Every few years, congestion at some major port becomes a news story, and the coverage treats it as a mystery or a scandal. It is neither. A port is a queueing system: arrivals (vessels), servers (berths, cranes, gates), and service times with variance. Queueing systems have a famous property that every operations textbook states and almost every capacity plan ignores: waiting time does not grow linearly with utilization. It explodes as utilization approaches capacity.
A terminal running at 70 percent utilization absorbs variance easily. The same terminal at 90 percent is one bad week from a spiral, because in a nearly full system every disturbance queues behind the last one. This is why congestion appears "suddenly." The system was not fine and then broken. It was quietly consuming its resilience margin for months, and the nonlinearity did the rest.
Implications for operators
For a carrier or shipper, the practical consequence is that port dwell should never be modeled as a fixed number or even a stable average. It is a state-dependent distribution whose tail fattens dramatically with terminal load. A routing or scheduling engine that ingests live terminal utilization, and knows the queueing curve, will start rerouting and re-sequencing before the delays materialize, because the leading indicator is the utilization, not the delay.
The last slice of utilization buys most of the waiting.
Push arrivals against berth capacity and watch the queue math.
Congestion is not linear in load. It is a queue, and queues explode near saturation.
Implications for the port
For the terminal, the queueing lens reorders the investment conversation. Adding berth capacity moves you down the curve, but so does reducing service time variance, smoothing arrival patterns through slot management, and pricing peak windows honestly. Variance reduction is routinely worth more than raw capacity, and it is dramatically cheaper. The terminals that treat arrival management as a control problem, coordinating vessel speed upstream so ships arrive just in time rather than anchoring in a queue, get more effective capacity out of the same concrete. The mathematics has been available since Erlang. The advantage goes to whoever operationalizes it first in each basin.