Retail media grew up.

Commerce media reached $178.2 billion globally in 2025, or 15.6% of all advertising, and overtook total television revenue, streaming included, for the first time (WPP Media, This Year Next Year, December 2025). Coresight Research independently sized the 2025 market within one percent of that figure (Coresight Research, January 2025; the study was Criteo-sponsored). The category that began as sponsored search on retailer websites is now the third structural pillar of advertising, after search and social.

The channel's operating reality lags its size. Growth has roughly halved from its 2023 peak, nine of every ten incremental US dollars now flow to just two networks (eMarketer, 2026), and only 15% of buyers report strong confidence in their retail media measurement (Skai, 2026).

Those three facts arrive together in 2026 planning season. The rest of this piece covers what changed, where the money goes now, and what a disciplined buyer does about it.

The model that sizes global retail media at $257 billion by 2028 also shows its growth rate falling from 23.8% in 2023 to 11.6% by 2028 (Coresight Research, January 2025; Criteo-sponsored). WPP Media pegs 2026 retail media network growth at 11.3% (WPP Media, December 2025). Deceleration on a doubling base is the profile of a maturing channel: it stops rewarding early adoption and starts rewarding allocation discipline.

The share keeps climbing even as the curve flattens. Retail media rose to 23.2% of global advertising in 2025, up from 20.6% in 2024 (Coresight Research, January 2025). In the US, eMarketer expects roughly $70 billion in retail media spend in 2026, up from $58.8 billion in 2025 (eMarketer, 2026).

The squeeze is sharpest at the funding source. US CPG ad spending growth slowed to 4.6% in 2025, below the national average for the first time in three years (eMarketer, 2025). The category funding retail media is under margin pressure of its own, so every retail media dollar faces sharper internal competition.

Concentration is the other defining number. Of the $10.5 billion in incremental US retail media spending expected in 2026, $9.4 billion goes to Amazon Ads and Walmart Connect (eMarketer, 2026). Amazon's ad business passed $68 billion in 2025, with Prime Video reaching 315 million ad-supported viewers in the fourth quarter (Amazon Q4 2025 earnings). Walmart's global ad business grew 46% to $6.4 billion in fiscal 2026, with Vizio ad revenue up triple digits in the fourth quarter (Walmart FY2026 earnings).

eMarketer projects Amazon's retail media business above $75 billion by 2028, roughly $65 billion ahead of the next-largest network (eMarketer, May 2026). The category grows while the set of choices that matter shrinks, and buyers gain negotiating power everywhere except where they spend the most.

Buyers now rank incrementality as their number one KPI: 71% put it ahead of ROAS, yet only 15% report strong confidence in their measurement (Skai, 2026). The KPI that decides budgets is the one buyers trust least.

When independent incrementality tests run, validated returns typically land 30 to 60% below the last-click ROAS networks report (The Trade Desk's The Current, 2026). Call that spread the proof gap. It works like a discount rate that buyers quietly apply to every reported number.

The standards already exist. The IAB and MRC published retail media measurement guidelines in January 2024, and Albertsons handed the IAB a full standardization framework in September 2024. Adoption still lags, because opacity protects network economics.

Buyers and networks are now negotiating the same number from opposite ends. Networks report the generous last-click figure, and buyers increasingly hold spend until an independent test confirms the incremental one. The spread between the two is where 2026 budgets are won and lost.

Brands now route 53% of retail media spend to off-site channels, the first off-site majority in dentsu's series (Dentsu Retail Media Industry Report 2025; survey of 100 US retailer and brand decision-makers, fielded August 2024). On-site sponsored inventory is finite, so concentrated demand inflates its price and pushes spend outward into CTV, social, and the store itself.

eMarketer puts US retail media CTV spend at $6.10 billion and retail-data-powered social spend at $7.76 billion for 2026 (eMarketer, 2026). The pattern behind both numbers is the same: where an impression runs matters less than which data prices it.

The store is the largest under-measured screen. Roughly 77% of US retail sales still happen inside physical stores, yet only a fraction of retail media dollars reach them (Coresight Research, 2025). Electronic shelf labels, TV walls, smart carts, and in-store audio are arriving ahead of the measurement standards, and dentsu already classes 28% of retail media network spend as off-site in-store, the fastest-growing slice in its survey (Dentsu, August 2024 survey).

Below the top two, the mid-tail is consolidating. More than 200 retail media networks have launched since 2010, a cumulative count reported by Mimbi via Coresight Research (Coresight Research, 2025), but only 35% of brands added a new network in 2024, down from 58% the year before, and two-thirds call the current count enough (Dentsu, 2025). 45% of networks now earn more than 40% of their value from non-media solutions: data, audiences, and insights sold without media attached (Dentsu, 2025).

Every added network also costs more than its insertion order. Duplicated teams, lookback windows and halo-revenue rules that refuse to reconcile, and joint business plans that credit $1,000 of self-serve spend as $750 all add up to a fragmentation tax (Dentsu, 2025). It is worth paying only where the audience is genuinely unique.

The model escaped the store. Chase Media Solutions launched in April 2024 as the first major bank media network, and PayPal, Klarna, and Revolut have built ad businesses on payment data (company announcements, 2024-25). United's Kinective Media arrived in June 2024 as the first airline network, and Marriott Media followed in mid-2025 with 237 million Bonvoy members and pilot advertisers including PepsiCo, Visa, and American Express (Mi3, 2025; company announcements).

WPP Media's accounting makes the scale point directly: retail, financial, and travel media together passed all of television by the end of 2025 (WPP Media, December 2025). Instacart's ads business alone reported $771 million in revenue through three quarters of 2025 (Instacart 10-Q, 2025). Commerce media has become a first-party-data story wearing whatever logo issued the loyalty card.

Agents are the newest surface. eMarketer expects AI platforms to mediate $20.6 billion of US retail ecommerce in 2026, about 1.5% of the total and nearly four times the 2025 figure (eMarketer, 2026), and McKinsey sizes the agentic commerce opportunity at $3 to 5 trillion globally by 2030 (McKinsey & Company). Amazon serves sponsored placements inside Rufus, Google surfaces sponsored products in AI Mode, and OpenAI is testing ads in ChatGPT with retail partners (platform announcements, 2025-26).

Beneath those surfaces sits a protocol race. Google's Universal Commerce Protocol counts Shopify, Etsy, Wayfair, Target, and Walmart among its partners, while OpenAI's Agentic Commerce Protocol, built with Stripe, has powered Instant Checkout in ChatGPT since September 2025 (Google; OpenAI, 2025-26). Agents bypass the search shelf that funds retail media, yet they reward exactly what networks hold: verified purchase data and the checkout itself.

Concentrate the portfolio on purpose. With roughly 89% of incremental US dollars already flowing to two networks (eMarketer, 2026), the first budget decision of 2026 is portfolio design, not network discovery. Hold a network for one of three reasons and exit the rest: a shopper signal no other network can sell you, incrementality the network will let you validate independently, or media tied to a merchandising outcome you can name.

Make proof a condition of spend. The measurement clause belongs in the contract: methodology disclosure aligned to IAB/MRC definitions before the first dollar moves, lookback windows and halo-revenue rules fixed in writing per network, independent incrementality validation on a test-and-control cadence, and spend tiers tied to disclosure tiers. Networks respond to conditioned dollars faster than to industry working groups, because disclosure becomes the price of the budget.

One budget, one owner, one scorecard. Retail media money still lives in several places at once: brand marketing holds it at 71% of organizations, ecommerce at 54%, product marketing at 46%, while shopper marketing's share fell from 37% in 2021 to 11% (Dentsu, 2025; survey of 100 US retailer and brand decision-makers, fielded August 2024). The result is duplicated buys and one channel measured five different ways. The fix is a single commerce center of excellence, one funding source, one cross-network scorecard built on validated incrementality, and joint business plans renegotiated to credit self-serve dollars at par.

Get agent-ready before it is urgent. At 1.5% of US ecommerce, agents leave 2026 plans mostly intact; at nearly four times annual growth, they will shape 2027 plans (eMarketer, 2026). The near-term work is concrete: structured, attribute-rich product feeds that agents can parse, a tracking posture on UCP and ACP as retail partners switch them on, early tests of sponsored agent surfaces while pricing is immature, and a measurement plan for discovery that never appears in a click path.

Retail media in 2026 is a structural channel with immature proof, and both facts should shape the budget at once. The scale argues for presence. The proof gap argues for conditions.

For independent agencies and the brands they buy for, the leverage is procedural. Scale sits with the top two networks and will keep sitting there, so a buyer's advantage comes from the things a buyer controls: the shape of the portfolio, the contract language, the validation cadence, and the scorecard. All four levers are available at any spend level.

The practical posture for the year follows from that. Hold fewer networks and instrument them harder. Write disclosure and independent validation into the terms before the first dollar moves. Measure every network, retail or otherwise, against the same validated-incrementality standard. Run small, early tests on agent surfaces while they are cheap. The channel grew up in 2025; the operating model that buys it should do the same in 2026.