Port congestion idles record 4.31m TEU of boxship capacity, above the 2022 pandemic peak

The Analysis

Spread the map out on the table and put your finger on one estuary. Outside Shanghai there are ships at anchor that between them account for more than 1.5 million TEU of capacity. One anchorage. More than a million and a half boxes of carrying capacity sitting in the same patch of water, burning days, going nowhere. Linerlytica counted 4.31m TEU of container capacity tied up in port congestion this week, which beats the 4.0m TEU that marked the pandemic-era peak of 2022. It is a record, and it landed in the middle of an otherwise ordinary week.

The number is less interesting than the ratio underneath it. In 2022 the stuck 4.0m TEU sat inside a world fleet of 25.3m TEU, which is 15.7 percent, a seventh of everything afloat parked and waiting. Today the 4.31m TEU sits inside a fleet of 34.4m TEU, which is 12.6 percent, and on that comparison the system looks calmer than it did four years ago. Anyone who reads only this line will tell you the problem is smaller now. They are reading the wrong line.

The right line is the one about slack. Of more than 5,400 container ships trading today, 55 are idle. That is 164,000 TEU, half a percent of the fleet. Half a percent is not a buffer, it is a rounding error. In 2022 you could still go looking for a ship: charter something, slow a service down a little less, pull an extra loader off a loop that had time to spare. There is nothing to pull now. The fleet grew by roughly nine million TEU in four years and all of that growth is already working. What has actually changed is not the congestion. It is the net. Ten years ago a queue this size would have held the front page for a month; today it is one paragraph on a Thursday.

Let me put 12.6 percent into the units I actually plan with. Assume you buy space on a typical Asia to North Europe loop, twelve ships, weekly sailing, roughly a twelve-week round trip including the return leg. Assume congestion removes capacity proportionally, so 12.6 percent of the capacity on that loop is unavailable in any given week. Twelve ships at 12.6 percent is about one and a half ships. On a weekly service that means you lose roughly one sailing every eight weeks. Not one in eight cancelled, which you would see on a booking screen, but one that was never there: absorbed silently by tighter allocations, by rolled bookings, by quietly dropping the least profitable cargo. If you have felt this year that space was harder to come by than the demand numbers justified, this is the mechanism.

Sea-Intelligence has the schedule side of the same story. Only about two-thirds of container ships are arriving on schedule, and the average delay among the rest runs past five days. Two-thirds means one sailing in three misses. Run four sailings a month on a lane and that is one and a bit late every month, every month, for as long as it lasts. Five days is not a rounding error on a thirty-two-day transit. It is a sixth of the door-to-door time, and it arrives on a distribution that never tells you in advance which sailing it lands on. The mental model to hold is not that my transit is thirty-two days plus five. It is that my transit is thirty-two days, and in one month out of three it is thirty-seven or worse.

Here is the arithmetic I would run, and I want the assumptions written out because the answer moves with them. Assume you move 300 TEU a month from East Asia into North Europe, which is ten TEU a day on a thirty-day month. Assume your nominal door-to-door is thirty-four days, thirty at sea and four across both ends. Assume the reliability figure holds, so one arrival in three is late and the late ones average five days. That gives an expected delay of about 1.7 days weighted across the month, but planning on the mean is exactly how people get caught.

The thing that costs money is the spread. Assume the standard deviation of arrival was about two days when berths were predictable and is now about six days, because the queue outside the berth is a variable you do not control. Assume you want a 95 percent service level, which puts z near 1.65. Safety stock is z times sigma times daily demand. Before: 1.65 x 2 x 10, which is 33 TEU. Now: 1.65 x 6 x 10, which is 99 TEU. That is 66 extra boxes of buffer you are funding permanently, for as long as the queue behaves this way.

Assume each box carries 40,000 euros of landed value and your annual carrying cost is 18 percent once you count capital, warehousing, insurance and shrinkage. That is 7,200 euros per box per year. Sixty-six boxes costs 475,200 euros a year. Spread across 3,600 boxes of annual volume, that is about 132 euros per TEU of pure buffer cost before anything else has gone wrong.

Now test whether changing the loading port fixes it. Assume an alternative port gets you a shorter anchorage wait and pulls arrival sigma from six days down to four. Safety stock falls to 1.65 x 4 x 10, which is 66 TEU, you release 33 boxes and you save 33 x 7,200, which is 237,600 euros a year. Assume the alternative costs two extra days of feeder or inland leg and 60 euros a box more in feeder, drayage and handling. Two extra days at ten TEU a day is twenty more boxes in transit, twenty times 7,200, which is 144,000 euros. The 60 euros across 3,600 boxes is 216,000 euros.

Total cost of switching: 360,000 euros. Benefit: 237,600 euros. You do not move. Turn the equation around and the threshold appears. The switch only pays once the new port's arrival spread is under about three days. At exactly three days you release 49 boxes and save 352,800 euros, which is still marginally short of 360,000. So the question for your forwarder is not whether the other port is cheaper. The question is what the observed spread of arrival times at the other port has been over the last twelve weeks, expressed in days. A port that is nominally faster but equally variable buys you nothing at all.

The geography matters more than the arithmetic, because the arithmetic assumes the map holds still and the map never holds still. Shanghai alone accounts for more than 1.5m TEU of the global 4.31m, which is over a third of all the stuck capacity on earth in one estuary, and part of that is typhoon season, which is a seasonal fact of doing business in this part of Asia rather than a failure of anybody's planning. Mumbai currently has the worst queue ratio in the world, measured as anchored ships per berth, and that says something different from Shanghai's absolute number: Shanghai is enormous and backed up, while Mumbai is short of berths relative to the ships calling there. Singapore, Busan, Colombo and several Chinese ports are carrying heavy queues at the same time. Four years ago you could read congestion as a single-port event and route around it. When five hubs on the same map are heavy simultaneously, routing around it stops being a tactic and becomes a sourcing decision.

Then there are the two narrow places. The Panama Canal authority has been cutting both daily transit slots and maximum draft, and the last restriction period is worth remembering: ships discharged containers and moved them by rail across the isthmus to make draft, which works until the rail side runs out of slots. Panama has said it will prioritise the large box ships when it allocates transits, and with the Pacific in an El Nino pattern the autumn delay picture there is expected to worsen rather than improve. On the other side of the world, every ship still routing around the Cape of Good Hope is consuming capacity for longer, and that is one reason Maersk and Hapag-Lloyd have been moving parts of the Gemini network back through Suez and the Red Sea, with MSC telling customers it will restore some Red Sea services as well.

The Suez Canal Authority puts the saving at up to fourteen days of sailing time on the return, plus the fuel. Fourteen days on a round trip is capacity handed back for free. Assume some of that returns over the next two to four quarters and the effective fleet grows without a single new ship being delivered. Assume it does not, and 12.6 percent stays where it is or drifts higher into next year.

A decade-scale view earns its keep at exactly this point. Berths, yards, gantry cranes, channel depth and the rail spurs behind a terminal are ten-year assets, and they are built against assumptions about fleet size and routing that were made long before anyone knew about any of this. Chinese port operators cleared the typhoon backlog quickly, and that operational recovery is worth more than any single scheduling trick, but operational recovery is measured in days while the queue is being fed by structural things: a fleet that grew faster than its berths did, a network permanently lengthened by rerouting, and demand that keeps arriving in lumps rather than in a smooth line.

New ships keep being delivered and some stuck capacity will come back that way. The expectation inside the trade is that congestion persists and the stuck number does not fall quickly, and the carriers have raised their 2026 guidance on the strength of a tight market, which is about as clear a signal as you will get that the people who own the ships expect this to run into next year. Plan against it lasting. If it eases early you have paid some carrying cost you did not need to pay. If you plan against it easing and it does not, you spend a year explaining stockouts.

It is worth saying plainly what the 2022 comparison does and does not tell you. A lower share of a bigger fleet means the industry has more absolute capacity to work with than it did four years ago, and that is real: 34.4m TEU against 25.3m TEU is a third more box carrying capacity afloat. What it does not mean is that any of that extra capacity is spare. Every point of that growth has been absorbed by cargo volumes, by longer voyages, and by the simple fact that a ship stuck at anchor for five days is carrying nothing for those five days. Percentage of fleet is a statistic for people writing about the market. Slack is the statistic for people moving cargo, and slack is half a percent. Hold both numbers in your head and do not let the comfortable one drive the plan.

There is also an inland consequence that follows from where the ships are waiting. When a third of the fleet is late by more than five days, the arrivals do not come evenly; they come in bunches, and a bunch of arrivals is a yard problem, a chassis problem and a labour problem long before it is a berth problem. Assume your discharge port can absorb the ships and that your own receiving operation cannot absorb three weeks of volume landing in four days. That is the failure mode I would insure against, and it is insurable: appointment windows, a named overflow yard, and a pre-agreed weekend receiving slot cost far less than the detention and the expedite fees that follow when the bunch lands and nobody planned for it.

One more consequence that people miss, because it does not show up on a freight invoice. When capacity is tight and the fleet has no slack, the cargo that gets rolled is chosen by someone else's margin logic, not by your urgency. Your protection against that is a volume commitment with a named allocation, and the price of that commitment is flexibility. Assume you are the cargo that rolls until you have signed something that says otherwise. The same applies to contract season: if you negotiate only on price, you are negotiating for the wrong thing in a year when the scarce good is space. Ask for allocation language, ask what happens on a roll-over, ask what happens when a sailing is omitted, and get it in writing at the same desk where you argue about the number.

So, concretely, and with dates. This week, before Friday 28 August, whoever owns inbound planning pulls the last twelve weeks of actual arrival dates by lane and by port and produces one figure for each lane: the standard deviation of arrival in days, not the average. If you cannot produce that figure you are not managing this, you are reacting to it. Deliverable is one page, twelve lanes, two columns, by close of business Friday.

By Friday 4 September, the procurement lead re-bases every open purchase order onto a transit assumption of nominal plus five days for the one-in-three late case, and re-dates the orders where that pushes the promised week out for a customer. Target is zero orders left in the system carrying the old transit assumption by close of business that day. Where a date genuinely cannot be held, the customer hears it this week, not in the week the ship is late.

By Friday 11 September, renegotiate the buffer. Take the sigma you measured, recompute safety stock lane by lane against a 95 percent service level, and take the delta to finance as a single working capital request with the 132 euros per TEU figure attached. Do not present it as an inventory problem. Present it as the price of operating on a network with half a percent of idle ships, because that is what it is and finance will understand that framing faster than they will understand safety stock.

By Monday 21 September, settle the port question. For each of your top three origin ports, get the twelve-week arrival spread from your forwarder in writing, in days, and apply the three-day threshold. Where no port clears three days, stop shopping for ports and start buying inventory: build two to three weeks of additional seasonal cover on the lanes that feed Shanghai, Mumbai and Colombo, funded now while carrying cost is a known number, rather than in November when the space is gone and you are paying for it in air freight instead.

Then set two triggers and put them in the weekly review, because a plan without a trigger is a plan you will not revisit. If the count of idle ships moves below 40 or above 120, re-run the whole calculation from the top: the slack reading is the one that decides whether a congestion spike turns into a capacity crisis or stays a scheduling annoyance. If the measured arrival sigma on any lane widens by more than two days against the twelve-week baseline, escalate that lane to daily monitoring and freeze new order placement on it until the spread narrows again.

On alternatives, and on where the money actually gets lost. Changing port is the obvious move and, as the arithmetic shows, usually the wrong one unless the arrival spread genuinely narrows. The trap is comparing berth waiting times instead of door-to-door variance, because a port with a short queue and a long feeder leg can easily be worse on the number that bites you. Changing mode is the expensive move: air freight solves a stockout and destroys a margin, and the honest way to think about it is as an insurance premium that you decide in advance to pay when a specific trigger fires, not as a freight decision you make under pressure on a Friday afternoon.

Changing sourcing country is the slow move, and it is the only one whose time scale matches the problem: if you are adding a second country to a lane because of congestion risk, you are making a two to three year decision, and you should be qualifying suppliers on that horizon rather than on this quarter's spot availability. Changing inventory is what most people do last and should do first, because carrying cost is knowable and berth queues are not.

The specific traps. Free time: demurrage and detention clocks run on calendar days, and five extra days of berth wait can consume the allowance before your truck reaches the gate. Get the free time in days, in writing, per port, and establish whether the clock starts at discharge or at gate-out, because with a long queue that difference is worth several days of charges. Congestion charges: establish what the surcharge is calculated on, whether per container, per TEU, per bill of lading, or as a percentage of freight, because a per-container charge and a per-TEU charge are not the same number on a forty-foot box, and a percentage charge moves with a freight rate you believed you had fixed.

Transit promises: an advertised transit is a port-to-port figure, and a five-day berth delay at the discharge end does not appear in it. Insist on a door-to-door figure in the contract and name the excluded legs. Allocation: assume you are the cargo that rolls until a signed volume commitment says otherwise. And the seasonal one, which cuts both ways: this week's 4.31m TEU includes typhoon disruption across Chinese ports. Some of it clears with the weather. If you rebuild an entire network plan around a typhoon week you will be over-insured by October and carrying stock you do not need. Separate the weather from the structure before you spend money on either.

I keep coming back to the idle number. Fifty-five ships out of more than fifty-four hundred, half a percent, 164,000 TEU of spare capacity on a planet that moves twenty million boxes a year through its container ports. A decade ago the trade ran with a visible layer of spare tonnage sitting in anchorages and lay-up berths and you could reasonably treat that layer as the system's shock absorber. It has been spent: partly on growth, partly on longer routes, partly on the ordinary fact that ships ordered in one market are delivered into another. What replaces it has to be built on land, in berths and yards and rail spurs, and until it is built the only other thing that can stand in for it is the buffer stock that a trading company keeps on its own balance sheet because the network no longer keeps any for you.

That is not a freight problem and it is not a procurement problem. It is a decision about how much of your own working capital you are willing to lend to a network that has none of its own. I know which way I would answer it, and I would rather answer it in September than in November.

↑ Back to the key points

— By Tomás Ferreira