OpenAI Ad Business at $1B Run Rate, Eyes $25B by Year-End

Full analysis

OpenAI says its ad business inside ChatGPT is running at a $1 billion annualized rate seven months after launch, and it projects $25 billion in actual ad revenue by the end of 2025, then $100 billion by 2030. The Digiday reporters covering it name five things standing in the way: measurement, privacy terms, inventory scarcity, small-business adoption, and ad plumbing. Criteo and Kargo are named as the plumbing partners; a Shopify product-feed integration is the most promising piece so far.

Here's the frame for an ad-tech operator. This isn't a decision you make. It's a bet you're being asked to place with budget, roadmap, and headcount. How hard is it to undo? Very easy. Spinning up a test line item in ChatGPT and pulling it back costs you a quarter and some frustration, nothing structural. What's actually being decided is not "does OpenAI reach $25B." It's "is high-intent commercial search moving off the query-and-click page fast enough that I need to reroute demand and measurement now." What sets the deadline is the 2026 budget planning cycle, when trading desks decide whether OpenAI gets a real line or a footnote.

One thing before the council. That $25 billion by end of 2025 number is already impossible to hit as I write this on the last day of September 2026. Either the figure is stale coverage of an old projection, or it always meant a run rate exiting 2025. Read it as ambition, not a forecast anyone should underwrite.

The Skeptic

A $1 billion annualized run rate from a standing start is the most flattering math in the business. It means one good month times twelve. It is not $1 billion collected. The $25 billion projection reads like a fundraising slide with a press release stapled to it. Look at the five hurdles Digiday lists. Those aren't friction. They are the business. Measurement doesn't get solved in a quarter. Privacy terms that scare an advertiser's lawyers don't get fixed by a partnership logo. And inventory scarcity means buyers committed money they cannot spend, which is a refund conversation, not a growth story. For the non-specialist: OpenAI sold ad space it doesn't actually have enough of yet.

The Market Analyst

The trade here isn't OpenAI, which is private. It's Criteo, which just got named as plumbing inside the hottest inventory story of the year. That's a real near-term bid for the stock. But the same thing that lifts Criteo can kill it. If OpenAI builds its own bidder once volume justifies it, Criteo is a bridge that gets burned after the crossing. The bigger read is what this does to Google and Meta. If conversational AI grabs even a tenth of high-intent commercial queries, the damage won't show up as a clean Google revenue miss. It shows up slowly, as the moment before someone types a search fades. In plain terms: the money doesn't leave Google in one quarter, it leaks over years, and the crowd will under-price that leak because admitting it means admitting their Google position is exposed.

The Operator

Tuesday morning, a trading desk lead spins up an OpenAI line and immediately hits the wall everyone will hit: there aren't enough impressions to spend the committed budget. Ninety days of chasing inventory that doesn't exist at scale, then a frustrated write-up. The Criteo and Kargo hookups give buyers a path in, but a path to an empty shelf is still an empty shelf. The Shopify feed is the one bright spot, because for a performance marketer selling products, the conversion can close without a custom measurement build. Everywhere else, the pain lands on measurement and finance teams at once. No third-party verification. No standard viewability signal. No clean way to tie a sale back to the ad. For the non-specialist: you can buy the ad, but you can't yet prove it worked.

The Customer / End User (the advertiser)

Nobody is actually asking for this yet. Advertisers are curious, and curiosity is not budget. The Digiday reporting says the terms and conditions are too rigid and load too much liability onto the buyer. That's the quiet killer. A brand's legal team will stall a signature for months over indemnity language, and no amount of "conversational commerce is the future" moves them. The buyers who committed spend did it to be early, not because the ROI penciled out. For the non-specialist: brands will test this with pocket change while their lawyers argue over who's on the hook if the AI says something wrong.

The CFO

Run rate is not revenue, and a CFO knows the difference cold. The question isn't the $25 billion headline. It's what realized number, actually collected, justifies rerouting spend or engineering time. Right now that number is small. The cost of a test is cheap, so a small experiment is defensible. Building bespoke measurement pipes into OpenAI before there's a standard is not, because you'll rebuild them the moment OpenAI ships its own. The disciplined move is to spend enough to learn and not a dollar more until inventory and measurement are real.

The tensions

Two disagreements matter. The Market Analyst sees Criteo as a winner; the Strategist and Operator both see Criteo as a temporary bridge OpenAI disintermediates once it has scale. Both can't be right for long. The near-term bid is real, the durable position is not.

The second is about speed. The Strategist's read is that search intent is structurally rerouting and the migration is underway. The Skeptic and the Customer say the five hurdles are the business, and none of them clear on a 2026 timeline. This is the real fork. If you believe the migration is fast, you build now. If you believe the hurdles are the business, you test and wait.

Synthesis

The whole thing hinges on two beliefs. First, whether high-intent commercial queries are actually moving off search fast enough to matter in the next 18 months. Second, whether OpenAI fixes measurement and privacy terms before advertisers lose patience. The council leans skeptical on the timeline and constructive on the direction. The intent shift is real and probably permanent. The pace is being oversold, and the $25 billion number is math, not money.

What to de-risk before committing anything real: run a small Shopify-feed test where attribution can actually close, and treat everything else as watching. Don't build custom measurement into a surface that has no standard yet. Don't move a Meta SMB budget on the strength of one integration. And if you're touching Criteo as a play on this, know you're buying near-term plumbing revenue with a disintermediation clock running underneath it.

Prediction: Criteo's full-year 2026 revenue, reported in February 2027, will not show a distinct, disclosed OpenAI-driven revenue contribution material enough to lift its total above its 2025 revenue.

Confidence: Medium. Inventory scarcity caps deployable spend faster than the plumbing can scale it.

Why: The Digiday reporting says buyers have already committed spend but can't find enough placements to deploy it, which means the ceiling on 2026 revenue through this channel is set by OpenAI's ad inventory, not by Criteo's ability to route demand into it. Criteo's core retargeting business has been roughly flat for years, so a genuine OpenAI contribution large enough to move the full-year total would have to overcome both that flat base and the scarcity ceiling in a single year. The opposite outcome, a visible OpenAI lift to Criteo's total, would require OpenAI to solve inventory, measurement, and privacy terms fast enough for real budgets to clear through Criteo's pipes in 2026, and the five hurdles named make that a 2027-plus story. The plumbing deal is real; the revenue timing is being pulled forward.

Revisit by 2027-03-01: We're right if Criteo's reported full-year 2026 revenue is flat or lower than 2025 and no OpenAI-attributable revenue line materially changes that. We're wrong if Criteo reports 2026 revenue above 2025 with OpenAI-driven demand cited as a contributor.