OpenAI announced a $1B annualized run rate for its ads programme in under 200 days (~6.5 months) after launch — announced August 31, 2026.
This basically means they earn $1B / 365 ≈ $2.74M in ad revenue per day. At a $50 CPM, that’s ≈55 million impressions monetized per day. If there’s one ad per conversation, that’s roughly 55 million conversations per day that carry a relevant ad.
Recently they also announced 1 billion weekly active users (same Aug 31 announcement — ads run on the free tier and Go tier, which make up the bulk of that base).
A year ago, Sam Altman mentioned 2.5 billion messages per day. If we expect similar growth, that’s at least 20% more today — ≈3 billion messages per day.
This means ≈1.8% fill rate, globally (55M monetized conversations ÷ ~3B daily messages). We have to assume the US converts at a much higher rate than other countries, and they’re aggressively adding countries this year. But how far can this go? And how fast?
10%? 25%? 50%? 100%? 500%?
When Facebook was growing, it increased ad load every year. So let’s start there: how much revenue can OpenAI generate with the same lever, even if usage doesn’t grow another inch.
$5B at a 10% average fill rate, globally.
$25B if 50% of global conversations carry one ad.
$250B if they show 5 ads in every conversation.
That’s not a forecast — it’s what falls out of simply turning the same dial Facebook turned, scaled off today’s $1B baseline.
Facebook itself gave us the framework for that dial. A 2016 analysis of Facebook’s earnings calls (Jan Dawson, Beyond Devices) breaks ad revenue into:
Users × Time spent × Ad load × Price per ad
And Facebook quantified ad load’s share directly: from 2013–2016, ad revenue grew 57–63% year-over-year, and its own CFO said ad load alone accounted for 10–30 percentage points of that. Strip it out, and growth would have been “only” 30–50% — still huge, just less.
On density: at Facebook’s saturation point in 2016, ads showed up roughly every 7 posts on desktop (~14% of feed) and every 10–12 posts on mobile (~8–9%). That’s the ceiling Facebook eventually hit — after which its CFO said ad load would “cease to be a significant driver” of growth.
Here’s the catch, though. Ad load is a percentage, and percentages need a denominator. Facebook’s denominator was enormous: the average user saw about 300 posts a day out of 1,500 potential in the feed — call it 2,100 posts a week. OpenAI’s denominator, per user, is nowhere close: at ~3B messages/day over a 1B-user base, and ~1.7 messages per conversation (per WebFX’s sample of shared conversations), that’s roughly 12.3 conversations — or about 21 messages — per user per week.
2,100 vs. 21. Facebook’s per-user content volume was roughly 100 times larger than ChatGPT’s is today. So it isn’t just that OpenAI’s ad load (1.8%) is far below Facebook’s saturated load (8-14%) — the entire surface it has to work with, per person, is two orders of magnitude smaller. Match Facebook’s load percentage exactly, and a ChatGPT user still generates a fraction of the ad opportunities a Facebook user did. The 500% figure only closes the load-rate gap; closing the volume gap needs people to use ChatGPT the way they used Facebook — often, and in longer back-and-forths — which is a behavior change, not a monetization setting.
But the volume gap isn’t just a fill-rate story — it’s a ceiling story. Back in 2013, when Facebook had that ~2,100-posts-a-week density, its global ARPU was around $1.35–$2.14 per quarter — call it $6-7 per user per year, of which roughly $6 was advertising. OpenAI today, at $1B in ads against 1B weekly active users, is sitting at about $1 per user per year. On paper, OpenAI extracts more revenue per message ($0.0009) than Facebook did per post ($0.000055) — roughly 16x. But that’s not OpenAI being better at monetizing; it’s what happens mechanically when you divide similar-order-of-magnitude revenue by a denominator that’s 100x smaller. Facebook’s real edge was never monetization skill — it was room. It had a vastly bigger canvas to keep adding ads to, one more post at a time. OpenAI doesn’t have that canvas. Its ceiling is capped by how few messages exist to interrupt in the first place, whatever the per-message price.
That’s the real open question sitting underneath all these numbers: does inserting an ad into a tool people use to get something done cost the same as inserting one into a feed people use to pass time? Facebook could raise ad load because scrolling past an ad barely interrupted anything. ChatGPT’s job is closer to finishing your sentence than filling your afternoon — and we don’t yet know what an ad costs it in trust. That’s the piece worth its own post.
Leave a Reply