Every agency-side media buyer has watched the same negotiation shift shape over the past two years: a client renewal that used to be a straightforward media plan now arrives bundled with a holding company’s AI stack, a data clean room, and \u2014 buried in the fine print \u2014 a chunk of guaranteed spend that never touches the open web.<\/p>\n
That structural bargain is the subject of AdExchanger’s Tuesday roundup<\/a>, which describes holding companies increasingly absorbing the cost of AI infrastructure \u2014 LLM licensing, compute, the internal copilots agencies now pitch as differentiators \u2014 in exchange for guaranteed client spend directed at their own principal media inventory. The roundup doesn’t name every account running this playbook, and the underlying agency-client contracts remain confidential, as most do. But the mechanism it describes tracks with what the three largest networks have said publicly about their own AI ambitions.<\/p>\n WPP has described Open Intelligence<\/a>, the marketing operating system it launched in mid-2025, as core infrastructure meant to sit across every client account rather than a standalone product sold separately \u2014 the kind of framing that only makes financial sense if the build cost is recovered through media commitments rather than software fees. Omnicom has similarly positioned Omni<\/a>, its AI-driven platform now absorbing the retail and commerce data from its Flywheel acquisition, as the connective layer clients access through their media relationship with the agency, not as a licensed tool sold independently. Publicis has made the same case for CoreAI<\/a>, the platform sitting on top of Epsilon’s identity and data assets, describing it in investor materials as an engine for retaining and growing client media budgets rather than a revenue line of its own.<\/p>\n None of the three networks has publicly disclosed the specific contractual mechanics \u2014 how much guaranteed spend gets tied to an AI licensing agreement, or over what term \u2014 and that gap is worth naming plainly rather than papering over. What’s documented is the strategic logic: all three have told investors and trade press that AI infrastructure is meant to be inseparable from the media relationship, which is a different claim than “we’re giving clients a rebate,” but one that produces the same downstream effect described in the AdExchanger roundup.<\/p>\n Principal-based media buying isn’t new \u2014 holding companies have owned and resold inventory positions for years, often controversially, as a hedge against margin compression in traditional commission models. What’s changed, per the roundup, is the bundling logic. Agencies are subsidizing the AI tooling clients want \u2014 forecasting models, creative generation, campaign optimization layers \u2014 and using that subsidy as leverage to steer guaranteed budget toward inventory the holding company itself controls or has pre-negotiated favorable terms on.<\/p>\n The pitch to a CMO is straightforward: cheaper or free access to AI infrastructure that would otherwise require a separate vendor contract, bundled with media execution the agency already handles. For the agency, it locks in guaranteed spend against inventory where margins are more predictable than open-exchange buys. For publishers who rely on programmatic guaranteed deals or direct-sold packages and aren’t inside one of those closed loops, the money increasingly routes elsewhere before it ever reaches an RFP.<\/p>\n The mechanism matters because guaranteed spend has historically been where independent publishers had the most negotiating room. Programmatic guaranteed and direct-sold deals carry premium CPMs precisely because they promise certainty \u2014 fixed impression volumes, brand-safe placements, first-look access. If a growing share of that guaranteed budget is pre-committed to principal inventory as part of an AI-for-access trade, publishers aren’t just competing for a shrinking pool of open-market dollars; they’re competing with less pricing power, because the buyer’s incentive structure was shaped upstream, in a contract they were never part of.<\/p>\n This isn’t quite the same story as walled gardens absorbing programmatic spend \u2014 a dynamic publishers have adjusted to for years by building direct relationships and first-party data products. The principal-inventory version is harder to route around because it isn’t platform-specific; it’s baked into the agency-client contract itself, before a media plan ever gets built.<\/p>\n For sophisticated media operators, the response isn’t panic \u2014 it’s an audit of where guaranteed dollars in their book are actually coming from, and how durable those relationships are if the agency’s incentive structure shifts further toward principal deals. Publishers with strong first-party data, retail media integrations, or commerce partnerships have alternative paths to guaranteed-adjacent revenue that don’t route through a holding company’s inventory desk. Publishers dependent primarily on programmatic guaranteed deals sourced through agency trading desks have less insulation, because that’s precisely the budget category most exposed to the bundling dynamic described above.<\/p>\n There’s also a timing question worth watching: how much of the AI-for-access trade gets formalized in multi-year contracts versus tested quarter to quarter. WPP, Omnicom, and Publicis have each framed their AI platforms as multi-year infrastructure investments rather than one-off tools, which suggests the incentive to tie those costs to long-term media commitments \u2014 not single campaigns \u2014 is structural rather than incidental. If that holds, the squeeze compounds: publishers lose access not just to this year’s guaranteed budget but to the renewal cycle that would have let them renegotiate.<\/p>\n Publishers watching this trend should be asking their agency contacts a blunter question than usual: not just what the CPM is, but whether the guaranteed line item they’re bidding for was ever genuinely in play \u2014 or whether it was functionally spoken for the moment the AI infrastructure deal got signed.<\/p>\n","protected":false},"excerpt":{"rendered":" Holding companies are bundling AI infrastructure subsidies with guaranteed spend on their own principal media \u2014 a trade that could wall off open-web publishers from guaranteed ad budgets.<\/p>\n","protected":false},"author":13,"featured_media":6514,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[1],"tags":[432,431,138],"class_list":["post-6515","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","tag-alternative-revenue","tag-commerce","tag-video"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/posts\/6515","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/users\/13"}],"replies":[{"embeddable":true,"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/comments?post=6515"}],"version-history":[{"count":0,"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/posts\/6515\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/media\/6514"}],"wp:attachment":[{"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/media?parent=6515"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/categories?post=6515"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/publir.com\/blog\/wp-json\/wp\/v2\/tags?post=6515"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}The Mechanics of a Closed Loop<\/h2>\n
What Gets Squeezed<\/h2>\n
Where the Leverage Actually Sits<\/h2>\n