The Trade Desk, which sells agencies and brands the software they use to buy advertising across streaming television, display and audio, told investors on 6 August that revenue this quarter would be at least $650m. That is a decline of roughly 12% from a year earlier, from a company that grew 17.7% as recently as last autumn. The shares fell about 22%.

The same week, Magnite — the independent sell-side platform that publishers use to monetize their ad inventory, and the technology partner for Disney, Netflix, Roku, Paramount and Warner Bros. Discovery — raised its full-year growth guidance to 13–14% from at least 11%. Its streaming revenue net of traffic-acquisition costs rose 36% to $97.1m.

Both companies are paid out of the same connected-television dollar. Only one of them is losing it.

The fee moved, the impressions didn't

The mechanism is price. Amazon's demand-side platform charges no fee on programmatic-guaranteed deals across Amazon-owned media and 1% on open-web publishers, against The Trade Desk's estimated 20% take rate. Amazon's share of global programmatic buying went from under 10% to just under 20% in roughly fifteen months.

Agencies have acted on it. Publicis stopped recommending The Trade Desk after an audit of its fees, billing and the behavior of Kokai, its artificial-intelligence buying tool; ad-tech sources say Omnicom agencies moved a double-digit share of third-quarter programmatic budget to Amazon. The Trade Desk kept the clients and captured less of their money — gross spend from existing customers declined year on year. Management said on the call that its take rate has been flat for a decade. That is the problem, not the defense: a flat 20% is now priced against 1%.

Magnite sits on the other side of that trade and does not care which buyer wins. Its unified ad server and exchange reach 99% of US streaming supply on a dollar-weighted basis. Second-quarter revenue rose 11.2% to $192.8m, and operating margin widened to 16.2% from 12.7%.

Where the open web is actually shrinking

Criteo, the Paris-based firm paid per retargeted click and per retail-media placement on retailer websites, is the clearest damage. Revenue fell 11.3% to $428.0m and operating income halved. It cut its outlook to a 10–12% constant-currency decline, the third position on that number in six months and each worse than the last, after two retail-media clients pulled $75m of scope. Strip those two and retail-media contribution still grew 20%, with 85 retailers now running auction-based display against 60 a year ago.

DoubleVerify, which is paid per impression it verifies as viewable and fraud-free, chose the exit: Nielsen agreed on 6 August to buy it for $13.60 a share in cash, about $2.15bn including debt. Revenue had grown 2.5%. At $13.27 the shares are a merger spread, not a verdict on advertising.

Digital Turbine, which pre-installs and recommends apps on carrier and handset-maker devices, supplied the cleanest description of where the audience went: management put open-web traffic down about 10% year on year on artificial-intelligence displacement, with some categories off 20–40%. Its revenue grew 26.8% to $166.0m and adjusted margin widened 640 basis points, though it remains loss-making on a reported basis. None of this is a boom. The Interactive Advertising Bureau (IAB) forecasts 9.5% US ad-spend growth this year; dentsu cut its global figure to 5.0%. Growth here is share, taken from somebody.

Does the price leave room

The Trade Desk at 13.45x forward and 15.79x trailing earnings, 7.02x trailing EV/EBITDA and a 13.6% free-cash-flow yield looks like the cheapest cash flow in advertising technology. The forward number is an artifact. Consensus 2026 revenue of $3.184bn against first-half actuals of $1.404bn and a third quarter guided to $650m would require about $1.13bn in the fourth quarter — a 33% jump, in a year the company is guiding down. What is real: $1.5bn of cash, $269m of buyback authorization left after $78m repurchased, and founder-CEO Jeff Green's roughly $148m open-market purchase in March, the largest insider buy in company history. It has not worked yet: the shares have held a downtrend since 13 November, 187 sessions with the 50-day average below the 200-day.

Magnite carries the opposite risk. Up 97.5% in 90 days, it trades at 23.09x trailing EV/EBITDA with a forward earnings multiple above its trailing one — the richest name here, against Criteo's 3.17x EV/EBITDA and 0.79x book. And its month is thin: remove Magnite's two best sessions, both in the week of 4 August, and a 29% gain becomes 1.7%.

The setup

Where it stands — The Trade Desk is losing fee share to Amazon's near-free DSP while Magnite, on the sell side, keeps earning on the same streaming impressions. Would confirm — Magnite's streaming revenue net of traffic costs holding above 25% growth next quarter while The Trade Desk prints at or below its $650m guide. Would invalidate — The Trade Desk reporting gross spend from existing clients back in growth, or a take-rate cut that stabilizes revenue. Watch next — Third-quarter results from both companies in early November; Nielsen expects to close the DoubleVerify purchase by the first quarter of 2027. Valuation — The Trade Desk 13.45x forward against 15.79x trailing on an unreachable estimate; Magnite 21.81x forward, above its 20.97x trailing.