Server.accept will block (wait) until a new connection happens. Once the callback (in the form of a Ruby block) completes the thread will end. So it starts a potentially infinite number of threads but only one per connection and each one is terminated pretty quickly. This is a pretty common way to write a server that can handle multiple simultaneous connections.
Ruby (and most languages) evaluates the arguments before passing them through to the function. So it first evaluates server.accept, which blocks until a new connection, then passes the return value through to Thread.new which spawns the new thread.
The parameters to Thread.new are just passed straight through to the block.