The Trade Desk Is Down 80%. Zuma Fixes the Product — Not the Trust

The Trade Desk rolled out Kokai Zuma this morning — its biggest platform update since Kokai launched in 2023 — and if you only read the release, the company looks like it's back on offense: agentic AI, one-click measurement, easier campaign management, and an average 32% improvement in cost per acquisition from upgraded forecasting. Read the market, and it looks like a company in free fall. The stock that sold near $88 a share in mid-2025 now trades in the low teens, a company worth roughly $6 billion. Revenue grew 18% last year, slowed to 3% in the spring quarter, and management is guiding the current quarter to a year-over-year decline — the first it has ever guided to as a public company. The gap between those two pictures is the entire investment question.

The toll booth and the pitch

The Trade Desk is the largest independent middleman for digital ad buying outside the walled gardens. Its economics are a toll. Last year advertisers pushed $13.4 billion of spending through the platform, and The Trade Desk kept about $2.9 billion of it — roughly 21 cents on every dollar, at an adjusted EBITDA margin of 41%. The pitch that justified a premium valuation was that a buyer whose budget leaks across thousands of publishers needs a neutral layer that optimizes against business outcomes rather than inventory supply; and because The Trade Desk owns neither the supply nor the demand, it can be honest about what actually works. Performance plus trust. The moat was never the interface. It was the claim that a disinterested referee with better data moves a dollar better than anyone else, and can prove it.

What actually broke it — and when

The 2026 collapse attacked both pillars, in sequence. The trust pillar went first. In March, Publicis Groupe — one of the largest ad holding companies on Earth, and one of The Trade Desk's biggest commercial channels — told its clients to stop using the platform. The trigger was a third-party audit by FirmDecisions that found The Trade Desk had applied its DSP fee on top of other charges in ways its contracts didn't describe — fee stacking, in effect. The company disputed the findings, but advertising is a trust business, and the charge struck directly at the neutrality claim. Rivals moved in publicly. WPP and dentsu stepped back from The Trade Desk's OpenPath supply product over transparency concerns, and Omnicom opened its own audit of the relationship.

Then the volume pillar sagged at the worst possible moment. Management blamed a "complex environment" plus its own slippage: auto advertisers — roughly a quarter of the business — and packaged-goods brands pulling budgets as tariffs and a squeezed lower-income consumer bit, while a CFO departure and a sweeping C-suite reset added noise. Q2 revenue came in at $715 million, up just 3%, missing estimates, and the stock fell another 22% in a day. By August it was down roughly 80% from its peak. The crowd's conclusion: the long-dethroned golden child of ad tech was a value trap, being squeezed by Google, Amazon, and agencies that would rather own the middleman than rent him.

The numbers the straight line ignores

Now run the contrary evidence — the stuff a "death of the open internet" story has to skip. Customer retention held above 95%, the same number it has printed for over a decade. The 217 clients on formal joint business plans grew 38% and are spending six times faster than the rest of the book. Publicis and The Trade Desk announced a settlement on June 12, terms undisclosed, with Publicis quietly resuming its recommendation. The CEO bought $150 million of his own stock in April. The company sits on more than a billion dollars of cash with no debt. That is not the fingerprint of an abandoned product. It is the fingerprint of a cyclical pullback layered on a trust dispute that was resolved mid-quarter.

Zuma, aimed at the thing that is actually scarce

Zuma is the company's answer on the performance side, and read through what is scarce in the market, most of it is aimed right. AI is no longer a differentiator in ad tech — Google's DV360, Amazon, Yahoo, Magnite, Mediaocean and PubMatic all market agentic AI now. Make execution abundant, and the scarce asset becomes proof of business outcomes delivered by a party with nothing to sell. That is exactly what Zuma's measurement stack does: one-click brand-lift and sales-lift studies through Lucid and NielsenIQ, geo-based incrementality, e-commerce reporting tied to retail purchase data. The Koa agents that build audiences and tune frequency are wrapped in a long-overdue interface reset — the hated periodic-table view is no longer the default. And the 32% CPA improvement is the load-bearing number; Karsten Weide, chief analyst at W Media Research, calls the performance gains "real numbers." If the tool genuinely moves CPA by a third, that is the performance edge the pitch always promised.

But keep the limits in view. The AI assistant is in open beta, some agents are still in closed beta, and the AI features are opt-in. None of this reverses a trust relationship in a single release. Zuma is the re-entry ticket to buyers' budgets, not a moat by itself.

The test

Zuma answers the product half of the equation. It cannot answer the trust half — and the market is not asking whether the product is good, because for the next quarter it already knows. Management guided Q3 to at least $650 million in revenue, down about 12% from the same quarter last year, with adjusted EBITDA margin guided to compress toward roughly 25% from 34% in the spring. That is a floor set deliberately below the current run rate, and it says plainly that neither the settlement nor the launch is expected to move revenue before the next report.

So the setup is a clean asymmetry. At about $6 billion of market value — roughly two times last year's revenue, five times its adjusted EBITDA — the market has written off both scarce assets at once: that the tool can still outperform, and that anyone trusts it enough to hand it money. Zuma makes a real, evidenced case on the first. The second is being repaired in negotiations, and the proof only shows up in reported quarters.

What resolves it is a short list. The actual Q3 number against the $650 million floor. Whether the EBITDA margin begins climbing back. Whether the agency channel that broke the stock — Publicis recovering, Omnicom's audit resolving, OpenPath partners drifting back — puts the dollars back where they were. If the decline turns out to be the anomaly, these prices are cheap in a way the market has not seriously entertained. If it is the new base rate, $13 was not a bargain; it was a mark. Zuma settles the question of whether the company can still build the best tool in the market. The next two reported quarters settle whether anyone is paying for it. The stock is currently priced as though both answers are no.