Newsletter · · Ashutosh Agarwal

Amazon's Ad Machine Compounds While the Trade Desk Slides and Ads Come to ChatGPT - Digital Ads & Retail Media Weekly - Week of August 9, 2026

For the week of August 3 to 9, 2026, Amazon Ads hit $19.8 billion and grew 26%, ten points faster than the marketplace beneath it, the Trade Desk's growth collapsed from 12% to 3% and drew three downgrades in a morning, and six separate shows converged on the race to put ads inside AI chatbots.

Digital Ads & Retail Media Weekly

Week of August 9, 2026: Amazon's Ad Machine Compounds While the Trade Desk Slides and Ads Come to ChatGPT

TL;DR

  • Amazon's advertising business is now the most important number in commerce. Amazon Ads hit $19.8 billion last quarter, up 26%, growing ten full percentage points faster than the marketplace it sits on top of. Sponsored-ad clickers convert 48% more often and spend 21% more than everyone else.
  • The Trade Desk had a brutal week. Revenue growth collapsed from +12% to +3% in a single quarter, with the company now guiding to a possible decline next quarter. Shares fell as much as 29%, three brokers downgraded it in one morning, and even bulls are now calling it a "value trap."
  • The single busiest theme of the week was ads inside AI chatbots. One expert called ChatGPT ads "the most targeted ad platform we may have ever seen in history." OpenAI is reportedly rolling out shopping-style "product carousel" ads for the holidays and partnering with a measurement firm, while, tellingly, taking a far more open, less walled-off approach than Google.
  • AppLovin got punished, down about 20% on earnings, cut in half for the year, as its cash generation stopped keeping up with its profits.
  • The big-picture debate, courtesy of valuation professor Aswath Damodaran: the ad businesses at Meta, Google, and Amazon are healthy cash engines, but they're now being used to fund an AI build-out that "nobody has any idea how it pays off."
  • Read-throughs: Pinterest is thriving by telling you to leave the app; Warner Bros. Discovery's TV ad revenue cratered 27%; and connected-TV inventory is quietly full of the same bot fraud that plagued the old web.

What's new

Amazon quietly built the third great ad platform

If you only remember one figure from this week, make it this one. On Selling on Giants (Aug 4), Bellavix co-founder Will Haire laid out Amazon's quarter with the eye of someone who spends other people's money on the platform every day: Amazon's total sales grew 20% to $200.6 billion, third-party seller services grew 16% to roughly $46.8 billion, and advertising grew 26% to $19.8 billion.

Why does the gap matter? Because, as Haire put it, Amazon "is increasingly growing its marketplace revenue through what brands pay to be seen, not only through what they sell." Advertising is compounding about ten points faster than the seller business underneath it. His warning to brands was blunt:

"If your 2027 plan assumes your [ad costs] stay flat while Amazon's advertising business keeps compounding faster than marketplace services, the spreadsheet is doing motivational speaking. It is not forecasting."

The engagement numbers explain why advertisers keep paying. Shoppers who click a sponsored listing convert 48% more often and spend 21% more than those who don't. Amazon's automated "ads agent" is delivering 8% lower cost per impression and 6% lower cost per acquisition. And Amazon's voice-and-AI shopping assistant reached more than 350 million customers over the past year, with active users nearly doubling in the quarter and interactions up more than fivefold, and those shoppers spend about 40% more per order in the US.

That $19.8 billion and +26% figure was independently confirmed a few days later by Forrester analyst Nikhil Lai on Marketecture (Aug 7), who added a striking read-through for the smaller players: Reddit's ad revenue grew 64% year over year.

(One plain-English aside: "retail media" just means the ad business a retailer runs on its own website and app, Amazon Ads and Walmart Connect are the giants. It's become one of the fastest-growing corners of all advertising because the retailer knows exactly what you buy.)

The land grab of the week: ads inside ChatGPT

Six separate podcasts touched the same idea this week, that the next big advertising platform is being built inside AI chatbots, making it the single most-discussed theme of the sweep.

The most quotable case came from Profound's Josh Blyskal on Content and Conversation (Aug 4):

"ChatGPT ads, however nascent, are going to prove to be one of the most important innovations in advertising in the last decade. I think it is going to be the most targeted ad platform we may have ever seen in history."

His kicker: "Until Google has a version of it."

Forrester's Nikhil Lai (Marketecture, Aug 7) added the detail that makes this real rather than hypothetical. The money already flowing into ChatGPT ads, he's hearing from brands, is mostly "display" dollars (the budgets used for banner-style awareness ads) rather than search dollars, "paid search remains fully funded." More concretely, OpenAI is reportedly:

  • launching "product carousel" ads, shopping-style catalog listings pulled straight from a retailer's product feed, heading into the Q4 holiday season, a format that looks a lot like what Criteo already runs across the web;
  • partnering with measurement firm AppsFlyer to bring in-app ad measurement to ChatGPT.

Lai's most interesting observation was strategic: everyone assumed OpenAI would copy Google's "control everything, let no data leave the building" model. Instead, "they're being a lot more flexible on data sharing," letting outside partners "nose into the tent." If that holds, it's a genuinely different, and more advertiser-friendly, posture than the walled gardens.

The other side of AI search is getting your brand mentioned by the chatbots in the first place. Danny Kirk of ReddiReach (Talk Commerce, Aug 4) explained why Reddit sits at the center of it: the AI models buy Reddit's data because subreddits are "little nation states" of passionate, constantly-updated human opinion. His striking data point on how fast this is normalizing: Reddit made up roughly 80% of all AI-search citations last September, but that's already fallen to about 20% as the models diversify their sources, and he expects it to keep dropping toward high single digits.

The "did I just buy that with AI?" illusion

A genuinely fun finding from Merkle's Global Chief Strategy Officer, Holden Bale, on the Retail Media Breakfast Club (Aug 6): 11% of consumers believe they have already bought something through an AI chat assistant like ChatGPT or Gemini.

That's essentially impossible today, the checkout plumbing and the sales volumes simply aren't there yet. But Bale's point is that it doesn't matter that they're wrong; it matters that they believe it. He's seen the exact same illusion with social commerce (where a majority of older consumers were convinced they'd bought through Instagram when they'd actually clicked out to a website). His conclusion: "If a consumer thinks they're shopping through a thing, they're shopping through a thing." The moment real AI checkout arrives, that pre-installed halo means shoppers are already primed to use it.

The debate

Is the advertising cash cow being milked to feed an AI bet nobody can justify?

That was the thrust of the week's heavyweight conversation: valuation professor Aswath Damodaran on Prof G Markets (Aug 7), in an episode bluntly titled "Big Tech Has No Idea How AI Pays Off."

His starting point is that the top-line looked great, Microsoft revenue up ~20%, Amazon up 20%, Meta up nearly 30%, but the cash tells a different story. Meta's free cash flow fell 91%. Amazon's and Google's both went negative, something Damodaran noted almost no one would have predicted a few years ago. His framing:

"These companies now are the equivalent of manufacturing companies. They're building huge capacity... and like all manufacturing companies historically, they're now going to be judged on whether they can deliver the earnings on this investment. Something they've never had to do historically."

He also flagged a number that should make anyone reading big-tech earnings pause: a huge chunk of reported profit isn't from operations at all. About 85% of Amazon's net income came from unrealized paper gains on its stakes in AI labs Anthropic and OpenAI; the figure was ~87% for Google (from SpaceX and Anthropic). That inflates net income and makes the P/E ratios look deceptively cheap. And much of the AI "revenue" the hyperscalers report is money going in a circle, Barclays estimates OpenAI and Anthropic account for 73% of Amazon's AI revenue; Wells Fargo puts it at 74% of Microsoft's. As Damodaran put it, "if people are not buying your end product... what difference does it make that you generate revenue from each other?"

The tell he keeps coming back to: Zuckerberg announcing that Meta will sell its "excess compute," and being unable to answer directly when asked how the AI spend will pay off. "I think they truly don't know. And I think they're afraid to say that."

The counter-view came from Morningstar's Meta analyst Malik Khan (via The Morning Filter, Aug 3). Even though Meta's stock fell 8% on the print (a miss plus soft revenue guidance), Morningstar held its $850 fair value estimate. Khan's thesis is that the core ad machine keeps getting better: AI-driven content and ad recommendations are letting Meta "drive higher price and volume levers in its ad business," and he expects operating efficiencies to keep margins in the mid-to-high 30s over the next five years. In other words: the ad engine is real and strong; the open question is whether the AI spending on top of it earns its keep.

And notably, Damodaran himself still owns five of the "Magnificent Seven" and can "live with" the valuations, his worry isn't collapse at the giants (their debt is manageable at roughly 1.5x EBITDA) but the "lesser AI" companies carrying real leverage, for whom an AI shakeout "will be catastrophic."

Stocks in play

The Trade Desk ($TTD), the week's disaster. The independent ad-buying platform had what analysts on Motley Fool Hidden Gems (Aug 7) called a sniff-test failure. Revenue growth went from +12% (Q1) to just +3% this quarter, and the company guided to a possible 12% revenue decline next quarter, for a business in a growing market. Shares fell around 20% on the report (down 29% pre-market per Schwab Network), leaving them roughly 90% off their highs and at "the lowest levels since 2019." Three brokers, Susquehanna, Evercore ISI, and Truist, downgraded it in a single morning.

CEO Jeff Green spent much of the call blaming macro conditions (soft spend from advertisers like Nike). The analysts weren't buying it: "This is an execution thing... blaming the macro conditions is always a red flag for me as an investor." The deeper worry is structural, the walled gardens (Meta, Alphabet, Amazon) are all growing double digits off much bigger bases, and even smaller ad-tech names like Magnite ($MGNI), PubMatic ($PUBM) and Zeta are posting strong growth, which makes The Trade Desk's stumble look company-specific. Despite a forward P/E of ~6.6 on a ~$5 billion enterprise value, the verdict was "value trap for now," with calls for Green to step up to executive chairman and hire a proper CEO.

AppLovin ($APP). Down about 20% after earnings and -50% year-to-date (from roughly $700 at the start of the year), per Luke Guerrero on InvestTalk (Aug 7). The whole business is staked on its AXON machine-learning ad engine, now trying to expand from mobile gaming into e-commerce. The metric that spooked the market wasn't growth, it was cash: free-cash-flow margin collapsed from 70% to 45%. "Cash generation is just not keeping up pace with earnings... one of the most brutal drawdowns I've seen in quite some time." (Worth noting the operator view is more constructive: performance marketers on the DTC Podcast said AXON genuinely works for them, it just "requires so much volume.")

Meta ($META). The stock fell 8% on the quarter but, as Damodaran noted, promptly clawed back about 70% of its prior six-month losses within a week, "the lessons aren't sticking." On the ground, performance advertisers remain constructive. Sam Polaro of The Moonlighters (My Amazon Guy, Aug 7) argued the much-feared "Andromeda" algorithm update has actually improved results, and made the case that obsessing over return-on-ad-spend is a mistake: "There's a button in Facebook to spend more money. There is not a button to get more ROAS." His rule of thumb, target a ROAS just above breakeven (e.g., 1.43 for a 10% first-order profit) and pour on volume. Meanwhile Taylor Holiday brought $1.4 billion of ad-spend data to The Andrew Faris Podcast (Aug 6) showing Meta's automated bidding is powerful in aggregate but noisy per account, a reminder that the platform's AI is doing more of the media-buying every quarter.

Read-throughs

Pinterest ($PINS), winning by telling you to log off. CMO Claudine Cheever (Brand New World, Aug 5) walked through the counterintuitive strategy behind the "the best thing you can find online is a reason to go offline" campaign. It's working: Q1 revenue up 18% to over $1 billion, monthly users up 11% to 631 million. Her pitch to advertisers is that intent beats attention, 96% of Pinterest's ~80 billion monthly searches are unbranded, meaning users are actively looking for products to buy. "What a great fertile place for an advertiser to be." The company is pushing its full-funnel performance product, Pinterest Performance Plus.

Connected TV / Warner Bros. Discovery ($WBD), Disney ($DIS), Roku ($ROKU). On Inside The Stream (Aug 7), the WBD numbers were ugly: studio revenue -39% to $2.3 billion, TV distribution -9%, and TV advertising revenue down 27% year over year, largely from losing NBA rights (plus March Madness and NHL). A combined WBD-Paramount would carry roughly $6 billion a year in interest on ~$80 billion of debt. Meanwhile Disney is exploring a free, ad-supported "FAST" service and third-party aggregation inside Disney+, and Fox is buying Roku, which would combine the Roku Channel (3.1% of US TV viewing) with Tubi (2.3%) for a 5.4% free-streaming share. The throughline: soaring sports-rights costs increasingly favor the deep-pocketed tech buyers (Amazon, Google, Netflix, Apple) over the legacy media companies.

YouTube. On Next in Media (Aug 6), agency executives said YouTube has reached "parity" with traditional TV among mid-size and independent agencies, now included in nearly every campaign, thanks to its targetability, ad formats, and measurability. That's a quiet win for Alphabet's biggest video ad asset.

Ad measurement (DoubleVerify $DV, Integral Ad Science $IAS). The Marketing Architects (Aug 4) made the case that "media quality" is a term nobody can define, and warned that a 20-year-old web scam called "reach extension" (padding campaigns with junk sites and bot traffic) has crept into connected TV, showing up even in direct deals with premium publishers. The practical advice: run independent verification (DoubleVerify, IAS) separate from your ad platform's own reporting, and demand log-level data. Their view on AI is balanced, it lowers the cost of producing junk inventory, but it's also the best tool for catching it.

Walmart Connect / Instacart ($CART). Retail-media practitioners (Better Advertising with BTR Media, Aug 4) described a busy back-to-school season spread across Amazon, Walmart, and Target's ad platforms, with 62% of shoppers already mid-journey and Walmart rolling out in-store QR codes and one-click "add your school's list to cart" tools. Separately, Walmart "finally gave advertisers a tool Amazon sellers have used for years" (tighter ad controls), per Selling on Giants.

What changed vs last week

  • The AI-search-ads story moved from theory to product. For months this was a "someday" conversation; this week brought concrete detail, ChatGPT product-carousel ads timed for Q4, an OpenAI and AppsFlyer measurement partnership, and evidence that display budgets (not search budgets) are the first to migrate.
  • The Trade Desk's slow decline became an acute crisis. A single-quarter growth collapse plus three simultaneous downgrades turned a "show me" story into an open "value trap or falling knife" debate.
  • Amazon's ad business printed a fresh milestone (~$20B per quarter, +26%), cementing it as the clear number-three ad platform and the fastest-growing part of the whole retail-media wave.
  • The framing around big tech shifted from "capex is scary" to "the cash flow already turned negative." The debate is no longer hypothetical, Meta's, Amazon's, and Google's actual free cash flow moved the wrong way this quarter.