A retailer’s search results page holds a fixed number of sponsored slots, and on the largest networks they are now mostly sold. On-site retail media, the listings that made the websites of Amazon and Walmart the fastest-growing ad medium of the decade, is a finite shelf: a retailer can raise the price of a slot, but it cannot add slots without burying the organic results a shopper opened the app to find. So the money, and the first-party data underneath it, is moving to the screen that still has unsold attention: the television.

The shelf filled because the channel matured faster than its inventory could grow. eMarketer’s forecast has US retail media reaching about $69 billion in 2026, after growth cooled to roughly 18 percent in 2025, well off the pace it set when sponsored search was still a novelty, by Adtelligent’s compilation of the figures. That deceleration is a supply ceiling, not soft demand: a search page and a product page hold only so many paid positions before the ads crowd out the result the shopper came for. A retailer can lift the price of a slot, and it has; it cannot conjure a slot that does not exist.

The destination is connected television, and what separates this turn from the long-running promise of shoppable TV is the data riding along. Off-site retail media is growing at twice the rate of on-site through 2026. Within that shift, one format is pulling away from the rest. Connected-TV retail media expanded about three times faster than retail-media search in 2025, because a 2026 television can be addressed with the same purchase records that made sponsored search work. Earlier streaming buys never had that input. The audience had moved to the couch years ago — only now can the retailer’s data follow it there.

Walmart did not wait to rent the screen; it bought one. In December 2024 it closed a $2.3 billion purchase of Vizio, whose SmartCast system sits in more than 19 million active accounts. That acquisition handed Walmart Connect two assets that used to live in separate files: the glass in the living room, and the content-recognition data that records what plays on it. The purpose is to join what a household watches to what that same household buys at Walmart. Around the same time, Walmart had fed its in-store purchase data to NBCUniversal so brands could target Peacock and broadcast audiences and tie the resulting sales back to the shelf; its retail-media chief called it a way of “tying our in-store purchase data to linear audiences.”

Instacart owns no television, so it rents space across three that do — NBCUniversal’s Peacock and linear channels, Roku’s home screen, and YouTube — and sells the one asset they lack. It matches its record of what shoppers actually buy, inside a clean room, against those audiences in deals struck across 2024 and into 2025. On Roku the loop runs in plain view: a viewer scans a code on the ad and the product can arrive from an Instacart retailer in as little as an hour. None of this is a media deal in the old sense, where a brand bought attention and hoped; the inventory being sold is the purchase record itself, and the screen is only where it gets spent.

The pattern is not a grocery story, even if grocery moved first, and fashion’s own networks show why it spreads. Nordstrom built a media network years ago on the purchase and browsing records of 32 million customers, and it has long sold off-site YouTube video beside its on-site listings, targeting shoppers by what they have bought. A department-store or beauty seller faces a sharper version of the saturation problem than Amazon does, not a gentler one: its site draws a fraction of the traffic, so its on-site shelf fills at a far lower ceiling, and renting another company’s screen becomes the only way to grow the network. The same finite-shelf arithmetic now faces every department-store and beauty retailer, and it points at the same living room.

The product these retailers ship to Roku and Peacock is knowledge of what a household already buys, rented out by the impression.

The strongest case against all of this is that it rests on measurement that may not survive an honest test. Connected-TV retail media is sold on closed-loop attribution, and the returns quoted to win budgets are scored by the seller’s own clean room. 51% of retail-media marketers already name measurement as a key obstacle to CTV investment; the eight-to-seventeen-times ad-spend figures from Instacart’s NBCUniversal beta are self-reported beta results, not incrementality studies. Incrementality has a way of deflating exactly this kind of number by stripping out the sales that would have happened anyway. If advertisers run clean holdout tests and the CTV premium shrinks toward the duller fact that television mostly reminds people of what they already meant to buy, then budgets snap back to the search bar, where intent is observed rather than inferred. That is the condition under which this thesis fails, and it is a real one. It is also beside the point, because the land grab never required CTV to beat search; it required the on-site shelf to be full, and a full shelf sends the next dollar to whatever inventory still carries the retailer’s data, mediocre returns and all.

The retailer’s real asset was never its website; it was knowing what a household buys, week after week, with a confidence no rival advertiser could match. Selling that knowledge into Peacock and Roku is a decision about how far it travels from the store, and the distance is now considerable: the loyalty card that used to sit in a wallet has been rebuilt as something that watches television alongside the household and recognizes the ads. Amazon, which already dominates overall retail media, is forecast to clear $75 billion in retail-media revenue by 2028, a reach the smaller networks cannot touch, which is why they are selling the one thing they own outright. For a fashion or beauty retailer, building a media network is no longer the decision; the decision is how much of its customers’ lives it will license to the screen in their living room to keep the network growing.