By The official story is that OpenAI is partnering with Criteo and holding talks with The Trade Desk to bring advertising into ChatGPT. A thoughtful, measured approach to monetization. Nothing to worry about.
Here's what's actually happening: OpenAI is hiring a monetization infrastructure engineer to build its ads systems from the ground up, an engineering manager to lead the team, a product designer to define what advertising looks like inside its AI assistant, a senior manager to own ad revenue accounting, and a trust and safety hire dedicated specifically to the ads product. All in San Francisco. All full time. According to Digiday, compensation bands for these roles run as high as $385,000.
Companies paying $385k for an ads engineering manager are not planning to rent their stack forever.
What's being built
Right now, OpenAI has no advertiser relationships, no measurement history, and no agency integrations worth mentioning. Criteo and The Trade Desk have all three. That's why the partnerships exist: they're not a strategy, they're a bridge.
The actual strategy is visible in the job listings. OpenAI wants a demand-side tool for buyers, a decisioning layer for yield optimization, and measurement infrastructure that integrates with third-party modeling platforms. That's a description of a walled garden. Not a ChatGPT plugin for some DSP's API.
Adweek reported this week that OpenAI has hired David Dugan, a former Meta vice president who led its global clients and agencies division, as VP and head of global ads solutions. Dugan reports to COO Brad Lightcap. Holding companies including Omnicom, WPP, and Dentsu are reportedly lining up to test inventory. Early brand partners like Best Buy and AppLovin have already run pilots. Minimum commitment to get in: $200,000 per brand, per Adweek.
The product has a name now too. OpenAI is testing an ads manager, the same type of self-serve interface that Facebook and Google used to turn their ad businesses from bespoke deals into printing presses.
Why this was always inevitable
LUMA Partners CEO Terence Kawaja made the arithmetic plain at a recent Marketecture conference: LLM infrastructure costs are large enough that subscription revenue alone can't cover them. OpenAI is expected to burn approximately $15 billion in cash this year, up from $9 billion in 2025, according to Digiday. Around 910 million users interact with ChatGPT weekly. Roughly 95% of them don't pay.
That math has one answer.
The historical parallel everyone in this industry should be reaching for isn't Google circa 2003. It's Netflix and Walmart. Both companies announced advertising businesses while simultaneously leaning on outside infrastructure to move fast. Netflix launched with Microsoft's ad tech. Walmart stitched together DSP partnerships. Both spent the years that followed systematically pulling the stack in-house, because margin is on the infrastructure side, not the media side. The companies that helped them launch were well aware of what they were signing up for.
Criteo's situation here is particularly instructive. The company is facing a $75 million revenue headwind in 2026 after two major retail media clients reduced scope, and its stock was near 52-week lows earlier this year. Being named OpenAI's first ad tech partner is the most credible validation it has had for its pivot toward agentic commerce. The problem: the better Criteo performs inside ChatGPT, the stronger OpenAI's case becomes for doing it themselves.
The incentive structure here isn't subtle. OpenAI needs ad revenue urgently. Criteo and The Trade Desk need the distribution and the narrative. But OpenAI keeps 100% margin once it owns the stack, and it loses that the moment a DSP is in the middle taking its cut.
What comes next
The vendor relationships probably have 18 to 36 months of genuine utility. That's enough time for OpenAI to build measurement infrastructure, develop the buy-side tooling, and sign enough agency holding company deals to have direct relationships that don't require a middleman.
Meanwhile, the holding companies lining up for these pilots should read the situation clearly. Jeff Green at The Trade Desk has argued that AI agents are well-suited to navigating programmatic complexity, which is a reasonable take if you believe AI will route budgets through existing infrastructure. The competing view, which OpenAI's hiring patterns seem to endorse, is that the company intends to be the infrastructure.
There's a version of this that ends with OpenAI as a large enough media property that it can command direct relationships with every major advertiser, bypassing programmatic entirely for its premium inventory and running its own auction for the rest. That's what Google built. That's what Meta built. That's what Amazon is still building. The job listings from February and March suggest OpenAI is at least attempting the same sequence.
The interesting question isn't whether OpenAI will build its own ad stack. It clearly will. The interesting question is whether the $15 billion burn rate gives it enough runway to build it before the subscription business has to carry more of the weight than it currently can.
For the vendors currently helping it get to market: congratulations on the deal. Check the expiration date.
Zach El-Amin covers ad tech for The Daily Vibe.
