Asian Port On-Time Rate Slumps to 32.2%; Shanghai 12.2%, Ningbo-Zhoushan 16.9%

The Analysis

Spread the map out along the Chinese coast and put a mark on every port that feeds the trans-Pacific. Ten years ago the question you asked about this coastline was how many boxes it could lift in an hour. The question that decides your delivery dates now is whether the ships that call here turn up on the day the timetable says they will.

Sea-Intelligence counted 32.2% of container ships arriving on time at Asian ports in August. Roughly one ship in three. The late ones were not late by a few hours: they averaged 7.28 days behind schedule. Inside that, China's port reliability fell to 23.9% from 35.8% in July, with Shanghai at 12.2% and Ningbo-Zhoushan at 16.9%. Singapore, normally at the reliable end of the region, slipped to 32.4% from 42.6%.

And then there is the number that reads like relief. Shanghai's queue fell from 129 ships on 7 September to 47 on 28 September. It is the figure most people will repeat this month, and it is the one to handle carefully, because berthing delays were still running beyond five days while it was being counted, and carriers were telling the trade to expect 15% to 20% less effective trans-Pacific capacity in October. A shorter line at the anchorage is not the same thing as a schedule that works.

Start with the number that matters, and it is not the one in the headline. 32.2% tells you how often. 7.28 days tells you how deep. A ship running a week behind is not a delayed ship, it is a ship that has missed its own next job, because in a liner network a vessel's life is measured in round trips rather than in calendar weeks. When the average late arrival is more than seven days late, the schedule grid is not bent, it is broken, and the sequence of calls that carriers sell, and that every plan downstream inherits, no longer describes anything real.

Here is the arithmetic, and it explains the 15% to 20% that carriers are quoting without anybody having scrapped a ship. Assume a trans-Pacific rotation of 63 days door to door: load in Asia, cross, discharge on the United States West Coast, and return. A weekly service on that rotation needs nine ships, because 63 divided by seven is nine. Now add an average lateness of 7.28 days and assume it is not recovered, which is what not recovered means in practice, since a ship that is a week down does not sail faster to make it up, it simply starts its next rotation late.

The round trip becomes 70.28 days and the same weekly service now needs 10.04 ships. One extra vessel, about 11.6% more capacity, purely to stand still. Add berthing delays still running beyond five days at the loading end and again at the discharge end, and the arithmetic arrives at the 15% to 20% carriers are telling the trade about. No capacity left the fleet. The fleet is the same size and the network is smaller, because the same hulls are turning slower.

Why now is not mysterious. A typhoon season landed on top of a peak season. The storm backlog was still sitting on the yards when the pre-holiday export push arrived, and both had to be worked in the same window. That is what put 129 ships waiting at Shanghai on 7 September. The driver worth naming afterwards is the carriers' own response: a scrambled rotation gets repaired by blanking sailings, and blank sailings are the mechanism by which a schedule problem becomes a capacity problem. The ship exists. The departure does not. Effective is the operative word in effective capacity, because 15% to 20% of it has not been scrapped, it has been cancelled.

Now the part I think is being misread, and it is the reason to be careful with the 47. Shanghai's queue fell from 129 ships to 47 in three weeks and will be quoted as recovery. A queue is a stock; a schedule is a flow. You can drain a stock by letting vessels berth out of sequence, with the berth window going to whoever is ready rather than to whoever was scheduled, and that is exactly what a congested port does. The line at the anchorage shortens while the arrival pattern at the destination gets worse, because vessels that were meant to land within days of each other land a week apart. Berthing delays still running beyond five days while the queue halved is the proof: the anchorage cleared, the berth did not. The congestion did not go away. It moved inside the terminal.

For an importer, out-of-sequence arrival is a worse problem than waiting, and almost nobody plans for it. Waiting you can absorb with safety stock. Sequence you cannot. Three vessels sold as one weekly string land on three different days, and what breaks is not the inventory, it is everything pinned to a date: the rail cut-off, the less-than-truckload consolidation, the labour booked at the distribution centre, and the single consolidated delivery promised to a customer who then receives three partial ones. A week of delay is survivable. Three unplanned receipts inside nine days is a different cost entirely, and it is invisible in every metric that measures how late a ship is rather than how far out of order it arrived.

There is a corollary, and it is the cheapest instrument on this page. If roughly two ships in three arrive outside their window, then the ETA printed on a booking confirmation is not a forecast, it is a placeholder. Any commitment made downstream that inherits that ETA inherits the error, and the error is now measured in weeks rather than days. Most delay in a supply chain is not created at the port. It is imported, weeks earlier, from a date that was never true, and then handed to a customer as though it were a promise. Changing what you promise costs nothing, and it is the only instrument in this whole situation whose price is zero.

Singapore deserves a line of its own, because it is the joint in the region's skeleton. It is not primarily an origin port; it is where boxes from secondary Asian ports are transhipped onto the mainline strings. When Singapore's on-time rate falls from 42.6% to 32.4%, the delay does not stay in Singapore. It is handed to every feeder that connects to it, and those feeders serve ports that never appear in the headline numbers. So 32.2% for Asia understates the reach of this: the ports you would divert to in order to escape Shanghai are, in many cases, connected to Shanghai through Singapore, and they inherit the same week. Spread the map over the region rather than over China and you see why diversion is less available than it looks on paper.

Is this a pulse or a turn? The honest answer has two halves. The typhoon is a pulse: storms pass, and the boxes they hold back get worked off within weeks. The 7.28 days is not all storm. It is what a shock produces when it lands on a network with very little damping, and the damping is what has gone. Ten years ago the same weather would have produced a two-day average, because there were spare ships to inject, slack built into the rotation, and alliance structures flexible enough to reshuffle mid-season. Today the spare tonnage is thin, the strings are fixed for the season, and there is nothing waiting on the sidelines. The shock is a pulse; the propagation is structural. Spread the map over ten years and what has changed is not the weather. It is that the network has less ability to absorb weather, and the same storm now costs roughly three times what it used to.

Spread the same map against ten years ago and the interesting thing is not the ports, it is the slack. In the middle of the last decade a disruption of this shape arrived into a market with too many ships, and the excess tonnage absorbed it: an alliance could add a vessel to a string, a charter could be extended, and the rotation recovered inside a fortnight. That spare layer is what a decade of consolidation, scrapping and disciplined capacity management removed. The network today is tighter, cheaper to run, and far less able to absorb a shock. Which is why the right unit of comparison is not the on-time rate but the recovery time, meaning how many weeks the grid needs to get back to itself. Ten years ago that answer was two to three weeks. With 7.28 days of average lateness carried into a peak season and no slack to absorb it, the honest answer is that recovery waits for the slack season.

There is a second-order effect that lands on the United States end and that almost nobody models. If the average Asian arrival is a week late in August, then discharge peaks arrive bunched and late at the destination terminals, which are asked to receive several weeks of volume inside a compressed window. The congestion does not stop at the origin. It travels, and it arrives at a port that planned its labour and its yard against a smooth flow. A shipper reading only the Asian on-time number is reading half of it: the other half is whether the discharge port is receiving a stream or a series of piles, and the piles are what produce dwell, chassis shortages and appointment backlogs at the destination.

One more piece of arithmetic, because it decides what happens this month rather than next year. If 15% to 20% of effective capacity is gone and the carrier protects its contract customers ahead of everyone else, the shortage shows up in the spot market, and it shows up as a booking that gets confirmed and then rolled. Rolling is not random; it lands on the cargo with the least committed relationship behind it. So the practical question is which part of your volume is protected and which part is exposed, and that can be answered in an afternoon by listing your bookings against your contract terms. If most of the seasonal peak sits outside any commitment, then the 15% to 20% lands almost entirely on the boxes you care about most.

Three conditions would turn this around, and they are worth separating because they have different owners. If import demand softens sharply into the first quarter, the schedule grid repairs itself in one or two rotations and the call that the peak runs into 2027 is wrong, because a network with slack time re-syncs on its own. If carriers re-baseline published rotations with genuine buffer in the contract season, on-time rates recover without a single new crane anywhere in Asia, since reliability is a timetable variable before it is a terminal variable and the timetable is the carriers' to set. If neither happens and berth delays stay above five days into November while the queue stays under 50 ships, then the constraint is inside the terminal, in yard density, gate hours and crane hours, and it will not clear with the weather. That last one is the test, and it is a two-week test.

Build the test yourself rather than waiting for the next monthly release. Take the five strings you actually use and record, for every sailing over the next six weeks, the advertised arrival at the load port and the actual one. Six weeks is enough to see whether the gap is closing or widening, and it is data about your own network rather than about Asia in aggregate. If the average gap is still above five days at the end of November, plan the whole of the first quarter on the assumption that it does not close until after Lunar New Year. If it closes to under three days, the buffer can be released in the January planning cycle. Either way the number you are running on is yours, and it is six weeks fresher than the one in the press.

Put numbers on it. Assume a shipper moving 60 boxes a month eastbound across the Pacific, 720 a year, at an average cargo value of $55,000 a box, with 12 of those 60 boxes tied to retail delivery windows. Assume 8% annual carrying cost. To cover an arrival variance of 7.28 days you need 7.28 days of extra in-transit cover, and at two boxes a day that is about 15 boxes. Fifteen boxes at $55,000 is $825,000 of capital parked in transit, which at 8% costs about $66,000 a year. That is the price of absorbing the variance instead of arguing with it.

Price the other instruments against it. If effective capacity is 15% short and space cannot be had, then on 60 boxes a month roughly 9 boxes a month are at risk of rolling, or 108 a year; at an assumed $400 a rolled box for extra storage, rebooking and one customer penalty, that is about $43,200 a year and rather more in trust. Diverting 20% of the lane to an alternate loading port at an assumed $250 a box inland differential is 12 boxes a month, about $36,000 a year. Air-freighting two of the twelve window boxes when a window is about to break, at an assumed $12,000 premium each, is $24,000 an event. These are not alternatives to choose between; they are layers, and the order is what matters. Time is the cheapest layer, optionality is the second, the emergency fix is the most expensive, and rewriting the promise sits underneath all of them at no cost.

On timing, this is already in the numbers you are being handed rather than in a forecast. 7.28 days is an August measurement. Shanghai's queue was still clearing on 28 September, and berthing was still beyond five days while it cleared. Carriers told the trade in October to plan for 15% to 20% less effective capacity. The transmission is not a question of whether; it is already sitting in your October and November arrivals. What is still open is how long it lasts, and the mechanism that decides that is the one I keep coming back to: a schedule grid carrying a seven-day debt per rotation can only be repaid in a slack window, and the next genuine slack window on the trans-Pacific is after Lunar New Year. That is the entire basis for the into-2027 call, and it is a statement about timetables rather than about demand.

What to do follows from that, and it is mostly a question of which layer gets built first. Build the free one immediately: quote a week rather than a date. Build the cheap one next: 15 boxes of cover on a 60-box lane, decided in the November planning cycle rather than improvised in December. Then buy the option, two test bookings a quarter out of an alternate loading port, Ningbo-Zhoushan or Qingdao, which costs a few thousand dollars and one person's attention and is worth nothing at all if it has never been run. Leave the emergency layer for the emergency. And in the contract season, stop buying transit times: ask for the published rotation with the buffer in it, and treat any transit quoted without buffer as unusable for planning, because that is precisely what it is.

The traps are worth naming because they are all shapes of the same mistake. Reading the falling queue as recovery, when a queue is a stock and the flow is still broken. Chasing the on-time rate month to month, when 12.2% for Shanghai in a single month measures depth rather than forecasting December, and a series that moves twelve points in thirty days is noisy. Solving a schedule problem with inventory alone, when inventory absorbs variance but does not move a rail cut-off, a consolidation window, or a labour booking already made. And assuming the peak ends in January, when a debt carried in the rotation rather than in the boxes only clears when there is slack, not when the calendar turns.

Spread the map out again in March. The ships will be where they always are, and the cranes will be lifting at the rate they always lift. What will have changed is whether the timetable you were sold was a description of the world or a wish about it. Ten years ago this coast was measured in moves per hour, and that was the right unit. Now it is measured in whether the date on the booking means anything, and that is a different kind of network: one where the scarce thing is not steel, or water, or even berths, but the willingness to publish a schedule that is true.

↑ Back to the key points

— By Tomás Ferreira