Rebuilding the CTV Ad Server: Momentiv's Sandro Catanzaro

Full analysis

Your draft

Sandro Catanzaro is pitching Momentiv, a CTV ad server built from scratch in 2025, and the interesting claim isn't the 18-hour prototype. It's that the switching-cost moat protecting publisher-side ad servers just got a lot thinner.

Here's what's actually being decided, and it's not really Momentiv's fate. It's whether the thing that has protected CTV ad server incumbents for a decade, the terror of a full "brain transplant" migration, still holds. Catanzaro says run two ad servers at once, watch the revenue A/B, then decide. That's easy for a publisher to undo. Try it, and if the numbers lie, walk away. When a decision is that easy to undo, the whole calculus changes, and the incumbents know it.

The Market Analyst. Momentiv is one data point, but it points at a category. Every dollar of agentic-buying investment has gone to the buy side, because that's where the money and the excess demand sit. Catanzaro is building the sell-side mirror image, and he's right that almost nobody else is. That gap is the whole thesis. For an operator, the read is this: if AI-assisted coding really does drop the cost of building an ad server from years to months, then the moats around FreeWheel, Google Ad Manager, and SpringServe were never about technology. They were about migration fear. Remove the fear and you've repriced the category. In plain terms: the reason nobody switched ad servers wasn't that the alternatives were bad, it's that switching could get you fired.

The Skeptic. What has to be true for this to work? The 30% revenue lift has to survive contact with reality, and right now it's a self-reported number from a pre-revenue startup whose podcast hosts are also its investors. No sample size, no duration, no publisher mix. Treat it as a hope, not a result. And Ari Paparo has spent years explaining why new ad-server businesses die: the incumbent integrations, the certification treadmill, the SSAI plumbing, the ad-pod logic nobody documents. The parallel-server pitch is clever, but running two ad servers isn't free either. Someone has to reconcile the logs, split the traffic honestly, and trust that Momentiv isn't cherry-picking the easy impressions in the A/B. That might be true. Untested. And will remain so until a named publisher says it out loud.

The Operator. Tuesday morning, the ad-ops lead has to actually wire this up. Parallel serving in CTV means duplicating your inventory forecast, your frequency caps, and your priority tiers across two systems that don't share state. Frequency capping breaks first: a viewer served by both servers sees the same spot twice, and the buyer complains. Then the direct-sold guarantees get messy, because your incumbent still thinks it owns 100% of the avails while Momentiv is skimming a test slice. The scene-level contextual pitch, champagne at minute 96, ad break at minute 98, is genuinely clever because the ad server already holds that data. But it only works if the content metadata is clean and the pod timing is deterministic, and in CTV neither is guaranteed across every app and device.

The Customer / End User. Two customers here, and they want opposite things. The publisher wants more money with no career risk, and the parallel-server model is aimed straight at that. Good. But the buyer, the agency or DSP, is the one Zawadzki flagged: he sees Momentiv as much a buyer tool as a publisher one, a way to route holding-company demand as close to supply as possible. That's the tension nobody on the podcast fully owned. A "publisher-only" ad server that lets buyer agents specify targeting in plain English and matches inventory natively is doing buy-side work. The moment a holding company leans on it to get "as close to the supply as possible," the single-constituency loyalty Catanzaro borrowed from Jeff Green starts to wobble.

The CFO. The thin-fee model is the right alignment and the wrong margin question. A publisher-only fee tied to publisher revenue is honest, but it also means Momentiv eats a brutal sales cycle, one premium publisher at a time, with no buy-side network effect to pull it along. The Trade Desk's loyalty-to-buyers model works because buyers aggregate; there are a few hundred that matter and they bring scale. Publishers don't aggregate the same way, and CTV publishers especially guard their stacks. The years of integration and certification burn before the 30% lift, if real, compounds into reference accounts. AI made the code cheap. It did nothing for the go-to-market, which Catanzaro admits is now the bottleneck.

Where the council splits

Two disagreements matter. First, the Market Analyst thinks the moat was always migration fear and AI just dissolved it; the Skeptic and Operator think the moat was the plumbing, the integrations and the ad-pod logic, and that AI writes code but doesn't write certifications. Both can't be right, and the answer decides whether incumbents lose sleep.

Second, Catanzaro insists Momentiv is publisher-only by design, while Zawadzki openly describes it as a buyer tool. If the buy side ends up steering it, the "loyal to one side" pitch that legitimizes the whole thing collapses into just another intermediary, the opposite of the supply-chain flattening Catanzaro forecasts.

What it hinges on

Three beliefs. One, that AI-assisted development genuinely resets the build cost of an ad server, which looks directionally true. Two, that the 30% lift is real at scale and not a demo artifact, which is completely unproven. Three, that parallel serving actually removes the switching risk without creating a new operational mess, which the Operator has real doubts about. The council leans that the strategic read is right, publisher-side ad serving is contestable again, while the specific company is a coin flip. For an operator, the move isn't to bet on Momentiv. It's to call your incumbent ad server rep and ask why you can't run a parallel revenue test yourself, and watch how fast they change the subject.

The Prediction

Prediction: At least one major CTV ad server incumbent (Google Ad Manager, FreeWheel/Comcast, or Magnite's SpringServe) will publicly announce native AI/agentic optimization or plain-English targeting features for its publisher ad server by the 2027 upfront season (May 2027).

Confidence: Medium. The competitive pressure is real, but incumbent roadmaps move slowly and the trigger here is a startup, not a customer revolt yet.

Why: Catanzaro's whole pitch names the exact capabilities incumbents lack: native identity, forecasting, agentic buying support, and scene-level context. Buy-side agentic tooling is already everywhere, so publishers will start asking their existing sell-side vendors why they can't do the same, and the parallel-server model gives sophisticated publishers a cheap way to make the comparison visible. Incumbents defend share by matching features on a slide well before the product ships, especially when the threat is framed as revenue lift. The opposite outcome, total silence from all three through an entire upfront cycle while "agentic" is the loudest word in the category, would require them to ignore a marketing layup, which incumbents under churn pressure rarely do.

Revisit by 2027-05-31: We're right if Google, FreeWheel, or SpringServe announces a named agentic or plain-English targeting capability for its publisher-side CTV ad server by the 2027 upfronts. We're wrong if all three go the full period with no such announcement.

One caveat worth stating: a feature announcement is not a feature shipped. Watch whether any of it reaches general availability, or whether it stays a keynote demo while the real product sits in private preview.