Ad tech's AI moat isn't better AI — it's exclusive data

Where does durable AI value come from in ad-delivery infrastructure? Not from the reporting copilots, auto-campaign builders, and bid optimizers every platform now ships — real gains, available to every buyer on identical terms, with Meta's Advantage+ alone at a $75B run rate #46. Durable value comes from invention: a proprietary decision engine compounding on data no rival can access. This report applies that investment lens across the ad-serving landscape; two multi-year engine builds show it in miniature.

AppLovin's AXON learns from closed-loop telemetry inside a network AppLovin owns end to end: $5.5B revenue, $3.3B net income, ~65% net margin #61 #104. The Trade Desk built an equally genuine eight-year engine, Kokai — on open supply and an open identity standard rivals can also use — and in March 2026 Morningstar stripped its moat rating to "none" #101. Same investment category, opposite outcomes. Engine plus exclusive data is a moat; engine alone is not.

The layer the money is rushing toward — agent-to-agent orchestration rails — is the one to be most skeptical of: two rival open standards are already commoditizing it the way Prebid.js commoditized header bidding #87–#92, and the protocols' own specifications contain no negotiation step, just faster discovery on the same auction #98. Meanwhile a US court has ruled the dominant ad-serving stack an illegal monopoly, with remedies pending on two continents #71 #73.

The diagnostic for every ad-tech AI initiative: is it automating what the industry already does, or inventing what it couldn't — and does the intelligence run on data competitors can't buy, or are you just paving someone's highway?

Full argument, buy-side and sell-side decision guidance, caveats, and the 112-source evidence log accompany the article.