Episode 181: Alex Kantrowitz on OpenAI and Big Tech's Latest Approach to Advertising

Full analysis

The episode is a survey shot of AI eating ad-tech from six directions at once: OpenAI staffing an ad team with Meta defectors and launching audience-matching, a fully AI-negotiated media deal at Vox, Cloudflare building a micropayment tollbooth for AI agents, Criteo drawing a private-equity buyout at a fat premium, and IAS swapping CEOs while it bleeds share to DoubleVerify. No single item is a bombshell. Together they sketch a clear direction: the buy side is getting automated and commoditized, the sell side and infrastructure layers are where control (and money) concentrate, and the AI labs are now credible ad-market entrants rather than distant threats.

What's actually being decided across the ecosystem: where value pools when both sides of the media transaction become agents. This is largely Type 1 (hard to reverse) for anyone building product roadmaps around it, but the individual news items are Type 2 (early, still optional to react to).

The Market Analyst — The Criteo bid is the loudest real signal here. A private-equity buyer paying a 50%+ premium for a company the market had left for dead tells you the "retargeting is dying" narrative overshot — there's a floor under businesses that own retail product feeds and a working identity graph. In plain terms: someone thinks Criteo's data is worth more than its stock price implied. Paparo's OpenAI-buys-Criteo thesis is the more interesting version: an AI lab needs a product catalog and conversion data to make agentic shopping work, and building that from scratch is slow. Meanwhile IAS getting a Slack alum as CEO signals the board wants a software operator, not an ad-industry lifer — an admission the ad-verification market has stopped growing.

The Skeptic — The load-bearing assumption in half these stories is that agents will actually transact at scale soon. One AI-negotiated Vox deal is a press release, not a paradigm. Cloudflare's stablecoin tollbooth assumes an "agentic web" where bots pay per API call — that world doesn't exist yet at volume, and publishers have watched infrastructure players promise them new revenue rails for fifteen years. Plainly: a lot of this is betting on a future that hasn't shown up. And the OpenAI "billion users" figure is unverified — the last confirmed number was 400 million weekly. Be careful which of these you build a 2027 budget around.

The Operator — The item that actually changes someone's Tuesday is the ADCP / publisher-agent shift. If the buy side collapses into a thin, automated layer that "doesn't really require a vendor," that's an extinction event for the mid-tier DSP and the managed-service trading desk. The people whose job is working a campaign through a UI should read that sentence twice. In plain English: the humans who manually place ad buys may be automating themselves out of a role. Second-order effect at 90 days: whoever controls the publisher agent controls the business logic — pricing, floors, packaging — which is why the sell side quietly likes this future. WBD building its whole ad stack on AWS Bedrock is the same story in a different suit: media companies renting their intelligence layer from a cloud giant that also sells ads.

The Customer / End User (advertisers & buyers) — Buyers should love AI-negotiated deals right up until they realize a sell-side-controlled agent is optimizing for the publisher's yield, not their outcome. Plainly: if the seller's robot runs the negotiation, guess who wins. The real question no one in the episode answers: who audits the agent? This is exactly where the transparency gap that already loses advertisers — no reporting on what ran or worked — gets worse, not better, when a black-box agent does the buying. Advertisers will demand independent verification of agent-to-agent deals, which ironically is a lifeline for measurement players if they can pivot fast enough.

The CFO — Follow the margin. Cloudflare's play is about skimming payment economics away from Visa/Mastercard on agent traffic — a toll on volume that doesn't exist yet but could compound if it does. Criteo at a ~$1.16B implied cap is cheap for the data inside it; that's why PE circled. The uncomfortable read for verification and single-purpose ad-tech: if your total addressable market isn't growing (IAS) and a bigger platform can bundle your function for free, your standalone value is a countdown. The cash question every operator should ask: is my product a feature that a lab or cloud giant absorbs, or a business they have to buy?

Sharpest tensions:

1. Is the buy side dying or just changing shape? The Operator and Market Analyst say value migrates to whoever owns the publisher agent and the data. The Skeptic says one Vox deal proves nothing and human buyers persist for years. This is the real fork.
2. Are AI labs buyers or builders of ad-tech? The bull case (Paparo) says OpenAI acquires its way into commerce data. The Skeptic says labs with anti-ad postures (Anthropic) and unproven ad revenue don't need to spend $3B yet.
3. Does infrastructure or application capture the money? Cloudflare and AWS suggest the picks-and-shovels layer wins; the ADCP story suggests the agent-orchestration layer does.

What it hinges on: whether agent-to-agent transactions move from demo to routine within roughly a year, and whether the sell side (not the buy side) ends up holding the business logic. If both are true, mid-market DSPs, trading desks, and single-function verification vendors are the losers, while publisher-agent platforms, retail-data owners, and cloud/payment infrastructure are the winners. The council leans toward "sell-side and infrastructure concentrate power; the automated buy side commoditizes."

The most concrete, checkable claim in the episode is the Criteo takeout — a live process with a named premium and a jumped stock.

Prediction: Criteo will announce a definitive agreement to be acquired (by a private-equity buyer or a strategic such as an AI/commerce platform) at or above the reported >50% premium before its Q4 2026 earnings report.

Confidence: Medium — Live PE process, stock already jumped ~20%, boards rarely walk back a public premium bid.

Why: Two PE firms have reportedly submitted a bid at a 50%+ premium and the stock re-rated ~20% on the news, which means the market now believes a deal is more likely than not — that price move is hard to unwind without a completed transaction. Criteo owns retail product feeds and an identity graph that are strategically valuable to both financial buyers (cheap cash-flow at a $1.16B cap) and strategics building agentic commerce, so there's competitive tension supporting the price. The opposite outcome — Criteo stays independent — would require the board to reject an above-market premium with no higher bidder emerging, which usually only happens when the offer is below intrinsic value, not at a fat premium.

Revisit by 2026-11-15: We're right if Criteo announces a definitive acquisition agreement at ≥50% premium to its pre-rumor price. We're wrong if Criteo remains independent with no signed deal, or formally rejects all bids and stays public.