When does a service hit its scaling limits?

hey everyone, i've been thinking about the point where a single service instance just cant handle more load, no matter how much you throw at it. we all talk about scaling horizontally, but there's got to be a hard limit per instance. what are the key indicators you look for? is it cpu, memory, network i/o, or something else entirely? curious to hear your experiences and how you identify that tipping point before it impacts users.