Thrad launches first supply-side platform for LLM ad inventory

Full analysis

A seed-stage startup called Thrad just declared itself the world's first supply-side platform for large language models. In plain terms: technology that helps publishers sell ad space, except the "publishers" are ChatGPT and a long tail of chat apps built on top of it, and the "ad space" is a sponsored suggestion dropped into your conversation with a bot. CEO Andrea Tortella's pitch to AI developers is that letting Thrad place ads will pay for the expensive compute those chats burn. Co-founder Marco Visentin is on the masthead; Brian O'Kelley, the AppNexus founder, shows up in the entity list.

This is easy to undo for everyone involved. No budget is committed, no standard exists, and no walled garden has said yes to third-party ad serving inside chat. What's actually being decided is not whether Thrad wins. It's whether an independent ad-tech layer can wedge itself into conversational AI before OpenAI, Google, and Meta build it in-house. The deadline is set by the big LLM owners' own monetization roadmaps, not by Thrad.

The Market Analyst No public ad-tech name moves on this today. Zero spend, zero measurement, zero signed walled-garden access. For a general reader: nobody is actually buying these ads yet, and the biggest platform named, OpenAI, hasn't agreed to anything. The real question for operators is ownership of the chat ad auction in 2027. OpenAI will almost certainly ship native sponsorship before it shares revenue with a seed-stage SSP. Whether a Magnite or PubMatic buys into this space in the next 18 months is the signal that institutional money thinks the surface is real. That purchase, not Thrad's press release, tells you something.

The Skeptic Three things must all be true at once for this to work, and none is likely. OpenAI has to accept a revenue split with an outside SSP instead of building its own. Users have to tolerate sponsored suggestions without abandoning the bot. And advertisers have to buy inventory with no deterministic targeting, in a market that is screaming for more signal. Tortella's own framing, that "ads to agents aren't advertising," is the kind of line you use when the category doesn't exist yet. Brian O'Kelley's name buys credibility optics. It does not buy a signed publisher contract. And Adotat is not WARC.

The Operator You need both sides of a marketplace before either side is worth anything, and Thrad has neither. On the buy side, trading desks have no approved deal IDs, no inventory taxonomy, no brand-safety signal from an accredited source. On the sell side, "AI walled gardens" in the deck may mean nothing contractually, because ChatGPT has committed to no third-party monetization partner. The thing that breaks first is publisher recruitment. Without 50-plus real AI apps showing genuine daily users, buyers won't touch it, and the inventory story falls apart. The clean "SSP for LLMs" tagline hides a cold-start problem the tagline doesn't solve.

The Customer / End User Two customers here, and both have a reason to balk. The advertiser is being asked to pay for placement next to a machine's answer, with no way to know who saw it or whether it worked. The person using the chatbot is the other customer, and a sponsored suggestion inside a conversation they trust is exactly the thing that erodes the trust. Search ads work because people expect commercial intent in a search box. A chatbot feels like a confidant. Slip a paid answer into that and you risk the product, which is why OpenAI will move slowly and carefully here, not hand the keys to a startup.

Where they split

The genuine disagreement is about timing and ownership, not whether chat becomes an ad surface. The Market Analyst and the Strategist both think the surface is probably real eventually; they just think the walled gardens eat it. The Skeptic thinks the business model itself is unproven and the hype is borrowed from last year's GEO and AEO noise. The second split is the long-tail bet: the Operator sees thousands of token-burning GPT-wrapper apps with no revenue as a real, underserved problem; the Skeptic sees a pile of apps with no audience, which is not inventory anyone pays for.

What it hinges on

One belief does the work: will OpenAI and its peers rent out their chat surface to an independent SSP, or build native sponsorship themselves? Everything else follows from that. Search history says platforms with this much query intent verticalize. Google did not outsource search ads. Meta did not outsource feed ads. The incumbent SSPs with existing agency relationships, Magnite and Index Exchange, are far better placed to move on a real chat surface than a seed-stage shop. The council leans hard: the surface may be real, Thrad is unlikely to own it, and a signed, named revenue-share deal with any top-five LLM plus a brand-safety standard from an accredited measurement source are the two things that would change that read. Absent both, this is a story, not a market.

Prediction: No top-five LLM by usage (OpenAI's ChatGPT, Google Gemini, Anthropic Claude, Meta AI, Microsoft Copilot) will sign a public revenue-share deal to let Thrad serve third-party ads inside its chat surface before the end of Q2 2027.

Confidence: High Platforms with this much query intent build ad sales in-house.

Why: OpenAI has made zero public commitment to any third-party monetization partner, and the one named platform in Thrad's pitch is exactly the one least likely to share a surface it can monetize itself. The pattern is consistent across every high-intent surface that came before: Google kept search ads, Meta kept feed ads, Amazon kept its on-site ads, because the auction on a surface you own is the most valuable thing you control, and renting it to a startup gives away margin and user-trust decisions you'd rather keep. OpenAI's own roadmap almost certainly reaches native sponsorship before revenue-share with a seed-stage SSP. The opposite outcome, a major LLM handing ad serving to Thrad, would require the platform to concede both economics and the brand-safety risk of a stranger's ads inside its most trusted product, which is the slow, careful decision no platform rushes.

Revisit by 2027-07-15: We're right if no top-five LLM has announced a revenue-share ad deal with Thrad by the end of Q2 2027. We're wrong if any of ChatGPT, Gemini, Claude, Meta AI, or Copilot publicly signs one.

The interesting evidence won't come from Thrad at all. It'll come from whether Magnite or Index Exchange spends real money in this space. If the incumbents with agency relationships move, the surface is real. If they don't, the "SSP for LLMs" line stays a deck.