The Trade Desk

When you open a news site, a streaming app or a podcast, an advertisement is often chosen for you in the tenth of a second before the page loads. An auction runs: websites and streaming services offer the space, and advertisers bid for it, through software. The Trade Desk makes the software the buyers use, called a demand-side platform. A ten-year-old could follow it: an advertising agency logs in, says "I want to reach young parents in Ohio on connected TV, at no more than this price", and the platform does the bidding, billions of times a day.

The Trade Desk does not own the websites or the viewers, and it does not take the risk of buying the space itself. It takes a fee on what its clients spend, historically about a fifth. In 2025 clients spent about $13.4bn through the platform, and The Trade Desk kept $2.9bn of it as revenue. That is the toll. Its great selling point has always been independence: unlike Google and Amazon, it owns no media of its own, so it has no reason to steer an advertiser's money towards its own properties.

For a decade that was enough to grow more than 20% a year with very high margins. In the last two years three things have gone wrong at once. Amazon built a rival platform and offered it almost free. The big agencies, which control most of the spending, began to demand discounts and to audit the fees. And The Trade Desk's own execution slipped, with two misses of its own forecasts and a parade of executives leaving. Revenue has stopped growing.

The first cell is the remarkable one. Even after losing nine-tenths of its value, a shareholder from ten years ago has beaten the index. That tells you how good this business was, and how much the price had assumed it would stay that good. It was removed from the S&P 500 in September, fourteen months after joining it, and dropped straight to the small-cap index.

The early history is one of the best in technology. Green and Pickles built a profitable company in a business full of money-losers, and they did it by refusing to own media, which meant advertisers could trust them. For years that trust was the moat.

The recent history is a lesson in what happens when a business priced for perfection meets ordinary difficulty. None of the individual blows was fatal: a missed quarter, a stronger competitor, an unhappy agency. Together, and with a management team that turned over repeatedly, they took a company valued at more than $50bn to about $7bn.

The Trade Desk's revenue is its clients' spend multiplied by its take rate. On the company's figures, clients spent about $12bn in 2024 and $13.4bn in 2025, and revenue was $2.45bn and $2.90bn. That makes the take rate about 20.4% in 2024 and 21.6% in 2025. Notice what that means: in 2025, spend grew about 12% but revenue grew 18%. A good part of the last good year came from charging more per dollar.

That lever is now moving the other way. Since late 2025 The Trade Desk has, for the first time, negotiated fee reductions and discounts inside "joint business plans" with the agencies, and the third-quarter 10-Q names those discounts as one of the reasons revenue fell. It also lists lower spend from existing clients and less use of its data products. When a toll operator starts cutting the toll to keep the traffic, the question is no longer how fast the road grows but what the toll will settle at.

The cost side is simple: gross margins of about 75–80%, and the biggest expense is people, much of it paid in stock. Stock-based compensation was about 25% of revenue in 2023 and is down to about 13% in the third quarter. That is still a large cost that the adjusted figures leave out.

The Trade Desk reported its third quarter after the close on 9 November. The shares had risen 15.5% that day, before the figures, with no company news, on heavy volume and very high short interest: about a fifth of the free float was sold short. After hours they rose further, then gave most of it back.

The beat is real, and smaller than it looks. Revenue of $736m beat the company's floor of $650m and the analysts' estimate of about $651m by a wide margin. But three things flattered it. First, $15m came from a strategic partner and will not recur; the CFO described it as about three points of growth, which is more than $15m would account for, a gap we cannot reconcile. Second, political advertising, which comes every other year, was a mid-single-digit share of the business. Third, a change in how some supplier costs are booked moved them out of revenue reductions and into costs. On UBS's estimates of those two effects, revenue without them fell roughly 8–9%.

The guide is a decline. For the fourth quarter The Trade Desk expects revenue of at least $755m and adjusted EBITDA of at least $280m, against $847m and $400m a year ago: down about 11% and 30%. That is well above what analysts had pencilled in, about $690m, and that is why the stock rose after hours. But the CFO also said not to expect the usual seasonal rise from the third quarter to the fourth, and that in election years the first quarter has historically been about a quarter lower than the fourth. The plan for 2027 comes in February, with expense growth to "moderate meaningfully" and more than $100m of savings from the job cuts.

Two accounting details. From April the company extended the assumed life of its servers from three years to four, which lowers depreciation by about $14m this year. And much of the year's strong cash flow came from collecting receivables faster than paying suppliers, which happens naturally when billings shrink. Neither is improper. Both make 2026 look better than the business underneath.

This chart is the whole story in one picture. A business that grew at 25–28% for years slowed every quarter of 2025 and turned negative in 2026. Some of that is cyclical: large consumer-goods and car brands, about a quarter of the business, are cutting budgets. Some of it is structural: discounts to the agencies and spend moving to Amazon. The honest answer is that nobody, including management, yet knows the split.

Buffett's question about any toll is whether the traffic has another road. For most of its life The Trade Desk's answer was good: advertisers wanted one place to buy across the whole open internet, and they did not trust Google or Amazon to be neutral, because both own media they would like you to buy. The Trade Desk owned none. That was a real advantage, and its customer retention, above 95% for more than a decade, shows that clients who come rarely leave.

The weakness is pricing power, and it is the one that matters most. Amazon's platform reportedly charges nothing on guaranteed deals for its own media and about 1% on the open web, against The Trade Desk's roughly 20%. Amazon also secured an exclusive integration with Roku, which reaches about 80 million American TV households, and Netflix inventory from late 2025. Walmart's exclusive arrangement with The Trade Desk ended. When a competitor with the deepest pockets in the world offers a similar product almost free, independence has to be worth a great deal to keep a 20% toll. For some advertisers it is. For others, clearly, it is not.

The second weakness is concentration. Two agency holding companies were about 30% of spend in 2025. Publicis stopped recommending the platform for three months this year after an audit claimed the fee had been applied to other fees; the two settled in June without disclosing terms. A business whose largest customers are also its loudest critics does not have a wide moat.

Our verdict: a narrow moat that is narrowing, kept alive by switching costs and by advertisers' reluctance to give everything to Amazon and Google. It is not gone. Netflix, Disney and most streaming services still sell through The Trade Desk. But the toll is now negotiable, and in a toll business that is the whole question.

Here is the question that decides the value of this company. Advertising is moving to platforms that own both the audience and the buying tools: Amazon with its shoppers and Prime Video, Google with YouTube and search, Meta with its apps. For them the buying software is a means to sell their own media, so they can give it away. The Trade Desk sells only the software, so it must charge for it. That is a fine position while advertisers value neutrality. It is a hard one if they come to see the software as a commodity.

Our reading. The Trade Desk is not going to disappear. Large advertisers will keep at least one independent platform so as not to depend entirely on Amazon and Google, and The Trade Desk is the obvious one. But "one of two or three platforms an advertiser uses" is a very different business from "the default", and it earns a lower toll. We think the likely path is a take rate that drifts down a few points, revenue that returns to modest growth with the open internet, and margins helped by cost cuts. That is a decent business. It is not the one the market was paying $117 a share for at the end of 2024.

Green was asked on the call whether the company should keep charging a percentage of spend at all. The answer was that the take rate has stayed close to its long-run average and that it would be lowered if that helped growth. We appreciate the candour. It also tells you who now has the upper hand in the negotiation.

It is instructive that the supply side of the same market is doing well. Magnite, which sells advertising space for publishers, grew revenue about 14% in the third quarter, and its connected-TV business, before traffic acquisition costs, about 39%. The open internet is not shrinking. The share of it that passes through The Trade Desk, at The Trade Desk's price, is. That is the strongest evidence we have that the problem is competitive and not only cyclical.

The governance, plainly. Buffett has always said that integrity is the first thing to look for in a manager and that its absence makes the rest irrelevant. We are not suggesting dishonesty. Green founded this company, built it brilliantly, and put $148m of personal money into the shares this year at prices well above today's. But three decisions of the last fourteen months show whose interests come first when they conflict.

First, the ten-vote shares were due to convert to ordinary shares in December 2025. Instead, with Green's own votes counted, the date was moved to 2035. On our arithmetic, if Green's votes all went in favour, ordinary shareholders voted against by more than two to one. Second, in October about 14.4m underwater options held by about 2,300 people, including 1.6m of Green's own ordinary options, were repriced to $12.26. Again, if Green's roughly 428m votes went in favour, about 95% of the other votes cast were against. Third, a new 7m-share option for Green at $14.97 was granted in September, as a "bridge" to the 2021 award whose targets are now out of reach. Each decision can be defended on its own terms: staff retention, continuity. Together they say that when the price falls, the people inside are made whole and the people outside are not.

The revolving door. Four people in the CFO role in about fourteen months, the strategy chief, the marketing chief and the head of the TV operating system gone, four directors resigning within weeks this spring, and a Nasdaq notice about committee composition. A business can survive turnover. It rarely improves during it.

Capital allocation. The Trade Desk has no debt and $1.65bn of cash, which is a real strength. But it spent about $2.3bn on buybacks from 2023 to 2026, including about $1.4bn in 2025 at an average of $52.60, and bought nothing at all in the third quarter of 2026, when the shares were near $12. On Monday it raised the authorisation to $1bn. Buffett's rule is simple: buy back shares only when they are below intrinsic value, conservatively worked out. The company did the opposite, buying heavily when the shares were dear and pausing when they were cheap. Meanwhile stock awards kept issuing new shares, so the share count fell only about 4% in two years. Outstanding awards and options equal about 16% of the shares.

Read across the "stock pay" column and then the "net income" column. From 2022 to 2025 stock pay was about $500m a year, more than GAAP net income in every year. That is the cost that the company's adjusted EBITDA and non-GAAP earnings leave out. Free cash flow looks generous because stock pay is non-cash, but it is paid with shares, and the company then spent its cash buying shares back to stop the dilution. An owner should count it.

The good news is in the same table: no debt, a decade of positive free cash flow, and stock pay falling as a share of revenue, from about 25% in 2023 to 13% now. This is a profitable company in a difficult patch, not a fragile one.

Which earnings are real? This is the reverse of the Microsoft or Alphabet problem we have written about, where heavy investment makes free cash flow understate the business. Here, stock pay makes adjusted earnings and free cash flow overstate it. If you count the shares given to employees as the cost they are, The Trade Desk earned about $0.78 a share in the last year, not $1.34. Of today's $14.93, about $3.49 is cash, so the operating business is priced at about 15 times owner earnings.

The model. We start from 2026 revenue of about $2.9bn, grow it to 2031, apply a margin for owner earnings (after stock pay), put a multiple on it, discount five years at 10% and add today's net cash per share.

The central value moves to about $13 at 14 times and $15.70 at 18 times; to about $14.90 at a 9% discount rate and $13.90 at 11%. If revenue simply stays flat, it is about $12.

What is priced in? At $14.93, on our central margin and multiple, revenue needs to grow about 6% a year to 2031. That is a modest recovery for a business that grew 25% a year until recently. In other words, the price is fair for the middle case. It is not cheap, because the range around the middle is very wide: from less than half of today's price to almost double.

A comparison. Booking Holdings, which we analysed last week, also sells a service that a giant could try to give away, in its case AI agents and Google. The difference is in the numbers: Booking's growth has held, its take rate has held, it trades at about 17 times owner earnings, and it has one class of shares, each with one vote. The Trade Desk's growth and take rate have both broken, and its governance has moved the other way. Similar fear, different evidence.

Why not a price at which we would buy? For most companies we give one. Here, at $10, the stock would be a third below our central value, and the numbers alone would argue for a look. But we would still be buying a minority stake in a company where the controlling holder has shown, three times in fourteen months, that the rules can be changed when they become inconvenient. Buffett's advice is to invest only with people you would be happy to have as partners, on terms that treat all partners alike. Until the governance changes, or the price falls to the cash and a little more, we would rather not be a partner. That is not a short call: a stock priced at its middle case can rise as easily as fall.