Business

The Business: A Toll Booth on Open-Internet Ad Spend

The Trade Desk is one of the highest-quality business models in software — a capital-light, buy-side-only operating system that skims a rising percentage fee off a $13.4 billion river of advertising spend it never owns, at margins no independent rival comes close to matching [1]. It is also, as of mid-2026, priced like a broken one: the stock has fallen roughly 78% from its 2025 high, and the market now pays about 3.3 times sales for a business still growing 18% with a 27% free-cash-flow margin. The entire investment question sits in that gap between quality and price — and it turns on one number the bulls and bears both watch: the take rate.

Sources: gross spend $13,394.7M, revenue $2,896.3M, Adjusted EBITDA per FY2025 10-K MD&A Executive Summary [2]; operating income $589.3M per Consolidated Statements of Operations [3]; free cash flow derived from operating cash flow $992.7M less capex $197.0M per Consolidated Statements of Cash Flows [4]; FCF yield computed on a $9.64B market cap at the $19.53 close (10-Jul-2026).

1. The economic engine — a percentage fee on spend it never owns

The single most important fact about The Trade Desk is that reported revenue is a thin slice of a much larger flow. Clients route their advertising budget through the platform to buy inventory, data and value-added services; on top of that spend the company adds a platform fee, generally a percentage of the client's total spend [5]. Because the company acts as an agent rather than a principal, it recognizes revenue net — it keeps the fee, not the media dollars. That is why $13.4 billion of gross spend became $2.9 billion of revenue in FY2025 [6].

The take rate — revenue divided by gross spend — is therefore the master variable. It rose to 21.6% in FY2025 from 20.3% the year before, driven by higher utilization of value-added services and data and higher platform fees [7]. A rising take rate is the crux of the bull case: it means the company is monetizing each dollar of spend more richly. A falling take rate is management's own stated warning sign — the filing explicitly cautions that revenue "may not necessarily grow at the same rate as spend on our platform" as pricing competition, volume discounts and channel-mix shifts bite [8].

Source: gross spend from FY2025 10-K MD&A and prior-year segment KPIs; revenue per Consolidated Statements of Operations [9] [10].

Source: derived from reported gross spend and revenue; the FY2025 take rate of 21.6% (up from 20.3% in FY2024) reflects higher platform fees and greater utilization of value-added services and data, per FY2025 10-K MD&A [11].

The take rate held in a tight 19–20% band for years and then stepped up to 21.6% in FY2025 — the visible signature of the company selling more high-margin data and measurement on top of media, not just routing more dollars. It is the number to track every quarter.

2. Why the revenue is recurring — and where the moat starts

Two structural features make this a durable toll, not a spot fee. First, the company contracts through ongoing master services agreements (MSAs) rather than one-off insertion orders, and reports a customer retention rate that has exceeded 95% for over a decade [12]. Retention above 95% for ten-plus years is rare in any software category and is the strongest single piece of moat evidence in the filing.

Second, the platform gets stickier the more a client uses it: buyers integrate their own first-party data into the platform to optimize campaigns, and the bidding engine lets them tune hundreds of granular bid factors against specific business objectives [13]. A team that has wired its data, its measurement and its optimization logic into The Trade Desk does not casually re-platform. The self-service model reinforces this: clients are trained to run campaigns independently, which deepens habit and lowers the company's own cost to serve [14].

The counterweight — and it is material — is client concentration at the holding-company level. Budgets are decided at the agency/brand level, so the client base looks fragmented; but if aggregated to the parent, two holding companies each represented more than 10% of gross billings in 2025 (up from one in 2024) [15]. The moat is real, but a couple of agency holding companies hold genuine negotiating leverage over the take rate.

3. The balance sheet is a mirage — read it as agency plumbing

A newcomer opening The Trade Desk's balance sheet sees something alarming: $3.77 billion of receivables against $3.01 billion of payables on a company with under $3 billion of revenue. This is not leverage or aggressive accounting — it is the agency float inherent to the take-rate model. The company invoices agencies for the gross media spend (media dollars plus its fee), then pays inventory and data suppliers. Because most spend flows through agencies, it carries "sequential liability": it often must pay suppliers on shorter cycles than it collects from agencies, and if an advertiser fails to pay its agency, the company must chase the advertiser directly [16].

Source: FY2025 10-K Consolidated Statements of Cash Flows and balance-sheet lines; receivables grew $432.7M in the year and payables $291.1M [17].

Two investor takeaways follow. (1) There is no financial-leverage risk — the company carries no funded debt, and cash plus short-term investments of roughly $1.3 billion exceed every liability that is not agency float. (2) Working capital is a permanent, growing use of cash: receivables climb with gross billings faster than payables, so a slice of reported profit is continuously reinvested into the float. That is why operating cash flow, though strong, does not simply track net income up in a straight line — the model funds its own growth in the receivable book [18].

4. Unit economics: real operating leverage, with an asterisk

A DSP is fundamentally software — once built, each incremental dollar of routed spend costs almost nothing to serve. That shows up as expanding operating margins as revenue scales. Margins are also noisy year to year because the FY2021 CEO stock-option grant runs through general and administrative expense, but the direction is unmistakable: from a ~7% operating margin in the 2022 trough to 20.3% in FY2025 [19].

Source: derived from FY2025 10-K Consolidated Statements of Operations (three-year detail) and prior filings; FY2025 operating income $589.3M on revenue $2,896.3M [20].

The asterisk is stock-based compensation. The company adds back $490.6 million of stock-based compensation to reach its headline Adjusted EBITDA of $1,196.4 million [21]. That add-back is roughly 17% of revenue and is a real economic cost — it dilutes shareholders whether or not it touches cash. The honest way to read profitability is to walk from the promotional number down to owner cash earnings.

Source: Adjusted EBITDA $1,196.4M and SBC $490.6M per FY2025 10-K Non-GAAP reconciliation [22]; reported free cash flow $795.7M (operating cash flow $992.7M less capex $197.0M) per Statements of Cash Flows [23]. Bridge steps are illustrative allocations of the same reported items; the endpoints — Adjusted EBITDA, FCF and SBC — are as reported.

The point is not that the numbers are fake — reported free cash flow of $795.7 million (a 27% margin) is genuine cash [24]. The point is that free cash flow adds back the $490.6 million of SBC, so a conservative "owner earnings" figure that treats stock comp as the real cost it is lands closer to $300–350 million. An investor who values TTD on unadjusted FCF and one who values it on SBC-adjusted earnings are underwriting two very different multiples — a distinction the valuation section returns to.

5. Capital allocation: the buyback treadmill

The company pays no dividend and reinvests almost nothing in fixed assets (capex was 6.8% of revenue even after a FY2025 step-up to build capacity), so the entire capital-allocation story is share repurchases [25]. In FY2025 it bought back and retired 26.2 million shares for $1.4 billion — and management is explicit that the program is "designed to help offset the impact of future share dilution from employee stock issuances" [26].

That framing matters. Despite spending $1.4 billion — nearly triple the year's SBC expense — the diluted share count fell only modestly, from 501.9 million to 493.6 million [27]. Much of the buyback is a treadmill: it mops up the dilution that SBC creates rather than compounding per-share ownership. The silver lining in 2026 is opportunism — buying back stock at $19.53 that traded above $100 in 2024 retires far more shares per dollar, so the depressed price makes the treadmill unusually accretive if the business is worth anything like its history suggests.

Source: repurchases and stock-based compensation per FY2025 10-K Consolidated Statements of Cash Flows [28].

One consequence worth flagging: cumulative buybacks now exceed cumulative retained profit, so the balance sheet carries a small accumulated deficit — retained earnings swung to roughly negative $591 million at year-end 2025 [12]. That is a bookkeeping artifact of returning capital, not distress — but it is why equity shrank year over year despite positive earnings.

6. The moat, pinned to mechanism and evidence

Moat claims are only worth as much as the mechanism behind them. Here is the honest scorecard.

Source: retention and philosophy [29]; data marketplace of over 370 vendors [30]; UID2/EUID identity strategy [31].

The data network effect deserves emphasis because it is the least obvious: a two-sided marketplace of over 370 third-party data vendors sits on the platform, and each additional buyer makes the platform a more attractive distribution channel for data, which in turn makes it more useful to the next buyer [32]. Its neutrality — buy-side only, owning no media — is a genuine differentiator against Google and Amazon, but it is the moat most exposed to erosion as Amazon leans into its own DSP.

Where the moat is not proven: Unified ID 2.0. The company is trying to make UID2 (and its European variant EUID) the open internet's replacement for the third-party cookie, alongside OpenPath, which routes buyers more directly to publisher inventory [33]. If UID2 becomes the industry standard it would deepen the moat enormously; today it is optionality, not a demonstrated advantage — and Google's 2025 decision to keep cookies in Chrome removed the near-term catalyst that would have forced adoption.

7. Competitive reality: the largest independent, dwarfed by the walled gardens

The competitive map has two tiers. Against the walled gardens the scale asymmetry is staggering — Google's advertising revenue alone is roughly twenty times TTD's entire spend — and the filing openly concedes the industry is "highly competitive and fragmented," naming Google and Amazon as its most formidable competitors [34]. TTD's counter is that it aggregates the rest of retail against Amazon's first-party garden: on its Q1 2026 call management argued the retailers in its data marketplace represent over 80% of the sales of top U.S. retailers, versus Amazon's own share of under 15% of U.S. retail spend [35].

Against the independents, TTD is not merely the largest — it is the only one that earns a serious margin, which is the clearest evidence that buy-side scale compounds into superior economics.

Source: TTD revenue, operating income and net income per FY2025 10-K [36]; Magnite income statement [37]; Viant income statement [38]; PubMatic income statement [39].

The read-through: TTD's ~20% operating margin dwarfs Magnite (14%), Viant (4%) and PubMatic (a loss) — and it does so at roughly four times the revenue of the next-largest independent. The genuine competitive threat is not on this table; it is Amazon's DSP, the only rival with the data, the scale and the retail-media flywheel to contest buy-side share directly. Share loss to Amazon is the core bear risk, and no amount of margin superiority over the small independents offsets it.

The growth engine underneath all of this is connected TV: video, which includes CTV, has grown to a high-40s percentage share of platform spend and continues to rise as the mix shifts [40]. The migration of TV budgets from linear to biddable streaming is the reason an 18%-growth business can still credibly talk about a long runway — and it is heavily U.S.-weighted today (the United States was 85% of FY2025 revenue), leaving international as the second leg.

8. Cyclicality and the self-inflicted 2024 stumble

Advertising is cyclical — budgets flex with the economy, the fourth quarter is the seasonal high and the first quarter the low, with extra lift in U.S. election years [41]. But the defining recent event was not macro. In Q4 2024, after eight years as a public company, TTD missed its own guidance for the first time, with the CEO calling it plainly "our fault" and attributing it to a slower-than-planned rollout of Kokai — its "most recent and major upgrade" to the platform — alongside an internal reorganization [42] [43].

9. Governance and alignment — an underwriting factor, not a footnote

Two governance facts change how an investor should size position and risk.

First, control is entrenched. The dual-class structure gives Class B shares ten votes each versus one for the Class A that public investors own, so insiders — led by founder-CEO Jeff Green — hold effective voting control regardless of their economic stake [44]. Public shareholders are, in practice, along for the ride on major decisions.

Second, the 2021 CEO Performance Option is one of the largest executive grants in corporate history and its incentives are now badly out of the money. It entitles the CEO to buy up to 16 million shares (a maximum of 19.2 million with a relative-TSR modifier) at a $68.29 exercise price, vesting only if the stock hits price hurdles ranging from $90 to $340 per share — against a $819 million grant-date fair value [45]. With the stock at $19.53, every tranche is deeply underwater and the first $90 hurdle is more than 4x away. The grant has also drawn shareholder litigation [46]. The alignment cuts both ways: the CEO makes nothing on the option unless the stock roughly quintuples, but the company keeps expensing it, and the size of the grant is a real governance red flag.

10. How to value it — the entire story is the de-rating

This is the section that matters. The Trade Desk was, for years, one of the most expensively valued stocks in software — the market paid 38 times sales at the end of 2021 and still 24 times sales at the end of 2024. By mid-2026 that had collapsed to 3.3 times sales. Nothing in the business broke commensurately: revenue still grew 18% and free-cash-flow margins held at 27%. What changed was the multiple.

Source: derived from year-end share prices (price feed) and reported annual revenue; 2026 point uses the $19.53 close on 10-Jul-2026 and FY2025 revenue [47]. Market-cap inputs from the numeric price feed, as reported.

At the $19.53 price the business carries a ~$9.6 billion market cap and, because it holds ~$1.3 billion of net cash and no debt, an ~$8.3 billion enterprise value. The valuation looks cheap on almost every conventional lens — and demanding on the one lens that treats stock comp honestly.

Source: multiples computed on a $9.64B market cap and ~$8.3B enterprise value at the $19.53 close; earnings, FCF, Adjusted EBITDA and SBC per FY2025 10-K [48] [49] [50].

The right lens. This is a capital-light, high-return compounder, so the correct frame is price-to-free-cash-flow and FCF yield, adjusted for stock-based compensation — not EBITDA (which flatters by adding back the SBC) and not revenue multiples (which tell you sentiment, not value). Read that way, the two poles of the debate are clear:

• The bull case underwrites the 8.3% reported FCF yield on a business still growing 18% with a rising take rate and a CTV tailwind — an extraordinary combination that has almost never been available this cheaply in TTD's history.
• The bear case notes that once you charge shareholders for the $490.6 million of annual dilution, the honest yield is closer to 3.2%, growth is decelerating, and Amazon's DSP is the one competitor that could compress the take rate the whole thesis rests on [51].

The decisive variables are the same three to watch every quarter: the take rate (is monetization still rising or is it rolling over) [52], CTV mix (the growth engine) [53], and share versus Amazon [54]. A high-quality business does not automatically make a good stock, but at 3.3 times sales the price has already done most of the work of pricing in the risks — which is precisely why the take rate is now the only number that decides the outcome.