Coupang

IThe business

Coupang is a Delaware company with a New York listing and a Korean business. It has about 108,000 direct employees, most of them in Korea, and reports two segments.² Product Commerce is the Korean retail business: goods it buys and sells itself, goods it sells for others, advertising, and the WOW membership. It was 85% of revenue in the TTM C (calculated). Developing Offerings is everything else: Taiwan, Eats, Rocket Now in Japan, Coupang Play, Farfetch and fintech.²³⁷ TTM on this page means the four quarters to 30 June 2026. The third quarter reports on 3 November 2026, after my price date, so every figure here stops at June unless I say otherwise.⁵⁶³¹

What it does

Coupang buys goods from suppliers, keeps them in its own fulfillment centers and delivers them with its own fleet. It calls this Rocket Delivery. The overnight version, Dawn Delivery, promises next-morning arrival for orders placed by midnight.²² The IPO prospectus counted more than 100 fulfillment and logistics centers, over 25 million square feet in more than 30 cities, with about 70% of the population within seven miles of one. That was at the end of 2020.¹⁶ The January 2026 petition from two shareholders uses the same 100-plus count and adds a fleet of more than 25,000 vehicles.²²

Few countries suit this as well as Korea. Most people live in a small area, many of them in high-rise blocks, so a van makes many stops per hour. Most retail networks are built on a promise of low cost or high speed. Coupang’s bet was that if it owned every step, it could offer both. The prospectus put the total Korean commerce market at $470B in 2019, of which $128B was online, and projected $206B online by 2024, both from Euromonitor.¹⁶ Bernstein, in a 2025 note reported secondhand, put online at 42% of retail, rising to 46% by 2027.³⁹

WOW is the glue. For 7,890 won a month, about $5.90 at today’s rate, members get free delivery, free Eats delivery and the Coupang Play video service. The fee rose 58% in 2024, and Coupang had about 14 million members in 2023.³⁸⁴⁸ The filings give no count. Management says only that membership is above its level before the breach.⁷

How it makes money

Revenue has two lines. Net retail sales, the goods it owns, were $26.3B in 2025. Net other revenue was $8.2B: commissions on other sellers’ goods, advertising, WOW fees, Eats and Farfetch. That second line was 13% of revenue in 2023 and 24% in 2025 C (calculated). Farfetch explains part of the jump, and fulfillment for outside sellers, called FLC, explains more. Gross margin went from 4.7% in 2018 to 29.4% in 2025 C (calculated), the best single fact about the company.¹⁶²

Product Commerce margin rose by 5.4 points in three years, then fell to 5.1% in the second quarter C (calculated).¹⁴⁶⁷ The TTM figures add 2025 to the first half of 2026 and take away the first half of 2025.⁶³¹⁸

What happened

In November 2025 Coupang learned that a former employee had taken data from customer accounts. The 10-K says about 33 million accounts were involved, and that names, phone numbers, delivery addresses and email addresses were obtained, but no card details or passwords.² Coupang first reported about 4,500 accounts. The Korean regulator later concluded 37.55 million users were exposed, and Coupang says the attacker kept data from about 3,000.²⁴ The numbers are still in dispute. The reaction is not.

Coupang issued about $1.2B of vouchers from January, netted against revenue as they were used, with the last redemptions in mid-April.³⁹ Korea’s privacy regulator announced fines of about $278M for the breach and $132M for an unrelated advertising data matter. Coupang booked the $410M in the second quarter and plans to appeal, but the fines are not stayed on appeal and are not deductible.¹²³ A US staff report counted more than 10 agencies, over 4,000 document requests and over 650 employee interviews.²⁴ Criminal complaints have been filed against some current and former executives and employees.³

Then, in July, a fire damaged a leased fulfillment center in Incheon. Coupang’s owned inventory, its fixed assets there and what it owes sellers for their stored goods come to about $246M, before insurance. Management says customer service was not much affected.³⁸

Where the customers went

The count fell by 0.7M from the fourth quarter to the first and recovered by the second. Revenue per customer fell 2% in dollars and rose 5% at constant currency.²³ Bom Kim says customers who stayed are spending at record levels, that most of the spend that left has returned, and that spend from everyone except those who left and did not return is growing about 16%.⁷ Reported Product Commerce growth was 8% at constant currency. He says most of the gap is the missing group, and that it drops out of the comparison in mid-2027.⁸ I cannot check this from outside. The company does not publish the cohort data. What I can see is that Product Commerce revenue growth at constant currency was 16% for 2025, 5% in the first quarter and 8% in the second R (reported).⁶⁵³

What comes next

Taiwan is the large bet. Coupang says its own network now delivers the vast majority of shipments there next day, seven days a week, that it reached Dawn Delivery in one year against four in Korea, and that early groups of customers follow the Korean curve. It also says selection is a fraction of Korea’s.⁷ It does not disclose Taiwan’s revenue or losses. Developing Offerings is guided to lose $950M to $1B in 2026, with Taiwan the largest part.⁷

Eats is further along. Management says Eats and Rocket Now in Japan together now pay their own way, and that food delivery in Korea has more than quadrupled since Eats launched.⁷ Outside estimates put Eats card payments at 11.4 trillion won in 2025, up 58%.³⁶ Farfetch, the luxury marketplace bought out of distress in January 2024, sits in the segment too. I cover what it cost under Management.

IIThe moat

There are three parts to what protects Coupang, and a fourth thing that does not protect it at all.

Why it is hard to copy

Next-morning delivery is a matter of stock in the right building before the order is placed. That takes forecasts, and forecasts take years of orders. Bom Kim describes the asset as a physical network, data from billions of orders, and direct ties to tens of millions of customers.⁷ A rival can rent vans. It cannot rent the fill rate on a route. Each added order on a route lowers the cost of the next one, and a rival with fewer orders pays more per box. That gap is what gross margin has been measuring.

The membership adds a switching cost that looks small and works well. A household that pays once for delivery, food and video has less reason to shop around. Eats benefits most, and its card payments rose 58% in 2025 on the outside estimate.³⁶ The Fair Trade Commission’s examiners put the sales related to the bundle at 5.26 trillion won.³¹ The commission held its first hearing on the case on 28 and 29 October.⁵⁴

The test

The breach was the first real test of switching costs. Customers had a reason to leave, public anger, and rivals with money to spend. Management’s account is that those who left tried the alternatives for months and came back at their old spend.⁸ WOW membership passed its old level by August.⁷ If that is right, it is the strongest evidence for the franchise I have seen, because it comes from a period when everything went wrong. The active customer count is the outside check, and it agrees: 24.6M, then 23.9M, then 24.7M.²³ The customer count says nothing about what share of each wallet came back. Only revenue will show that, and it has not yet.

Evidence it is narrowing

Naver is the rival that matters. WiseApp’s card-payment estimate for the first half of 2026 has Naver and Naver Pay at 96.5 for every 100 at Coupang, and Baemin at 32.8.³⁷ That is an outside estimate, it counts Naver Pay used anywhere and it is not revenue. Still, Korean shoppers do have a real second choice. Bom Kim concedes more elevated competitive activity, and says the step-up in marketing is a one-time cost of winning customers back, not a response to rivals.⁸ Marketing spend that was forced by competition and marketing spend that was chosen look the same in the accounts. The test comes in 2027, when he says it falls.

Some of the protection also depends on rules Korea can change. Dawn Delivery rests on night work, and the 10-K says that the Korean Confederation of Trade Unions has supported a ban on nighttime deliveries and that Dawn Delivery would then no longer be permitted.² A tentative March deal capped night delivery at 48 hours a week with two days off after five, and Coupang fought for the higher limits.³⁴ The talks then broke down over 46 hours against 50. On 25 August a ruling-party lawmaker introduced bills that would set night work at 10 hours a shift and 48 hours a week, with no more than three nights in a row and 12 a month.⁵⁵ I could not confirm that either bill has cleared a committee. If the monthly limit survives, it would bite on Dawn Delivery harder than the weekly one. Limits are a cost, and one that rivals pay too. A ban would be a different matter. I treat a ban as unlikely and a tightening as likely.

The franchise is strong with customers. Its weak side faces the government, and customers cannot help with that.

IIIThe competitive advantage

Coupang’s advantage comes down to delivering more boxes per hour than anyone in Korea, because it owns the network and has the volume to fill it. Being first and being big helped, but the cost per box is what lasts. The comparison below uses the best figures I could find. The rivals report differently, so I keep to what can be read across.

Where Coupang wins

It wins on frequent, everyday goods, the kind a marketplace of many small sellers finds hard to deliver by morning at a known cost. Coupang also knows what it will sell, so it can buy in bulk, and it says it is missing volume-based supply chain savings this year that it expects to recover next year.⁷ That admission shows how much scale is worth. Lost volume, discounts and extra marketing took Product Commerce margin from 8.4% in 2025 to 5.1% in the second quarter C (calculated).⁶⁷

It also wins in advertising and fulfillment for others, since a seller who wants Rocket’s badge has to pay for it. Management calls these margin-accretive and says they are early in their scale.⁷ Amazon’s retail business is the obvious reference. North America earned an 8.2% operating margin on my Amazon page.⁴⁹ Coupang’s Product Commerce earned 8.4% in 2025 before stock pay and depreciation, which take about 3 points, so the gap to Amazon’s level is about 3 points C (calculated).⁶

Where it loses

It loses on breadth. Selection on Rocket is limited by what it can stock, and Bom Kim says a meaningful share of what customers want is still not available there.⁹ Naver, with many sellers and no warehouses to fill, has more of the long tail and spends less capital. Temu and AliExpress compete on goods that can wait, and the petition argues that Korean regulators treat Chinese companies more gently, a claim I cannot judge.²² Coupang also has more to lose from any new rule, because the rule will be written with the biggest player in mind. The 10-K lists what could reach it, including night work, platform rules, payment terms and the handling of settlement funds.²

Abroad

Taiwan has one feature in common with Korea: density. It has local rivals, a smaller market, and suppliers that have to be won over brand by brand. The filing says Developing Offerings gross margin improved in the second quarter, with Taiwan the main cause, while for the half it fell.³ Management says the path is not linear, and that sometimes it steps back from a category to rebuild it.⁷ It may work. There is no number in the filings that lets an outsider say so, and the size of the loss is itself withheld. I come back to that in Valuation, because it decides how much of the Korean cash reaches an owner.

Agents

Bom Kim says it is not yet clear what form agent-led shopping will take, and that any winner will need more than AI.⁸ I agree with the second part. A shopping agent still needs a box at the door by morning, which favours whoever owns the network. It could still weaken advertising, since an agent does not look at sponsored results. I cannot size that, so I leave it out of the numbers and flag it here.

IVManagement

Bom Kim founded Coupang in 2010 and is chief executive and chairman. He took it from a daily-deals site to a national logistics network, through losses of more than $1B a year as late as 2021, and to a first operating profit in 2023.¹⁶¹⁴ Whatever else is said on this page, few people have built something that large from nothing. Gaurav Anand is chief financial officer. Harold Rogers is general counsel, and has also been interim head of the Korean subsidiary since December 2025.¹⁰

The structure comes first. Each Class B share has 29 votes, and Kim holds all of them. At 31 March 2026 that gave him 74.3% of the votes.¹⁰ He owns about 158M shares, about 8.8% of the company, worth about $2.3B at the close C (calculated).³¹ SoftBank’s funds held 289.5M Class A shares at that date, far more economically, with 4.6% of the votes.¹⁰ So nobody can outvote him, and nobody can remove him. The rest of this section follows from that.

His total pay of $3.2M was 95 times the median employee’s $33,831, a modest ratio by US standards.¹⁰ The incentive is simple and in one way ideal: his money is in the stock, and his pay barely matters. I like that. It also means that no board measure, return target or vesting date asks him to show a return on the next billion spent. He answers to himself, and he has a stated preference for the long term. The same structure that protected Coupang through ten years of losses protects Taiwan now.

The other named officers do get stock. Their performance units tie to business objectives that the company declines to publish, for competitive reasons.¹⁰ Since they are paid in shares that vest with time and these undisclosed targets, there is no way to tell whether the target includes any measure of capital return.

What he has done with capital

The buyback deserves a closer look. In the first half of 2026 Coupang bought 43.7M shares for $850M, about $19.45 each, the largest return in its history C (calculated). Free cash flow in that half was −$59M, and short-term borrowings rose by about $1.0B.³⁴ It borrowed to buy, in a period when it was also facing $410M of fines, unknown legal claims and a law that will cut its supplier float. The price since then is 26% lower C (calculated).¹ A founder who believes his stock is low is entitled to buy, and $907M remains authorized.³ But a company with $3.9B of suppliers’ money behind its cash balance has less room than the headline suggests. I would judge this one on the Q3 report, which will show whether it kept buying at $14 to $16.

Who else is in, and who left

Insiders have sold in small amounts. Harold Rogers sold 192,110 shares at about $16.18 in August, mainly to cover tax, and the chief accounting officer sold 5,519 at about $13.65 in October under a pre-arranged plan.²⁰²¹ The two most recent filings I opened were both sales. The proxy says shares held by Anand, a further officer and Mehta are pledged against personal debt, with committee approval.¹⁰ Pledged shares are a small thing at $30 and a larger one at $14.

Governance

The proxy listed eight nominees, and Kevin Warsh had said he would leave if confirmed as Federal Reserve chair. The June results show seven, all elected.¹⁰¹¹ Say on pay passed with 5.49B votes for and 96M against.¹¹ Kim’s Class B alone carries about 4.58B of those votes C (calculated). Take those out and the rest still voted about 10 to 1 in favour C (calculated). Kim’s brother works at the company, earning about $491,000 in 2025 plus 41,510 RSUs, and a sister-in-law about $300,000.¹⁰ The sums are small. In Korea they carry weight, because the brother’s role is what the Fair Trade Commission used to name Bom Kim the group’s controlling person. Seoul High Court suspended that order in July until 30 days after the main ruling.³⁵

The Korean subsidiary has had an interim head since December 2025, and the parent is a US company whose founder holds the votes.¹⁰ Most of its workers, customers and revenue are in Korea.² That split has become part of the political argument, and I cannot judge what it costs. It is a real exposure that no filing measures.

VFinancials

All figures are in US dollars as reported. Most of the business runs in won, so reported growth can differ widely from real growth. In 2025 revenue grew 14% in dollars and 18% at constant currency.² In the second quarter of 2026 it grew 4% and 10%.³ I give both where they matter. The filings begin with 2018, so the table has eight years. Gross margin rose from 4.7% in 2018 to 16.5% in 2019, as XBRL confirms for the later year.¹⁶¹⁷

Blank cells are figures the documents I opened do not give. 2018 to 2020 are from the prospectus and XBRL, later years from the 10-Ks, and the 2019 and 2020 free cash flow are the prospectus’s rounded figures.^1617151424 The 2020 operating loss differs slightly between the prospectus and later filings, and I use the later one.

Margins

Gross margin is where the story of this company sits. It went from 16.0% in 2021 to 29.4% in 2025 C (calculated). That rise came from scale in the network, automation, and the growth of higher-margin revenue such as advertising and FLC.¹⁵²⁷ In the TTM it is 28.4% C (calculated). Operating, general and administrative cost rose from 23.4% of revenue in 2023 to 28.0% in 2025 C (calculated), as Farfetch and the ventures came in.²

Management’s account of 2026 is mechanical. Capacity, supply contracts and staff were sized for a demand curve that the breach bent, so costs are spread over less volume, and they chose not to cut them.⁸ Some of this is plainly true. It has happened before: coming out of COVID, a similar shift hurt margins and they recovered.⁸ The guide is that Product Commerce returns to about pre-incident margins by mid-2027, and that consolidated margins fall 300 to 400 basis points in the third quarter.⁷ Anand says the pressure is temporary and nothing has structurally changed. My base grants most of that and not all of it.

Cash, and what an owner can take out

Reported earnings say little here. In 2025 net income to holders was $208M. Operating cash flow was $1,773M. Take off capital spending of $1,251M and stock pay of $475M and an owner is left with $47M C (calculated).⁶² That is the true cash for the year, after everything, and it is close to nothing. In 2024 the same sum was $574M, and in the TTM it is −$434M C (calculated).²⁴

That number is unfair in one way, since most of the capital spending is for growth. If I charge only depreciation of $517M as the capital needed to stand still, an owner gets about $781M in 2025, against reported earnings of $208M C (calculated). It is unfair the other way too, since the operating cash flow leans on suppliers paying for growth, a source that is about to shrink.

The figure that best explains the company is the Korean business on its own. In 2025 Product Commerce made $2,485M of Adjusted EBITDA. Charge it 86% of stock pay and depreciation by revenue share, and tax it at the 25% management expects long term, and it earns about $1.2B after tax E (estimate).⁶⁹ The ventures then took about $1B away before any tax relief, because losses in Taiwan and Japan give no shield. That is why tax runs at 75% to 80% this year.⁹ So $1.2B of Korean earnings became $208M. An owner sees the second number. The first is what the business could send home if the founder stopped investing, and he has said he will not.

TTM cash also has not yet carried the fines. Accrued expenses rose $361M in the half, and at June the $410M sat in that line.³⁴ When it is paid, the −$434M becomes worse by that amount. Stock pay also rose to 1.5% of revenue from 1.4%, and Q2 was $164M against $113M a year before.⁴

The balance sheet

At 30 June Coupang held $6.1B of cash against $2.6B of borrowings: $1,985M short-term and $618M long-term, including $750M drawn on the revolver.⁴³ That looks like $3.5B of net cash. I do not read it that way, and this may be the most important table on the page.

Suppliers and sellers fund the business for more than the whole of its net cash. That is normal for a retailer that grows, and it is a real advantage. It holds only as long as the rules let it, and they just changed.

On 17 September the National Assembly passed an amendment to the Large-Scale Retail Business Act. It cuts the limit on paying direct-purchase suppliers from 60 days to 35, and on consignment-type deals from 40 to 20. It takes effect one year after promulgation, with details set in rules still to come.³² An FTC survey of 132 large retailers found an average of 27.8 days for direct purchases. Coupang averaged 52.3.³³ The 10-K warned this may reduce accounts payable and cash.²

I can size it roughly. Direct purchases are about $25.6B a year at today’s won, taking net retail sales less a 15% product margin, and a 17-day cut from 52.3 to 35 releases about $1.2B of cash to suppliers E (estimate). That is a third of net cash, gone once, around the end of 2027. It may be larger, since an average of 35 is not the same as a limit of 35, and the limit may not apply to every contract. Monthly settlement has a separate rule.³² I use $0.8B to $1.8B across my cases.

There is a second, smaller version. From December 2026 at least 60% of settlement funds in electronic payments must be managed by third parties, rising to 100% by December 2028. The 10-K says this may raise restricted cash or require insurance.² The balance is not disclosed, so I cannot size it. The filings also show $475M of supplier finance inside payables, which behaves like debt.³ Other commitments include $3.0B of operating leases and about $1B of expected severance benefits, and I treat those as operating costs.⁴²

Currency

This matters more than most readers will notice. The won averaged about 1,500 in the second quarter, and closed June at 1,549. By late October it was 1,325.⁴⁸ Most of Coupang’s revenue, costs and cash are in won or Taiwan dollars.² The Product Commerce revenue of $7.4B reported for the second quarter would be about $8.4B at today’s rate C (calculated). Reported fourth-quarter growth may look very strong for this reason alone, against a guide given in constant currency, and the two are easy to confuse. The June balance sheet also carries won cash at 1,549. I have not marked that up, because the filings do not say how much of the cash is in won, and much of the debt is in won too. If half of net cash is in won, it adds about $0.3B E (estimate).

VIValuation

What would ruin it

Before any numbers I ask what would most likely destroy this, and work back from there. I see three ways it goes badly wrong for an owner.

1. Korea changes what the business is allowed to do. A ban or tight limit on night delivery would take away the service customers pay for, and the August bills are the nearest version of that.⁵⁵ A KFTC ruling could force WOW and Eats apart, and the fine is the small part of that.²³ An antagonistic state has many tools, and the petition says a revocation of the business licence has been raised.²² The franchise survives most of these. It earns much less after some of them.
2. The margin does not come back. Management guides to Product Commerce margins near their pre-incident level by mid-2027.⁷ That was 8.4% for 2025 and 9.0% in the second quarter of 2025, against 5.1% now.⁶¹⁸ I use about 8%. If the customers who left stay gone, if Naver keeps rising, or if marketing has to stay high, then fixed costs built for a larger Coupang stay too large, and the supplier float shrinks at the same moment. In that case the company has thin margins, a large fixed base and less free cash.
3. The cash never reaches the owner. Nothing needs to go wrong in Korea for this one. Korea earns well, and the founder, with 74.3% of the votes, keeps sending it to Taiwan and whatever comes next for another fifteen years. He may be right. A shareholder has no way to insist.¹⁰

The first two make up my bear case. The third is why I show what the business is worth with and without the ventures.

What the price implies

At $14.37 Coupang is worth $26.1B on 1,820M shares C (calculated): 1,639.7M Class A and 157.8M Class B on the 10-Q cover at 31 July, plus the 22M the 10-Q shows as left out of diluted shares because of the loss.¹³ There were also 59M RSUs and 13M options at year-end.¹⁹² If every RSU vested at no offset the count rises about 2%. The stock is down 57% from its 2025 closing high of $33.53 on 17 September, 39% this year and 59% from the IPO C (calculated).¹ The lowest close of the year was $13.75, on 26 October.¹

Earnings multiples are not much use in a loss year. Nasdaq’s mean is −$0.47 for 2026, on two estimates, then $0.34 for 2027 and $0.72 for 2028, on four and three.⁴² Here are the peers on those numbers.

Coupang’s 2028 multiple sits with Amazon’s and MercadoLibre’s, on earnings that do not exist yet.⁴⁴ So the market is not simply dismissing it. A better way to read the price is against the Korean business. Take the $26.1B, subtract net cash of $3.1B after the fines, and $23.0B remains. That is about 19 times the $1.2B the Korean core would earn in a normal year E (estimate). That uses headline net cash. If the supplier float is treated as owed, net cash less the fines is about −$0.8B, and the multiple is about 22 times E (estimate). Either way the buyer gets everything else for free, including a business that loses $1B a year.

A normal reverse DCF starts from today’s cash and grows it. That breaks here, because today’s cash is negative. So I hold my base revenue path, $95.9B in 2036 at today’s won, and ask what free cash flow after stock pay the price needs. If the margin climbs in a straight line from zero in 2026, the price needs 3.5% by 2036 C (calculated).¹ That is $3.3B a year on that revenue.

Two other ways to say it. Inside my model, the price matches if Product Commerce margin peaks at 9.7% and not 11%, or if Product Commerce grows 2.8 points a year slower than I assume, about 4.8% a year on average for ten years E (estimate).

The price assumes a real recovery and some permanent damage, about 1.3 points of margin.

The discount rate

I use 10.5%, the rate on my Amazon page, which is the closest business on this site.⁴⁹ The capital asset pricing model gives about 9.9%, on a 4.94% Treasury, Damodaran’s 4.20% premium and a five-year beta of 1.17 against the S&P 500 C (calculated).^4546471 The three-year beta is 0.45, which would give 6.8%, and I ignore it. It reflects a stock falling on Korean news while US markets rose, and that is just the sort of risk a beta misses. I add about half a point over the CAPM for one thing: nearly all the cash comes from one country, under one regulator in open dispute, with no vote for outside owners.

The base equals the price at a 12.0% rate E (estimate). That is the plainest way to say what the market believes. It wants about 7 points over Treasuries for this cash flow, roughly 2 points more than the CAPM on its own beta.

What I cannot judge

Five inputs drive the value and I cannot know any of them today. I show them as ranges for that reason.

• Korean politics. The president, the prime minister and agencies have spoken and acted against the company, and US lawmakers have answered.²²²⁴ I cannot say how this ends. I have no edge here, and I doubt anyone does.
• Taiwan. The company does not give its revenue or loss. Developing Offerings at zero loss and no growth is worth $19.2 a share in my model; ventures that lose $800M a year forever make it $11.4 E (estimate). That $8 range sits inside my base.
• The KFTC case on WOW and Eats. The 10-Q put the hearing in September.³ Korean press says it began on 28 October and ran two days, and I found no decision by my price date.⁵⁴ The examiners’ related sales are 5.26 trillion won. The most a 4% rate allows on that is 210.4B won, about $158M at 1,330.³¹ The fine range is small. A forced split of the benefit would be the costly outcome.
• The size of the float unwind and the settlement-fund rule, which depend on rules still to be written.³²²
• The won. Each 100 won is worth about $1 a share E (estimate).

Three cases

Every case runs in won at a fixed 1,330 per dollar. Product Commerce starts from 2025 revenue at the 2025 average rate plus about 7% constant-currency growth in 2026. I model Product Commerce and Developing Offerings separately, take off stock pay at 1.4% of revenue, depreciation at 1.6%, and tax at 25% on profit with no credit for losses abroad. Capital spending is depreciation plus 10 to 12 cents of property per added dollar of revenue, in line with property at about 11% of sales today C (calculated). Working capital brings in 3 to 6 cents per added dollar, before the one-time unwind. After 2036 I grow at 3.0%. Values are at 1 January 2027, with cash flows at year end.⁴

The base is a recovery, with a mark left on it. Product Commerce grows 12% in 2027, less than the 16% underlying figure management gives, and fades to 5%. Margin reaches 7.5% in 2027, a little short of the guide, and 11% by 2034. After stock pay and depreciation that leaves about 8%, close to Amazon’s North America.⁷⁴⁹ Taiwan and the rest cross break-even in 2033. In the near term I am below the Street. My base earns about $0.08 of after-tax operating profit a share in 2027 and $0.35 in 2028, before interest income, against the means of $0.34 and $0.72 E (estimate).⁴² The difference is mostly tax and a slower margin.

The bear is a Coupang that remains the largest retailer in Korea and earns too little. Limits on night work and a forced change to WOW hold Product Commerce margin at 7%, and growth slows as Naver takes the next dollar of online spend. The ventures do not pay until 2036. The fine, the full tax claim, a further $0.45B of fines and litigation and a $1.8B unwind all land. Revenue still doubles from today. Free cash flow stays below $1B a year, and an owner at $14.37 loses 4.5% a year E (estimate).

The bull is the story management tells. Lapping the breach lifts growth to 14%, advertising and automation take margin to 13.5%, and Taiwan follows Korea with Developing Offerings at $33B of revenue and 8% margins. It is worth $33, more than twice the price, and it needs nearly everything to go right.

The base has $8.5B of present value in the ten years and $21.3B after them, plus $2.8B of net cash E (estimate). Weighted 25%, 50% and 25%, the cases give $18 a share E (estimate). At 20%, 50% and 30% the figure is $19.6, and at 30%, 50% and 20% it is $16.6 E (estimate). Against the price the skew is −76% to my bear and +131% to my bull E (estimate).

The same dollar, elsewhere

A ten-year Treasury pays 4.94%.⁴⁵ My base on Coupang returns about 13.4% a year from here and my bear loses 4.5% a year E (estimate). A return that high with a loss that large is what a contested situation looks like.

The closest business on this site is Amazon, which has the same model with a better second business. It sits 47% above my base at the same 10.5% rate.⁴⁹ So the dollar here buys a lower-quality franchise at a much better price. The fairer competitors are the other cases on this site where the price is below my base. Uber is 24% below, Tencent is 24% below and neither has a dispute of this kind with its home government today, though Tencent has its own history.⁵⁰⁵¹ Uber has also offered for Delivery Hero, which owns Baemin, Coupang’s main rival in food. Uber’s owner cash today is about $4.7B.⁵⁰ Coupang’s is negative. A careful reader would ask what Korean politics adds for a similar discount. The only answer is the size of the bull, more than double the price, and the cost of that answer is the bear.

How good is the business, and how good is the price

These are separate questions. The business is good and not great. Customers love it and the network is real, but it earns thin margins, it depends on one country and on rules it does not write, and most of what it earns goes into a founder’s next project. On Mastercard or Visa nobody has to ask whether the margin will come back. The price looks better than the business: the market values the Korean core at about 19 times normal earnings and adds nothing for anything else. The business earns thinner margins than the best retailers, and the price pays for part of that. The range of outcomes is wide, from $3 to $33.

What closes the gap

Two lines in that table matter most. One point of Korean margin is worth $2.75 a share, more than any other single input E (estimate). And the ventures, in my base, cost about $1.20 a share against simply stopping them. That says a lot. All of Taiwan’s promise, taken at my estimates, is worth less than nothing today, because the money comes first and the return comes late. If Taiwan never pays, the value falls below the price. The market may be pricing just that, and given how little it is told, the market may be right.

Why the market may be right

The price is below my base, so here is the case against me. My model treats 2026 as a dip. It might be a reset. The company has told outsiders that margins recover, but it has also told them nothing about Taiwan, and it spent $850M at about $19 while its own float was about to shrink. Long-term holders such as Baillie Gifford have sold. Regulators have many cases open, and each one can return. Base-case numbers built on 11% margins have never been reached: the best year was 8.4%. If a company has one proven year near 8% and a state hostile to it, a 12% discount rate is not irrational.¹⁰³⁰³ Part of my gap also comes from the late-October won. At the 2026 average rate, about 1,451, my base is $16.6 E (estimate).⁴⁸

How the odds shrink

The bull needs many things at once, and these are my judgments of each.

That comes to about 6% E (estimate). The six are not independent, since peace with Seoul would help several, so the true figure is higher. The price needs far less. It needs the first two lines, 53% together, and a margin peak of at least 9.7%, which I put at 75%. Together that is about 39% E (estimate). A simple argument here is easier to make than a long one. If Korean margin comes back to about 8% and the state lets the business run, today’s price leaves room. The rest is optional.

The margin of safety

• Price$14.37Close, Q4 2026 [1]
• Base value$18My estimate E (estimate)
• Margin of safety below$13the base value less 30% C (calculated)

I set it at the base value less 30%, more than the 25% on Mastercard, because the cash comes from one country in open dispute with the company and outside owners have no vote. Unrounded, it is $12.56. At that price my base returns about 15.3% a year and my bull 24.6% E (estimate). My bear still loses 3.2% a year, so even at $13 the owner is not paid if the bear arrives. The price is 10% above $13. The lowest close of the past year, $13.75, is close to it.¹ The market may be right that $14 is fair for what is still unknown, and I may be wrong about the recovery. The first evidence arrives on 3 November.

Where this could be in 5 and 10 years

Each figure is my model’s value at that date, in US dollars: the present value then of the owner cash after it, at 10.5% with 3.0% growth after 2036, divided by today’s 1,820M shares. Owner cash is free cash flow after stock pay. The $2.8B of net cash today is counted at the start, as in the yearly returns above, and not added again.C (calculated)

All per share. Owner cash is the sum over ten years, not discounted, and the bear’s is slightly negative because it needs cash early. Return a year is the year-10 value plus that cash against the $14.37 price.C (calculated)

For the base to stand at five years, revenue must reach $68B in 2031 from about $40B this year, and Product Commerce margin must reach 10%. At ten years revenue is $96B, that margin is 11% and Developing Offerings earns $0.95B. Only $1.4 a share of cash arrives in the first five years, and the whole year-10 value is cash after 2036.E (estimate)

The bear keeps margin at 7% and slows growth to 3.5%, with revenue of $57B in 2031 and $70B in 2036. The bull reaches 11.5% in 2031 and 13.5% by 2036, with revenue of $79B and $119B. The rate matters too: the base year-10 value is $39.7 at 9% and $26.5 at 12%.E (estimate)

The most likely way the base is wrong is margin. The best year so far is 8.4% and the second quarter was 5.1%. If it is one point lower every year, the year-5 value is $22.3 and the year-10 value $27.8.C (calculated)

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

The won has moved about 12% in Coupang’s favour since the June quarter.⁴⁸ Reported revenue and cash were struck near the weakest rate in over 15 years, while the stock trades today at the strongest rate of the year. At 1,330, TTM revenue is about $39B and not $35.5B, and the price is about 0.6 times that after net cash C (calculated).

Eats has stopped drawing on the portfolio. Management says Eats and Rocket Now together are sustainable, so the Developing Offerings loss is now largely Taiwan.⁷ That loss fell by $110M from the first quarter to the second, though the first quarter also carried its share of the vouchers.⁷⁹

The customers tested the alternatives. They had months and a reason. The count recovered to 24.7M within two quarters, a better natural experiment than any survey.³

The bull case

Bom Kim gave it in August. The underlying customer base is growing about 16%, the missing group leaves the comparison in mid-2027, margins recover because the costs are volume-driven, and Coupang’s share of Korean retail spend is still below what peers reach in markets such as the US.⁷⁸ These are company claims. Anand adds that nothing has structurally changed.⁸

Nasdaq’s mean target is $22.93, in a range of $12 to $29.⁴³ One report has BofA repeating a $24 target on 12 October, without giving its reasoning.⁴⁰ Bernstein’s 2025 view, before the breach, was that Coupang wins as online rises to 46% of Korean retail.³⁹

Greenoaks and Altimeter argued in January that Korea is mounting a "whole-of-government assault" on a US champion, in a petition that puts their combined stakes above $1.5B.²² They withdrew it in March, keeping the arbitration route, and three more investors joined.²³ This is a complaint about treatment. It is a bull case only by implication. Altimeter’s 13F listed 15.7M shares at 31 December and none at 31 March.²⁵²⁶ Greenoaks listed 25.7M at both March and June, Tiger added to 36.5M, and Duquesne held 2.7M. These are positions, and I found no published view from them.²⁷²⁸²⁹ I found no statement on Coupang from the other investors I looked for.

Against reality

If Coupang returns Product Commerce to about 8% by mid-2027, grows it 12% or more, and the state settles down, a value of about $23 is possible. In my model that takes either a 9% discount rate on my base, or margins about 2 points above my base E (estimate). Today’s numbers are a way off: 5.1% against 8%, 8% growth against 12%, and $219M of quarterly venture losses still running.⁷⁴ The third-quarter guide is for margins down 300 to 400 basis points again.⁷ None of this is disproved, and none of it is shown yet.

What a 10x would take. Ten times the price is $143.70, a value of $261B C (calculated). At 25 times earnings that needs $10.5B of net income. At a 6% net margin, about what Amazon’s North America earns after tax, it needs $174B of revenue C (calculated). By 2036 that is 26% a year in the share price and about 17% a year in revenue, from $35.5B C (calculated). My bull has $119B in 2036 E (estimate). This is not close to impossible in the way it is for a trillion-dollar company. A $26B company has room. Eli Lilly’s shares rose 19 times in ten years from a far larger base.⁴¹ The limit here is the market. The prospectus projected all Korean commerce, including restaurants and travel, at $534B for 2024, so $174B would need about a third of it if it all came from Korea.¹⁶ It cannot. A 10x needs a second and third home, with Taiwan and Japan as large as Korea at Korean margins. I put the odds at a few percent E (estimate). It is also 4.1 times the IPO price, and that took five years to lose.

Where I come down

The returning customers moved me. I give the base more growth than I would have in January, and the won adds about $2 against the rate in June E (estimate). The 16% figure did not move me, because I cannot see it. The law and the buyback pulled the other way. I would raise my values if the margin passes 8% in the first half of 2027 with marketing falling, or if Taiwan’s numbers are disclosed and look like Korea’s. One point of margin would make the base about $21 E (estimate).

VIIIRisks, and what would change my view

1. Korean regulation. Fines, the bundling case, night work limits, platform rules. The 10-K notes executives can face criminal liability under Korean law.²³
2. Customer trust. A second breach would be far worse than the first. The 10-K says the company may again experience data incidents.²
3. Payment terms. The new limit shrinks the float by an amount I estimate at about $1.2B E (estimate).³²³³
4. Settlement funds. At least 60% must be managed by third parties from December 2026 and all by December 2028.²
5. Legal claims. A US securities class action, a privacy class action and derivative suits, with losses the company cannot estimate. A $208M tax claim is in dispute.³
6. Control. Kim’s 74.3% of votes means capital allocation is his alone.¹⁰
7. Taiwan and Japan. Losses give no tax shield, and the filings do not show the economics.⁹
8. Currency. Most revenue and cash are in won, and the buyback and the revolver are in dollars.²³
9. Fire. The company has now had two large fulfillment center fires, in 2021 and 2026.¹³³
10. Korea itself. Tension with North Korea and an ageing population sit behind every number. The 10-K lists the first.²
11. Trade politics. The dispute has reached USTR and Congress. Support from Washington could help the company or make it a target.²⁴²³

What would raise my base value: Product Commerce margin back to 8% with marketing down, and any honest disclosure on Taiwan. What would lower it: a forced end to the WOW bundle, a limit on Dawn Delivery that bites, or a second year of borrowing to fund buybacks. A lower price would not change my base. It would change the answer to the third way of getting hurt, because paying less is the one protection an outside owner has.