Retail Media's Concentration Problem: What $196 Billion in Ad Spend Reveals About Platform Power

A market that reached $196 billion, and who actually owns it

Retail media is now one of the fastest-growing lines in the entire advertising economy. Worldwide spend is forecast to rise about 12.4 percent to roughly $196.7 billion in 2026, with the US market alone projected between $69 billion and $71 billion, up from around $58.8 billion in 2025. By these estimates retail media approaches 18 percent of total US digital ad spend and about 16 percent globally, and more than 80 percent of digital advertisers report an allocated retail-media budget.

Growth of that scale invites a familiar story about a democratized new channel. The distribution of the money tells a different one. The same forecasts expect Walmart and Amazon together to capture more than 89 percent of the incremental retail-media dollars added in 2026. The category is enormous and expanding, and almost all of the expansion accrues to two firms. Before drawing any conclusion from those figures, it is worth being precise about what they are and what they are not.

What retail media networks are, and why they grew so fast

A retail media network is the advertising business a retailer builds on top of its own store. When a shopper searches "moisturizer" on a marketplace and the first results are labeled sponsored, that placement was sold in an auction the retailer operates. The retailer brings three assets that classic search and social platforms historically could not match at the point of purchase: first-party data on what real buyers actually bought, control of the digital shelf where the decision happens, and closed-loop measurement that ties an ad impression to a verified sale inside the same system.

That combination is why the category scaled so quickly. It attaches advertising to the moment of transaction rather than to an earlier stage of intent, and it does so with measurement the advertiser cannot easily dispute, because the platform owns both the exposure and the outcome. The result is a surface that behaves like sponsored search, priced by auction, but sits directly on the retailer inventory a buyer is already browsing.

The concentration problem, read precisely

The load-bearing figure is the 89 percent share of incremental dollars, and the word incremental matters. It describes the new money entering the category in 2026, not the entire installed base of spend. A concentration this steep in the flow of new dollars is the sharper signal, because it describes where the category is heading, not only where it has been. When almost nine in ten new dollars route to two destinations, the practical market for everyone else is not the $196 billion headline. It is the remaining sliver of growth that the two largest networks do not absorb.

This is a concentration of gatekeeping power in the ordinary economic sense. Two platforms own the shelves where a large share of purchase-intent advertising now clears, they set the auction rules, and they hold the measurement that grades the outcome. An advertiser transacting there is not choosing among many comparable venues; it is renting position on infrastructure a competitor also controls as a retailer. The size of the pie is not the story. The ownership of the ovens is.

The source matters for reading it correctly. These are eMarketer forecasts, corroborated by WPP Media and WARC commentary, and they are syndicated-research estimates rather than independently audited or peer-reviewed measurements. They should be treated as directional, and any specific percentage cited long-term needs re-verification against primary data. The concentration pattern is well-supported across the industry trackers; the exact decimals are the part that will move.

Why platform power here is a feature of the mechanism, not an accident

It is tempting to read the concentration as a temporary head start that competition will erode. The auction theory underneath sponsored placement suggests otherwise. Retail media inventory is sold through position auctions, the same family of mechanisms that governs sponsored search. In the generalized second-price auction that dominates this design, the highest bidder wins the top slot but pays the second-highest bid, and so on down the ranking.

Two properties of that mechanism explain why scale compounds rather than dissipates. First, the value of a slot depends on the quality of the targeting and measurement wrapped around it, and both improve with the volume of first-party purchase data the platform holds. A network with more buyers produces better matches and more defensible measurement, which draws more advertisers, which funds more data, a loop that favors whoever is already largest. Second, the generalized second-price auction is not incentive-compatible: it has no dominant-strategy equilibrium, and truthful bidding is not optimal, so competing well requires modeling the behavior of rivals rather than simply declaring willingness to pay. That modeling is easier for the platform, which sees the whole auction, than for a small advertiser bidding into it.

What it means to be a smaller advertiser on someone else's shelf

For a small or mid-size advertiser, the concentration figure translates into a concrete strategic constraint. The largest, highest-intent surfaces are exactly the ones where the auction dynamics favor incumbents with deeper data and larger budgets, so the marginal cost of visibility on those shelves tends to rise fastest for the advertisers least able to absorb it. Being "priced out of the biggest surfaces" is not a complaint about fairness; it is a straightforward reading of who wins a position auction that rewards scale.

This does not mean the surface is useless for smaller advertisers. It means the real planning question is not "should we be on the biggest retail media network" but "where does our budget buy defensible position rather than renting the most contested inventory at the price the largest bidders set." That is a portfolio question about where a constrained budget competes, and it is the same discipline that governs any spend under uncertainty: concentrate where the return is measurable and defensible, and refuse the surfaces where the mechanism guarantees you are the marginal bidder subsidizing someone else's data advantage.

The surfaces opening while these concentrate

The retail media story runs alongside a countervailing move. In 2026 the newest paid surface, advertising inside AI chat, went the other direction on access. OpenAI began testing ads in ChatGPT in February 2026 and, in May 2026, opened a self-serve ad platform that eliminated the prior $50,000 minimum spend, bringing the surface within reach of small and mid-size advertisers for the first time. Ads there are matched to conversation context and labeled as sponsored, visually separated from the organic answer.

The contrast is instructive rather than reassuring. A new surface can open cheaply to smaller advertisers and still inherit the same auction and disclosure logic as sponsored search, which means it can concentrate later as data advantages accumulate, exactly as retail media did. The durable lesson is not that one surface is friendly and another hostile. It is that every one of these venues is an auction on someone else's inventory, and the party that owns the data and the measurement holds the structural advantage regardless of how open the door looks on launch day.

How to read a forecast like this

A $196 billion figure and an 89 percent share are the kind of numbers that get repeated until they harden into fact. They should not. They are syndicated industry forecasts, useful for direction and dangerous when quoted to the decimal as if they were audited results. The academic literature on advertising is emphatic that platform-reported and forecast figures systematically overstate precision, and the disciplined response is to separate the parts of a claim that are well-evidenced from the parts that are estimated.

Here the split is clean. The auction mechanism that makes concentration durable is established, peer-reviewed economics. The direction of the retail media market, large and consolidating toward two platforms, is consistent across every major tracker. The specific magnitudes, the exact dollar totals and the exact concentration percentage, are the emerging, industry-forecast layer that will move and must be re-verified. Reading the three tiers separately is what lets a business act on the genuine signal without betting on a decimal. The pattern is the decision-grade fact. The precise number is a footnote with a shelf life.