of those simulated computers would chew up enormous amounts of compute in the "real world", and eventually forward progress would grind to a halt.
If I were running a simulation and it just kept getting slower and slower, I would eventually pull the plug.
So it's not a crash that I'm worried about. It's a purposeful shutdown by whoever runs the simulation.