China Post, COSCO Shipping and SF Express make 2026 Fortune Global 500

The Analysis

Saturday morning, 06:20, and the depot is already half full of parcels that failed on Friday. The driver pushes the cage in, we scan what came back, and the pile splits itself into three: the ones the morning wave can still carry, the ones that need somebody to make a call, and the ones that should never have been loaded at all. Everybody in the room knows what happens to that third pile. What the customer asks is always the same question in the same words: can it be delivered today. The honest answer depends on three things nobody outside operations thinks about: which network the parcel is sitting in, how many waves that network runs today, and how many hours it has already spent in the hold area.

On 28 July 2026 the Fortune Global 500 came out, and three Chinese logistics and shipping groups are on it: China Post at 119th, COSCO Shipping at 237th and SF Express at 372nd. Separately, SF Holding completed the first tranche of its 2025 A-share buyback, repurchasing roughly 160 million shares for about 5.999 billion yuan against a 6 billion yuan ceiling, with the shares to be cancelled and registered capital reduced. I read both items the way a shift supervisor reads them, which is to say I skip the pride and go straight to the operational question.

What I want to work through here is the half that never makes it into a press release. A revenue ranking tells you how much network exists. It does not tell you what happens to your parcel when the first attempt fails at 14:20, or how many hours you lose at the handover between an ocean leg and a last-mile leg, or how many slots you can actually buy in the last eight weeks before peak. Scale is real. Scale is also not a delivery promise, and the sooner you separate those two in your own planning, the less money you lose in the second half of the year.

Start with what the list actually measures. The Fortune Global 500 ranks companies by revenue, so 119th, 237th and 372nd are statements about scale, not about on-time performance. I am not being cynical about that, I am being precise, because the two get mixed up in planning meetings constantly. Somebody puts three rankings on a slide and concludes that a parcel handed to any of these groups is safe. It is not. What scale buys you on the ground is density: more acceptance points, more waves in twenty-four hours, more vans to redeploy when a route goes wrong, and a genuine ability to run a second attempt without wrecking the rest of the route.

What scale does not buy you is flexibility about your particular parcel. Large networks run on rules, because rules are the only way to hold that many stops together. Address wrong, the rule applies. Phone number missing, the rule applies. The bigger the network, the less room there is for a human being to make an exception for you at 18:40 on a Friday.

The other thing that matters is that these three sit on three different kinds of network. China Post is a postal network, built around universal coverage and a very large number of acceptance and collection points. COSCO Shipping is an ocean network, built around vessel schedules, port calls and container gates. SF Express is an express network, built around sort waves and door-to-door delivery. One shipment of yours very often crosses two of them, and sometimes all three: an ocean leg to the port, a handover to a domestic or overseas carrier, then a postal or express last mile.

Each leg keeps its own clock, and the clocks are set by different physics. An ocean leg runs against a closing time that can sit hours before the vessel moves. An express leg runs against a sort wave that happens two or three times in twenty-four hours. A postal leg runs against a collection and delivery round that may happen once. Put all three on one tracking page and you have a promise nobody can actually manage.

Most of my week goes on the middle of that, the handover. Here is the shape I plan against, and all of it is my assumption rather than a published service level. Line-haul or container arrives at the gateway at 04:10. Handover scan completes by 06:30. Sort window opens at 08:00. Out-for-delivery scan by 11:00. That is roughly seven hours from arrival to the parcel being on a van, three in a good week and eighteen in a bad one, and the reason is never dramatic: one document that does not match, one label that will not read, one cage pushed into the wrong lane.

Write those hours down per lane and set an alert above them. My thresholds are the ones I use everywhere else: escalate when arrival-to-sort passes four hours on a lane where normal is under two, and escalate when out-for-delivery to final scan passes eight hours. If you cannot name the person on the carrier side who owns that handover window, you do not have a lane, you have a hope.

The buyback gets one paragraph from me and no more, because I am not an investor and I am not going to play one. The announced facts are these: roughly 160 million shares repurchased, about 5.999 billion yuan spent, a 6 billion yuan ceiling, shares to be cancelled and registered capital reduced. Two things fall out of that by arithmetic. Execution against the ceiling is 5.999 over 6, which is 99.98 per cent, leaving about 1 million yuan of the ceiling unused. And dividing spend by share count gives an average price of roughly 37.5 yuan per share, which is my arithmetic on their published numbers, not a figure they published.

How you read that is a job for people who are paid for it. The only read I will allow myself, because it touches my side of the business, is this: money going into buyback and cancellation is money not going into unlimited capacity expansion. So do not build your peak plan on the assumption that capacity is freely available in the second half. Lock your slots early, and assume the network will protect its premium products and its largest accounts before it protects you.

Failed delivery behaves differently depending on which of these networks the parcel ends up in, and this is where a lot of shippers get hurt. Inside an express network a failed attempt usually generates a second attempt, often within the following twenty-four hours, and the parcel stays in the van network. Inside a postal network a failed attempt very often converts into a notice and a hold at a collection point, and from that moment the clock that matters is the storage clock, not the delivery clock. Both are legitimate ways to run a network.

Neither is what most shippers believe is happening. The sentence I repeat to my team until they are sick of it: the window closes and that is it. Once a parcel converts to a hold, you are not recovering the promise, you are racing a storage clock. So the rule is four hours. From the failed-attempt scan, somebody decides: second attempt, divert to a collection point with the recipient notified, or return. Four hours, not tomorrow morning. An exception still undecided at the end of a shift has already lost both clocks.

Underneath all of it sits the same cheap fix I have been pushing for years, which is a phone number and a clean address. A driver who can call ahead converts a failed attempt into a delivery. A recipient who gets a message twelve hours before the window rearranges their day. I will not quote you a conversion rate, because I do not have a clean one and I am not going to invent one. Measure it yourself: four weeks, tag the parcels carrying a reachable mobile on the label against the ones without, compare first-attempt success, set your own target. The expensive version of this problem is the address itself: no house number, a postcode that belongs to the next street, a company name sitting in the recipient field. Retries never fix those. They get fixed at the checkout form, validated against the destination country's format, which is a product change rather than a logistics one.

Exceptions need a clock and a name, and both have to exist before peak rather than during it. I split them into five families: address problem, recipient problem such as refusal or not at home, damage, handover or customs hold, and missed connection. One clock for all five: classify within two hours of the exception scan, decide the disposition within six, close or escalate within twenty-four. One named owner per lane and never a shared mailbox, because a shared mailbox is where parcels go to die, and forty-eight hours later you are issuing refunds instead of making deliveries.

And the threshold I trust most: if one exception code passes two per cent of a lane's daily volume across two consecutive twenty-four-hour periods, it is a defect rather than an exception, and it goes upstream to whoever owns the data. Below that, treat it as floor noise. Above it, you are fixing a form, a product page or a carrier's sort logic, and no amount of chasing individual parcels will fix it.

Peak is where all of this gets tested, and the ranking does not help you there at all. A large network in peak behaves like a large network: it protects its premium products, it protects its largest accounts, and everyone else gets what is left, at whatever hour is left. I book eight weeks ahead, by lane, with a written cutoff and a written overflow route, and I treat booked volume crossing ninety per cent of the agreed allocation for two consecutive weeks as the trigger for the overflow conversation rather than the trigger for panic. If you wait until the week your volume spikes, you are not buying capacity, you are asking for a favour, and favours get granted last. The other half of peak planning is boring and it works: clean the address data before the volume arrives, re-weigh and re-measure the cartons, and get the exception owners named and rehearsed while the depot is still quiet enough to train people in it.

Let me put the arithmetic on the table so you can check it, because everything that follows is either their published numbers or my assumption, and you should be able to tell which is which. Published: China Post 119th, COSCO Shipping 237th, SF Express 372nd, on a list that ranks by revenue, published 28 July 2026. Published: roughly 160 million shares repurchased for about 5.999 billion yuan against a 6 billion yuan ceiling. My arithmetic on those: execution of 5.999 over 6, which is 99.98 per cent, leaving about 1 million yuan unused; and 5.999 billion divided by 160 million, which is roughly 37.5 yuan per share on average. The spacing between the rankings is worth a glance too: 118 places between the first and the second, 135 between the second and the third. Those are revenue gaps, and they tell you the three networks are not the same shape. Which is exactly why a single cutoff assumption across all three will eventually fail you.

Now a handover sum, and every number in it is my assumption, so put your own in. Suppose your shipment crosses two legs, an ocean or line-haul leg and a last-mile leg. Suppose each handover needs six hours of buffer in normal conditions, measured from arrival scan to sort complete. Two handovers is twelve hours. Suppose your customer promise is seventy-two hours. Twelve over seventy-two is about 16.7 per cent, so nearly a fifth of your entire promise is spent standing still at handovers, and that is on the assumption that nothing at all goes wrong. Get each handover down to four hours and the total is eight, which is about 11.1 per cent, and you have bought four hours of promise back without touching the transport. That is the cheapest capacity you will ever find, and it lives in a document, not in a truck.

My own book, all assumptions again. Thirty thousand parcels per 720 hours, call it a month, moving through a mix of postal and express last miles. Failed first attempt rate of 4 per cent, so 1,200 parcels. Assume 10 yuan of handling on each second attempt, and that 10 yuan is my number, not a published tariff, which gives 12,000 yuan per 720 hours. Now suppose the failure rate comes down to 2 per cent: 600 parcels, 6,000 yuan, a saving of 6,000 yuan per 720 hours and roughly 72,000 across a year. The work that produces it is a mandatory mobile number, an address validator and a message twelve hours before the window. Set 72,000 a year against the cost of one more person on the exception desk and the decision is not close.

One more, on peak surcharges. Assume the peak surcharge on the lanes you use lands at 2 yuan per parcel, which is my assumption. On 30,000 parcels that is 60,000 yuan per 720 hours of peak, and peak runs longer than one month in most calendars. Assume that committing volume eight weeks early lets you hold the pre-peak rate on the committed portion, and assume that portion is 70 per cent of your volume, also my assumption. That is 42,000 yuan you did not spend, bought with a planning conversation eight weeks earlier. That is the whole argument for booking early, and it has nothing to do with being pleasant to your carrier.

Alternatives, and what each one costs. Switching the last-mile channel, express to postal or postal to express, changes the failure behaviour far more than it changes the price, and that is the part people miss. Postal gives you reach and a collection-point network. Express gives you tighter waves and a second attempt that actually happens. Test it with 500 parcels on one lane, measure first-attempt success and the exception mix, then decide. Switching the gateway or the port of entry moves your handover window by hours in either direction, and it costs you a learning curve: I allow four to six weeks for a new lane to stabilise, and through those weeks the exception rate sits above steady state.

Do not make either switch inside eight weeks of peak, or you will hit the busiest weeks of the year with a lane that is still learning its own failure modes. Stock position is the quieter alternative. If the last mile is where you bleed, holding two weeks of buffer closer to the recipient turns a seventy-two hour promise into a twenty-four hour one and lifts those orders out of the peak queue entirely. You pay carrying cost and you take mix risk. Against peak exception rates and service credits, that is usually the cheaper of the two.

The traps, in the order I watch them bite. Storage free time: on a postal hold the clock nearly always starts at the arrival or hold scan, not at the moment somebody told you, and whether weekends and public holidays count varies by country and by product. Get it in writing, in hours. Surcharge basis: peak and oversize charges are usually per parcel and on chargeable weight, not per shipment on actual weight. Assume your cartons run at 1.4 times volumetric, my assumption for soft goods, and you are paying forty per cent more than your own scale suggests, discovered on the invoice.

Cutoff clocks: on a shipment that touches two of these networks, vessel closing, pickup cutoff, sort cutoff and line-haul departure are four different clocks, and sales teams quote whichever one sounds best. Promise clocks: delivery attempted is not delivery, and the promise usually stops at the attempt, so a parcel can fail inside a promise that was technically met. Proof of delivery: postal and express produce different evidence, and if your contract or your marketplace needs a signature, confirm which channel actually gives you one before you move the volume.

Where I land on all of this is simple. Three Chinese groups sitting on a revenue ranking is a real statement about how much network stands behind your parcels. It is not a statement about your delivery window, and it never was. What decides your window is the cutoff you hold internally, the hours you lose at handovers, the four-hour decision on every failed attempt, and the slots you booked before everybody else wanted them. When a parcel comes back on a failed scan, we get the parcel back first and argue about who pays later. That is the whole job. Everything else in that announcement is scale, and scale only helps you if the hours underneath it are yours.

↑ Back to the key points

— By Nadia Pryce