United Parcel Service

UPS picks up packages from 1.6 million businesses every working day, carries them through a network of sorting hubs, trucks and aircraft, and delivers them to more than ten million addresses — in 2025 an average of 20.8 million packages a day, in more than 200 countries. It is, in effect, a giant machine for turning fixed costs (buildings, aircraft, sorting equipment and about 460,000 people) into a stream of deliveries, priced per package. A ten-year-old's version: you pay UPS to take your box anywhere, by tomorrow if you like, and it is cheaper for UPS the more boxes it can drop on the same street.

That last sentence is the whole economics of the business. A brown truck that stops at a hundred houses on one road costs only a little more to run than one that stops at sixty, so the profit lives in density — packages per stop, stops per mile — and in mix: a next-day business parcel pays several times what a two-pound bag of dog food sent to a house does. For most of the last decade UPS chased volume, especially e-commerce volume from its largest customer, Amazon. In January 2025 it reversed course: it agreed to cut Amazon's volume by more than half, and set out to build a smaller network that earns more on every package.

The shares, at $94.57 on 9 October 2026, sit 21% below February's closing high and pay a 6.9% dividend — one that has not been cut since UPS listed in 1999, but has not been raised this year either, and now absorbs essentially all of the company's free cash flow. This report asks two questions: is the smaller network a better business, and can it carry the payout it inherited?

Two lessons for an owner. The first is that UPS's history is a history of density: every great step — common-carrier rights, nationwide coverage, the air network — added stops to routes it already ran, and every recent problem has come from volume that filled the trucks without paying for the stops. The second is that the last five years flatter nobody. Measured from the end of 2020 to the end of 2025, a shareholder lost a quarter of their money including dividends while the S&P 500 nearly doubled; over ten years, with the price lower than in October 2016, the return is about 2.6% a year, almost all of it dividends, against more than 15% a year for the S&P 500. The decisions that set the current dividend and the 2022 buybacks were taken at the peak of a boom that has not come back.

The unit economics, in one line. In the second quarter of 2026 UPS's US network earned $14.24 of revenue per package and spent $13.09 per package on an adjusted basis — a spread of $1.15, multiplied by 16 million packages a day and about 63 working days, which is roughly the $1.19bn of adjusted operating profit the segment reported. Every strategic question about UPS can be put in those terms: does a decision raise the revenue per piece, lower the cost per piece, or add pieces to routes that are already being driven?

Why the costs are so stubborn. Most of UPS's cost is people: about 385,000 hourly workers, nearly 80% of the US workforce represented by unions, with the Teamsters' national master agreement running to 31 July 2028. Pay rises under that contract are fixed regardless of volume, so when volume falls, cost per piece rises unless buildings close and hours are taken out. That is why the Amazon decision came with a second, harder programme — the closure of 93 buildings' daily operations in the first nine months of 2025 and 45 more in the first half of 2026, roughly 48,000 jobs cut in 2025, and two rounds of buyouts for full-time drivers — $150,000 each, capped at 7,500 drivers under an April 2026 settlement with the Teamsters — the second costing about $1.1bn.

What pays best. Business-to-business deliveries (dense, predictable, signed for), next-day air, international express and temperature-controlled healthcare shipments pay best. Residential ground parcels pay least, and the cheapest of them — UPS's old SurePost, now Ground Saver — were brought in-house in January 2025 and, within a year, handed back to the US Postal Service for final delivery. The flip-flop cost money in 2026 but shows the direction: UPS wants the parcels where its network is an advantage and is willing to give away the rest.

A parcel network is a classic Buffett moat in one respect: it is enormously expensive to replicate, and the incumbent's cost per stop falls as its density rises. For fifty years only two private networks in America could deliver anything anywhere overnight, and UPS was the more profitable of them. Buffett's 2007 letter puts the test simply — "an enduring 'moat' that protects excellent returns on invested capital" — and on returns UPS has always passed: return on invested capital in the high teens to 20s through most of the 2010s, falling to about 9% now.

What has changed is that a third network now exists, and it was built by the customer. Amazon delivers more parcels in America than UPS does, and in May 2026 it launched Amazon Supply Chain Services, offering freight, warehousing and parcel delivery to any business. That does not destroy UPS's moat — Amazon's network is built for residential delivery of goods it stores, not for business-to-business freight, overnight air or regulated medicines — but it caps the price of the commodity half of the market. We score the moat 6: wide where UPS is choosing to concentrate, narrowing where it is choosing to leave.

Why UPS fired its biggest customer. In 2024 Amazon was UPS's largest customer, at roughly an eighth of revenue; in 2025, 10.6%. On the day the cut was announced, Carol Tomé put it bluntly — Amazon was UPS's largest customer but "not our most profitable customer." Its parcels are light, residential and sent at deeply discounted rates, and Amazon was steadily taking the best of them into its own network, leaving UPS with the overflow. Cutting that volume by more than 50% removed about a million packages a day by the end of 2026 — volume that, UPS argued, filled trucks without paying for the stops.

The evidence so far. In the second quarter of 2026, with the glide-down complete, US volume was down only 3.3% on a year earlier, but revenue rose 6.0% because revenue per piece rose 9.3%. Adjusted cost per piece rose 8.0% — the separation payments, lost density and higher fees to the Postal Service for Ground Saver — so the spread per piece widened from $0.91 to $1.15, and the segment's adjusted margin rose from 6.6% to 8.0%. In July UPS raised its 2026 guidance to about $91.2bn of revenue, $8.65bn of adjusted operating profit and adjusted earnings of about $7.22 a share. That is the first evidence that the strategy works.

Our reading. The arithmetic of the strategy is sound: a parcel network that carries fewer, better-paying packages through fewer buildings with fewer hours can earn more per package, and the second quarter showed it doing so. What it has not yet shown is more profit in total. Adjusted operating profit in 2026 will be roughly flat on 2025 and still well below 2022–23, International is moving the wrong way, and the savings are being partly spent on separation payments that will not recur but have to be paid in cash this year. And there is a new question that did not exist when the plan was drawn up: on 4 May 2026 Amazon began selling the very network that took UPS's volume to every other shipper in America. UPS's bet is that the business it is keeping — B2B, small-business, air, international and healthcare — is the part Amazon cannot easily serve. That is plausible. It is not yet proven, and the third-quarter results (expected in late October) are the next data point.

The pattern. For most of UPS's life its competition was FedEx, and the two behaved like a sensible duopoly: annual price increases of about 5–6%, announced within days of each other. The market now has four national networks — UPS, FedEx, the Postal Service and Amazon — and two of them are not run for profit in the ordinary way: the Postal Service has a universal-service mandate, and Amazon's network exists to serve its retail business, with outside customers as a way to fill it. That is a worse structure for pricing than the old duopoly. UPS's answer is to compete where the other three are weakest: in time-definite air, in business-to-business routes, internationally and in regulated healthcare. The comparison on our board is Amazon itself — the company whose logistics spending UPS once counted as revenue and now counts as competition.

Integrity: no veto. UPS has been unusually plain-spoken about its strategy, and Tomé's description of Amazon as large but unprofitable was the kind of candour Buffett likes. Capital allocation: poor at the peak, good in the trough. The 2022 dividend rise and the 2022–23 buybacks were sized to earnings that turned out to be a pandemic high, and they are the reason the dividend is now uncomfortable. The 2025–26 decisions — cutting Amazon, closing buildings, stopping buybacks — are the right ones, taken late. Credibility: dented. The targets set at the March 2024 investor day for 2026 ($108–114bn of revenue and an adjusted operating margin above 13%) will be missed by a wide margin, and guidance was withdrawn altogether for most of 2025. We score management and capital 5.

Three kinds of earnings, and why they differ. UPS's adjusted earnings ($7.16 a share in 2025, about $7.22 guided for 2026) exclude the costs of the transformation — building closures, buyouts, write-offs of the MD-11 fleet. Its GAAP earnings include them: $5.38 over the last twelve months, depressed by about $1.3bn of after-tax charges in the first half of 2026 alone. The adjustment is fair for genuinely one-off costs, but UPS has reported "transformation" charges in most quarters since 2022, so a sceptical owner should count part of them as the ordinary cost of running a network that keeps having to be reshaped. Free cash flow, which is what pays the dividend, sits in between: about $5.5bn, or $6.45 a share, over the last twelve months.

The capex question. UPS expects to spend only about $3.0bn on capital projects in 2026, 3.3% of revenue, against $5–6bn a year in 2019–2023, when it was building automated hubs. Lower spending flatters free cash flow today. Some of it is genuinely lower need — fewer buildings, less volume — but a network that spends less than its depreciation (about $3.6bn) for long enough eventually has to catch up. In our valuation we assume capex drifts back to ~$3.5bn.

The verdict on the dividend: not safe in the way a 6.9% yield should be, and not about to be cut either. The payout was set in 2022 at half of the pandemic-peak earnings; those earnings fell by 40% and the dividend did not move, so today it takes about 91% of adjusted earnings and all of the free cash flow. UPS's own policy — about 50% of the prior year's adjusted earnings — would justify roughly $3.60 a share on 2025's $7.16. The board's choice is to freeze rather than cut and wait for earnings to grow back into it: at 50%, that needs adjusted EPS of about $13, a level analysts' consensus does not reach until 2030 or later.

So the realistic paths are three. If the restructuring works and free cash flow climbs to $7–8bn by 2028, the dividend becomes covered ~1.3–1.5× and stays frozen for years — a 6.9% bond-like income with no growth. If profits stall, the board keeps borrowing modestly to hold it, which is tolerable for a few years at single-A credit and not beyond. If a recession arrives before the recovery, a cut to something near the policy level is possible; the board has never done it, and it would be read as a verdict on Tomé's strategy. We put the chance of a cut in the next two years at low but real — perhaps one in five — and the chance of a meaningful raise before 2028 at close to nil. The comparison on our board is Verizon, whose ~1.7× cover makes a similar yield far better protected.

What is it worth? We discounted owner earnings of about $7.00 a share at 10% for ten years. If the restructuring delivers what management promises — earnings growing ~5–6% a year for five years as margins recover toward 11%, then 3%, ending at 12–13 times — UPS is worth about $105. If Amazon and the Postal Service cap prices and the margin stalls at ~9% — 1% growth, 10 times at the end — about $67, close to where a dividend cut would probably send the shares. If the network re-levers fast and analysts' 2029–30 numbers prove right — 9% then 5%, 15 times — about $150.

At $94.57 the price sits about 10% below our central value. For a business we score 5–6 on quality and whose dividend is not covered, our rules ask for a wider margin of safety than that: buy below about $82, roughly the low of October 2025 and 22% under our central value, where the 8% yield would compensate for the risk of a cut. Existing holders can keep collecting a frozen dividend while the restructuring plays out — the analysts' mean target of $118.67 assumes it works — but the case for buying more here rests on management's targets, which have not been reliable since 2024.