Newsletter · · Ashutosh Agarwal

Retail Media Eats the Funnel as Walmart Ads Jump and Instacart Storms Cannes - Digital Ads & Retail Media Weekly - Week of August 23, 2026

Digital advertising and retail media newsletter for the week of August 23, 2026. Walmart Connect's roughly 70 percent margins carried a weak quarter, Instacart pushed shopper data onto the open web at Cannes, and ChatGPT switched on its first ads with Criteo as the early partner.

Digital Ads & Retail Media Weekly

Week of August 23, 2026: Retail Media Eats the Funnel as Walmart Ads Jump and Instacart Storms Cannes

Week of August 16–23, 2026. Everything below is drawn from podcasts published in the last seven days. Where a speaker runs a business, I've flagged them as an operator; where they're an analyst or investor talking their book, I've said so.

TL;DR

  • Walmart's retail-media business is the story, not its soft sales. The stock got hammered on the weakest US comparable sales in six years, but advertising revenue jumped 38% and one analyst pegs Walmart Connect's margins near 70%, versus 3–4% for the core store. The pitch: "$5 in ads adds as much to the bottom line as $100 in groceries."

  • Amazon's next leg of ad growth is "non-endemic", airlines, banks and carmakers that sell nothing on Amazon but want its data. US off-site retail-media spend is set to hit $17.05B in 2026, up 30%. Amazon's ad platform is quietly becoming "one of the world's largest consumer behavior graphs."

  • Instacart made the loudest noise in retail media this week, a Cannes Lions panel on pushing shopper data out onto the Trade Desk, Roku, Pinterest and TikTok (McCormick reached an audience that was 93% new, and cut ad prices 20%+), plus a deep dive on its AI shopping carts.

  • Connected TV is being re-sold as a performance channel, not a branding toy, you can now track what a viewer does after the ad, and irrelevant ads make people flip the channel 60% more.

  • AI is all over ad earnings calls; AI revenue mostly isn't. ChatGPT turned on ads, and Criteo, its first ad-tech partner, doubled its ChatGPT advertisers to 2,000 in a quarter, while warning investors it won't move the needle this year.

  • The open-internet trade got a beating. The Trade Desk drew a brutal aside, "the wheels have absolutely come off", while walled gardens keep compounding.

What's new

1) Walmart's ad engine ran hot even as the store cooled

Walmart reported and the market didn't like it, but dig past the headline and the interesting part is the ad business. On The Rundown (Aug 20), Public.com's Zaid Admani walked through the quarter: revenue up 6% to $188B, adjusted earnings of $0.81 against the $0.74 Wall Street expected, US e-commerce up 24%, membership income up 17%, and, the line that matters for this newsletter, advertising revenue up 38%.

The ugly number was US comparable sales (sales at stores open at least a year): up just 2.6%, the slowest in more than six years, dragged partly by a new federal rule that caps drug prices and hit Walmart's pharmacy. Strip that out and it's 3.4%, still light. As Admani put it, Walmart "is becoming a much more interesting company than just a big store that sells cheap stuff … building these high-margin businesses around advertising and membership and marketplace."

Why do 38% ad growth and 2.6% sales growth belong in the same story? Because of the math on margins. On Schwab Network's Ca$htag$ (Aug 19), LikeFolio's Landon Swan spelled it out: Walmart Connect, the in-house ad platform that lets brands pay to show up in Walmart search and across its properties, runs at an estimated ~70% margin, against Walmart's 3–4% net retail margin. His framing: "Walmart selling a hundred dollars worth of groceries might add the same to the bottom line as them selling $5 in ads." That's the whole bull case for why a low-margin retailer trades at a ~40x price-to-earnings multiple, a 161% premium to its retail peers. Swan's warning was equally blunt: "This stock at this price cannot afford just pretty good … it's got to be almost perfect."

Plain-English note: "retail media" means a retailer selling ads against its own shopper data and store traffic, Walmart Connect, Amazon Ads, Instacart Ads. "CPM" is the price of 1,000 ad views; "ROAS" is revenue earned per dollar of ad spend.

2) ChatGPT switched on ads, and Criteo is the early winner (for now)

The AI-search ad format everyone has been waiting for arrived. Two podcasts confirmed it from very different angles.

On Ad Age Insider (Aug 21), senior tech editor Garett Sloane reported that Criteo is the first ad-tech partner for OpenAI's ChatGPT ads and, in the roughly three-to-four months since, doubled its ChatGPT advertisers from about 1,000 to 2,000 in a single quarter. Impressive growth rate, tiny absolute base: for scale, Meta has more than 10 million advertisers, and Criteo told investors ChatGPT "would not be a significant portion of its earnings this year." Sloane's wider point is worth holding onto: AI dominated Q2 earnings calls, but "you're not seeing the bottom line shaped by AI" yet, on the ad-tech side, "agentic" (AI-run) buying is maybe "tens of millions of dollars … you're talking about a trillion dollar industry."

From the practitioner side, marketing agency owner Lori Werner told Medical Spa Insider (Aug 19) that "ChatGPT actually announced ads" the prior week, complete with a verification process and targeting, and reminded listeners that OpenAI, "from what I hear … [is] not even profitable," which is the not-so-subtle reason the ads are showing up now.

3) Instacart owned Cannes, and showed where retail media goes next

The most substantive retail-media conversation of the week was The CPG Guys' live-from-Cannes-Lions episode (Aug 22) with Instacart's Adam Silverblatt (operator) and independent consultant Andrew Lipsman.

Lipsman's framework: retail media's "third wave", brands buying ads on a retailer's own website, is maturing into a "fourth wave: performance TV," where retail-media data flows out across the open web and streaming. His line: "the future of retail media is the future of all media."

Silverblatt made it concrete. Instacart now lets brands take its first-party shopper data (who actually bought what, tied to real delivery addresses) and activate it off Instacart, inside the Trade Desk, Roku, Pinterest and TikTok, in the brand's own ad-buying seat, with mid-campaign optimization. Two case studies:

  • McCormick ran a Cinco de Mayo campaign on the Trade Desk (with Flywheel). Of the taco-seasoning audience Instacart supplied, 93% could not have been reached with McCormick's own data, only a 7% overlap, and CPMs came down more than 20%.

  • Deep Indian Foods turned on an Instacart conversion signal inside the Trade Desk mid-flight and doubled its ROAS while cutting ad prices.

Lipsman also pressed his favorite point, incrementality, i.e. sales you wouldn't have gotten anyway: a reported $4 ROAS "isn't showing up in their financials," whereas an incremental ROAS of $1.50 "is a real 50% increase." Instacart claims 79–90% of its digital audience is incremental to a brand's physical brick-and-mortar shoppers.

The debate

Is the "open internet" ad thesis broken, or just having a bad year?

The bull story for independent ad-tech has always been that dollars would flow away from the walled gardens (Meta, Google, Amazon) toward the "open internet", every other site and streaming app, and that The Trade Desk (TTD) would be the toll booth. That thesis took hits this week.

On Investing Unscripted (Aug 19), one host used TTD as a cautionary tale about companies inflating their total addressable market: the "trillion dollar" open-internet framing is "total BS … the wheels have absolutely come off. Business is grinding to a halt. And all of the ad revenues at all of the walled gardens is growing like gangbusters." He called it "an execution story" under Jeff Green's "revolving door of a management team." His co-host pushed back, still holding, believing "the fundamentals are largely there" and preferring to ride out the volatility.

Here's the twist: even as TTD's standalone narrative wobbles, the open internet is very much alive as a destination for retail-media dollars. Instacart is routing McCormick's audience through the Trade Desk. So the pipe may be fine even if the company owning it is struggling. That's the tension to watch: is TTD a broken business, or a healthy toll road with a management problem?

Does anyone actually make money from AI in advertising yet?

Sloane's Ad Age reporting is the honest counterweight to the hype. The industry is busily building the plumbing for AI to buy ads on its own, "MCP ad servers," agent-to-agent buying, so that one day you tell your agent "go buy me media across Magnite, PubMatic, Criteo, Google and Meta" and it does. But today, "what we're seeing this year is just the foundations of that." PubMatic touts 80-plus "agentic" campaigns; the real question, Sloane joked, is "what do we consider fully agentic? … at the very least, Claude was involved somehow." The takeaway for investors: AI is in the pipes and back-office of the ad companies, but it is not yet a revenue line, and anyone claiming otherwise is "sprinkling a little AI dust into the earnings."

Stocks in play

  • Walmart (WMT). Ad revenue +38%, membership +17%, e-commerce +24%, but US comps at a six-year-low 2.6%. The debate is entirely about whether the ~70%-margin Walmart Connect / membership / marketplace stack justifies a ~40x multiple while the core store slows. (The Rundown, Aug 20; Schwab Network, Aug 19)

  • Amazon (AMZN). The under-covered growth vector: non-endemic advertisers (see Read-throughs). Amazon Marketing Cloud now lets advertisers look back five years; Amazon DSP has become a "consumer behavior graph," not just a retail ad buy. (Retail Media Breakfast Club, Aug 19; Ad Age Insider, Aug 21)

  • Alphabet / Google (GOOGL). The quiet kingmaker. Google's AI Overviews have cut click-throughs to source websites by ~60%, Google now has ~1 billion Gemini users, and it pays Reddit just $60M/year to license data, a number Reddit thinks is far too low. Google management claims it can monetize AI answers as well as classic search. (The Investor's Podcast, Aug 16)

  • Reddit (RDDT). Down ~20% after earnings. Hit a 30% operating margin by end-2024 (years early), but only 9% of monthly users log in daily (vs 35% for TikTok, 50% for Instagram), it leans on Google for over half its daily traffic, and it just stopped disclosing logged-in user counts. Bulls say it's still under-monetized with 300M users; skeptics see a user-growth problem being papered over. (The Investor's Podcast, Aug 16)

  • The Trade Desk (TTD). Under fire on execution even as its rails carry retail-media dollars. (Investing Unscripted, Aug 19; The CPG Guys, Aug 22)

  • Criteo (CRTO). First mover on ChatGPT ads; advertiser count doubled to ~2,000, but management is managing expectations for this year. (Ad Age Insider, Aug 21)

  • Warner Bros. Discovery (WBD), Disney (DIS), Netflix (NFLX), EchoStar (SATS). See CTV read-through below.

Read-throughs

Amazon's next act: advertisers who sell nothing on Amazon

The sharpest single insight of the week came from Retail Media Breakfast Club (Aug 19), where Kiri Masters downloaded her own Amazon ad-data file and found roughly 400 "advertiser audiences", Pfizer, Geico, Emirates, Chase, McDonald's, an Australian bank, companies that sell nothing on Amazon but are paying to reach her based on Amazon's behavioral data. This is "non-endemic" advertising, and it's where mature retail-media networks go once they've saturated the brands that actually sell on their shelves.

The numbers and examples:

  • US off-site retail-media spend will reach $17.05B in 2026, up 30% year-over-year (eMarketer's Sarah Marzano).

  • Home Depot's Orange Apron Media put 20+ first-party audiences into a beta with Pinterest, the first social platform where brands Home Depot doesn't even stock can target Home Depot shoppers.

  • Dick's Sporting Goods is in "very real conversations" with financial-services partners, using its GameChanger youth-sports app audience.

  • Kroger ran Chevrolet's Equinox EV launch through the Yahoo DSP based on grocery-basket signals rather than car-buying intent.

  • The catch, per an IAB white paper: chasing non-endemic dollars is only viable for networks willing to invest $100M+ over multiple years, a game for Amazon, Walmart, Kroger and Costco, not everyone.

Connected TV grows up: from billboard to performance channel

Two operators made the same argument: CTV (streaming TV ads) is being repositioned from a brand-awareness play into a measurable, direct-response channel.

  • On The Agile Brand (Aug 19), AdRoll's Allison Clark said marketers can now "see what a consumer did after they saw a CTV ad, did they do a branded search, go to the website, convert?", target down to the individual account or contact, and optimize mid-flight. Her clients' repeated finding: "you add CTV into the plan, and the rest of your media performs better and more efficient."

  • On AdTechGod Pod (Aug 19), Whirl's Mikey Garcia (Whirl was acquired by Apple) shared a striking stat: users change the channel 60% more when the ad is contextually irrelevant to what they're watching, "we call that tuning out." His pitch for FAST channels (free, ad-supported streaming) is that everyone watching is at the same point in the stream, so ads can be matched to the scene in real time, something video-on-demand can't do. His nuance: use contextual targeting with audience data, not instead of it.

And the DTC angle from DTC Podcast's "Harness the Halo" (Aug 20): Keen's Justin Jefferson (who sees 450+ brands and $45B of ad spend) says the Meta/Google "pop a dollar in and watch your brand grow" era is over, auctions are "a lot more competitive," and larger brands are moving budget up-funnel into CTV, video and audio. One golf-apparel brand saw ~+23% revenue in year two after leaning in; a brand that over-rotated "zero to 100" lost sales in year one. CTV is "the biggest top-of-funnel channel we're seeing growing," with audio the "dark horse." TikTok Shop, he adds, now rivals Meta spend for some brands.

The CTV earnings scoreboard

The Dan Rayburn Podcast (Aug 16) ran the numbers across the streamers:

  • Warner Bros. Discovery: streaming profit +75% to $512M, revenue +10% to ~$3.1B, but ad revenue fell 22%, almost entirely because the NBA left Turner. Also booked a $1.1B pre-tax acquisition/restructuring charge.

  • Disney: streaming operating income doubled to $712M on $5.5B revenue (+11%). Notably, subscription revenue grew 15% while advertising grew just 3%, "impressions higher, rates lower." A useful corrective to the "ads are the streaming growth engine" narrative: at Disney, subscriptions are still doing the heavy lifting.

  • Netflix: the co-hosts' industry sources say Netflix CPMs are "similar to the other premium content sites … they've not been able to get a higher tier" despite premium content. Pause ads are where the innovation is going.

  • EchoStar: Sling TV down to 1.7M subscribers (-80k), plus -161k satellite and -59k broadband, the slow bleed continues.

The in-store frontier: Instacart's AI carts

On The AI Why (Aug 20), Instacart's Chief Connected Stores Officer David McIntosh described the Caper smart cart as a new ad and data surface: shoppers stare at the cart screen for 30+ minutes, the most-loved feature is the running total, and location-aware coupons and recommendations drive discovery. A simple "did you forget the yogurt?" prompt at checkout produced a ~0.9% sales lift (disclosed on last month's earnings call). The carts are in 100+ cities and tripled year-over-year, powered by 1.6B lifetime online orders of recommendation data. The ad angle: "the ads that appear on the cart are more relevant … and generating more revenue for our retailer partners." Instacart's Card Assistant (agentic meal-planning) now runs inside ChatGPT and on retailer sites, with Sprouts rolling it out.

What changed vs last week

A note on honesty: there isn't a prior week's issue on file to diff against line-by-line, so here is what genuinely shifted in the conversation this week rather than a manufactured comparison.

  • Retail media stopped being an abstraction and became a P&L story. The debate moved from "retail media is growing" to specific margin math (Walmart Connect ~70%; "$5 in ads = $100 in groceries") and specific dollar forecasts ($17.05B off-site, +30%).

  • AI-search advertising went from theory to a live product with a first customer. ChatGPT ads are on; Criteo has real (if tiny) advertiser numbers.

  • The open-internet/CTV trade split in two. The independent toll-booth (TTD) is under execution fire, even as the use case for its rails (retail-media data flowing to the Trade Desk, Roku, TikTok) is accelerating.

  • CTV's pitch flipped from "awareness" to "performance," backed by hard behavioral data (the 60% tune-out stat) and mixed earnings (WBD ad revenue -22%, Disney ads +3%).