Saturation Point
The load level beyond which additional demand produces queueing and latency growth rather than additional completed work.
Throughput rises with load until a resource saturates. Beyond that point, adding load does not add throughput — it adds latency, and eventually it reduces throughput as contention, context switching and retries consume capacity.
The shape matters more than the number. Below saturation, latency is roughly flat. Approaching it, latency rises sharply — the queueing delay term grows as 1/(1−utilisation), so the last 10% of capacity costs far more latency than the first 80%. Past it, the system may collapse rather than plateau.
This is why capacity targets sit well below measured maximum throughput. Running at the saturation point means running with no margin for a traffic spike, a slow dependency or a lost instance.
Finding it requires a load test that increases load progressively and measures latency alongside throughput. A test that reports only "handled 5,000 requests per second" is useless without the latency at which it did so, because handling them at 30 seconds each is not handling them.
The resource that saturates first is frequently not CPU: connection pools, thread pools, downstream quotas and disk I/O bind earlier and are less often instrumented.