Erik Hovanec on Ad Tech M&A

Full analysis

Erik Hovanec, the former Chief Strategy Officer of Magnite, sat down with Corey Ferengul and Joe Zawadzki to explain how header bidding nearly killed Rubicon Project and forced CEO Michael Barrett into an eight-acquisition rebuild that became Magnite. Along the way he laid out a theory of ad tech that cuts against the industry's actual behavior: middlemen who charge fat, hidden margins are hanging a target on their own backs, and AI does not change that math.

This is light on news and heavy on how deals actually get done. The implication for operators is a warning about where margin lives and how long it survives.

How hard is this to undo? Nothing here is a decision you make this week. It is a lens on positioning. The reversible part, pricing posture and how you present a business to a buyer, you can adjust anytime. The hard-to-undo part, a reputation with your own customers as extractive, is exactly what Hovanec says you cannot fix once a buyer's diligence team starts dialing them.

What's actually being decided: whether your take rate is a moat or a liability, and whether AI makes your product defensible or just cheaper to copy.

The Market Analyst. Hovanec is talking his book on SpotX and SpringServe, and the fact check is right to flag it. Magnite's stock fell hard after that 2021 CTV peak, and the company carried real debt to get there. "Bargain" is a retrospective told by the guy who negotiated it. But the structural point survives the self-interest: the SpotX call option on SpringServe meant whoever won SpotX got both, and that bundle logic is live again right now. CTV ad-serving, curation, and identity are consolidating at once. Any operator holding one of those layers should assume a buyer is pricing the adjacent layer into your deal, with or without you in the room.

In plain terms: when you sell one piece of the plumbing, the buyer is often really after the piece next to it.

The Skeptic. The clean story is that transparent, low take rates keep competitors out and opaque 20 to 30 percent margins invite them in. Is that actually true? The whole programmatic complex has run on hidden fees for fifteen years and the incumbents are still here. Transparency has not displaced them. What Hovanec is really describing is an acquirer's preference, not a market law. A buyer folding you into public-company guidance hates a margin that could collapse on contact. That is a diligence problem, not a competitive one. Plenty of extractive businesses thrive precisely because nobody can see the fee. The moat is the opacity, right up until someone can measure it.

In plain terms: hidden fees are not punished by the market, they are punished by the person trying to buy you.

The Engineer. The runtime claim is the part operators should actually stress-test. Tens of trillions of ad requests a day, 5 to 7 milliseconds to respond. That number is unsourced and probably a reach on the "tens of trillions" scale, but the order of magnitude is the point. You cannot run a large AI model inside a 7-millisecond bid window at that volume and afford it. The economics break before the latency does. This is why "AI-native bidding" pitches tend to work in a notebook and die in production. The realistic path is AI offline, setting the parameters, training the model, writing the bid factors, with something small and cheap firing at request time. Anyone promising live model inference on every bid is selling you a demo.

In plain terms: the smart AI runs the night before. The thing answering each ad request in milliseconds has to stay dumb and cheap.

The Customer / End User. Here the customer is the publisher or advertiser on the other side of the take rate, and Hovanec's advice reads very differently from their seat. He tells founders that a buyer will call your customers before closing, and being seen as "too clever by half" is the kiss of death. Flip that around: your customers already know what you extract from them, even when your rate card does not say it. The curation layers and private-marketplace operators booking fat spreads today are building exactly the customer resentment that shows up in someone else's diligence call in three years. The buyer is not confirming a secret. They are hearing what the market already whispered.

The CFO. The uncomfortable line for anyone running a P&L on programmatic complexity: AI lowers your build cost and does nothing for your margin. If a competitor can rebuild your product in three months, your pricing power has a clock on it. That means a take rate that looks like healthy gross margin today is really a melting asset if the only thing protecting it is that the product was hard to build. The businesses that survive this are the ones whose moat is scale or a direct relationship, not a fee nobody has gotten around to competing away yet. Model the commoditization into your forecast now, because a buyer absolutely will.

Where the council splits:

The Skeptic and the Market Analyst disagree on whether transparency is actually protective. The Analyst takes Hovanec's moat argument at face value: lean pricing keeps entrants out. The Skeptic says the market has rewarded opacity for a decade and the only party that punishes fat margins is a would-be acquirer. That gap matters, because it changes whether you clean up your take rate for the market or only when you decide to sell.

The Engineer and any AI-native founder split on where AI lives in the stack. If inference has to happen at request time, the runtime cost ceiling is real and most AI-bidding pitches are fiction. If the intelligence can move offline into parameter-setting, the ceiling mostly disappears and the "AI changes nothing" line is too strong.

What this hinges on: whether hidden margin is a moat or a countdown. Hovanec's own evidence points to countdown. His whole eight-acquisition story starts with header bidding exposing Rubicon's pricing power, which is the exact mechanism he warns founders about, a technology that made a hidden advantage visible and competed it away. The council leans toward the Skeptic's framing with the Engineer's caveat: your fee is safe only as long as nobody can see it or rebuild around it, and AI shortens both clocks.

What to de-risk before you act on any of this: measure your own effective take rate the way a buyer's diligence team would, and know what your top ten customers would say about it on a reference call. If either number embarrasses you, you have a problem that no roadmap fixes.

Prediction: At least one CTV infrastructure company among the independent ad-serving, curation, and identity vendors will be acquired by a larger sell-side or measurement platform by the end of Q2 2027 earnings season, reported in August 2027.

Confidence: Medium. The consolidation pattern is live, but the timing of any single deal is hard to pin.

Why: Hovanec's SpotX-SpringServe account describes the exact mechanism now repeating: a buyer chasing one CTV layer prices in the adjacent layer and takes both, because an intermediary sitting between giants like Disney and Netflix cannot survive without scale. The streaming ad stack today has three layers consolidating at once, ad-serving, curation, and identity, and several of them are still held by sub-scale independents whose margins depend on complexity that larger platforms are actively trying to absorb. The less likely outcome is that these layers stay independent, and that runs against both the scale logic Hovanec lays out and the public sell-side platforms' demonstrated appetite for buying their way into CTV infrastructure rather than building it.

Revisit by 2027-08-31: We're right if a larger sell-side, ad-serving, or measurement platform announces the acquisition of an independent CTV ad-serving, curation, or identity vendor before the end of Q2 2027 earnings season. We're wrong if no such deal is announced in that window.