skip to content
The Weighted Average

AI Economics for Operators

ChatGPT Ads Earn $1 a User; Meta Earns $68

OpenAI's ad business hit a $1B run rate in under 200 days — about $1 per weekly user a year, against Meta's $67.56 per person.

By Stephen Van Tran 10 min read Filed under View more posts about  AI Economics for Operators , View more posts about  Policy & Geopolitics
Large illuminated billboards on a building at dusk
Large illuminated billboards on a building at dusk. Photograph by John Cameron

OpenAI said on Monday that ChatGPT Ads reached a $1 billion annualized revenue run rate in fewer than 200 days, a milestone CNBC reported alongside OpenAI’s $852 billion valuation and its expansion of self-serve buying across India, Europe, the Middle East, and North Africa. Divide that run rate by the more than one billion weekly active users OpenAI says ChatGPT now reaches, and the number that matters appears: roughly $1.00 of advertising revenue per weekly user per year. Meta, monetizing an audience of comparable scale, earns $67.56.

Hours before that announcement, the European Commission designated ChatGPT a Very Large Online Search Engine. The same week that OpenAI proved its ad business works, Brussels attached the compliance regime that governs how ad businesses of that size behave. Both facts land on the same operator question: whether the conversational surface is a real acquisition channel yet, and what it will cost to buy on it once the rules arrive.

A dollar a head, and a regulator at the door

The gap is not a rounding error; it is two orders of adolescence. Meta’s second-quarter results filed with the SEC report $60.801 billion of revenue against 3.60 billion daily active people, which works out to $16.89 per person for the quarter and $67.56 annualized. OpenAI’s ad run rate spread across “more than 1 billion weekly active users” yields about a dollar. Meta extracts roughly 68 times more per human it touches, on an ad machine two decades older, with impressions up 14% and average price per ad up 12% year over year in the same filing.

Meta earns 68 times more per user than ChatGPT's ad business

Annualized revenue per user: Meta per daily person, ChatGPT ads per weekly user

MetaChatGPT ads$0$10$20$30$40$50$60$67.56$1.00
MetaChatGPT ads$0$20$40$60$67.56$1.00
OpenAI via CNBC; Meta Q2 2026 SEC results · Sep 2026

That ratio is the honest frame for every media plan being drafted this quarter. ChatGPT is not an underpriced Meta; it is a channel at the very beginning of its yield curve, with monetization confined to the free and Go tiers while the paid work tiers stay clean — the structural boundary this paper described when ChatGPT Ads first turned intent into a market. The upside case is precisely that the dollar has 67 dollars of headroom above it. The risk case is that conversational intent may never carry feed-grade frequency, because a user who asks one question and leaves sees one opportunity, not forty.

There is a second asymmetry hidden in the denominators, and it flatters OpenAI rather than Meta. Meta’s $67.56 is computed against daily people; OpenAI’s dollar is computed against weekly users, a looser definition that counts anyone who showed up once in seven days. Normalize toward daily engagement and the gap widens rather than narrows, because a weekly cohort contains far more low-frequency visitors than a daily one. The comparison is generous to the newer business and the newer business still loses by a factor of 68.

Regulation arrived on the same day. The Commission’s designation of ChatGPT as a Very Large Online Search Engine — alongside Reddit and Roblox as Very Large Online Platforms — gives OpenAI four months, until January 2027, to assess and mitigate systemic risks spanning illegal content, minors, wellbeing, fundamental rights, elections, and public security. The Commission’s register of designated services lists ChatGPT at 159.1 million average monthly EU users, well above the 45-million threshold, with Reddit at 57.2 million and Roblox at 46.6 million. The Commission’s press release is the formal record. Under the DSA text itself, the ceiling for breach is 6% of worldwide annual turnover — a number that scales with OpenAI’s revenue, not with its European ad take.

Follow the run rate, find the ramp

The ramp is the strongest fact OpenAI disclosed, and it is faster than the ad industry’s own precedents. The pilot began in the United States in February. By late March, CNBC reported that the ads business had passed $100 million in annualized revenue in under two months, with more than 600 advertisers, 85% of US free and Go users eligible, and fewer than 20% shown an ad on any given day. Five months later the figure is ten times larger and the advertiser count is “tens of thousands.”

Two derived numbers fall out of that sequence. The first is intensity: a 10x increase in run rate across roughly five months, which is not a mature channel’s growth pattern but a supply expansion — more eligible users, more countries, more inventory per session. The second is proportion. Bloomberg Law reported in August that OpenAI’s total revenue run rate topped $40 billion, a figure Epoch AI’s independent tracking of AI revenue puts in the same range by July. Against that base, advertising is 2.5% of OpenAI’s revenue — a diversification story for the prospectus, not yet a business line that changes the company’s unit economics.

There is a third number worth deriving, because it sets the ceiling on how fast this can compound. Meta’s filing shows ad impressions up 14% and average price per ad up 12% year over year — a mature machine growing revenue per person by roughly a quarter through volume and price together. OpenAI has neither lever fully engaged: it is not running a price auction anyone can see, and it is deliberately suppressing volume. Growth so far has come almost entirely from supply expansion — more countries, more eligible users, more sessions carrying an ad. Supply expansion is finite, and OpenAI is already in more than 40 countries. When the geographic runway ends, the run rate has to grow through frequency or price, and both of those move the product’s trust surface rather than its footprint.

Timing explains the emphasis. CFO Sarah Friar told employees in August that OpenAI “will be a public company in 2027,” possibly sooner, with enterprise run rate up 50% quarter to date and Anthropic’s own run rate reported at $65 billion. A subscription-and-API company walking into public markets against that comparison benefits enormously from a third engine that scales with users rather than with GPUs. The $1 billion line is aimed at that audience.

It is also a reversal executed in public. Anthropic built its first Super Bowl campaign around mocking the very idea, running an ad whose premise was that Claude has no ads weeks after OpenAI announced it would test advertising with US free and Go users. Seven months later, one company has a billion-dollar run rate and a European regulator; the other has a positioning statement. Which of those is the better asset depends entirely on whether the answer layer stays uncontaminated. The same discipline governs the infrastructure layer, where today’s brief on what a point of benchmark gain costs to buy shows the paper’s preferred test: divide the spend by the thing it actually produced.

The ways a dollar stays a dollar

Three things could keep ChatGPT’s revenue per user from ever approaching Meta’s, and none of them is speculative.

The first is inventory physics. Meta earns $67.56 per person because it shows an enormous number of impressions per session and prices each one in a live auction. OpenAI has deliberately capped exposure: ads appear for free and Go users only, under-18 accounts are excluded, and the company keeps them away from politics, health, and mental-health contexts. Fewer than one in five eligible US users saw an ad daily as of March. A product whose value proposition is a trustworthy answer cannot raise frequency without eroding the asset it monetizes.

The second is the compliance drag now attached to scale in Europe. Systemic-risk assessment, independent audits, and researcher data access are not one-time filings; they are permanent engineering and legal overhead applied to ranking, recommendation, and ad-delivery systems. The paper has tracked this cost curve since the EU AI Office staffed up its enforcement bench and since three of four EU transparency duties landed without a grace period. A VLOSE designation converts ChatGPT’s browsing and answer paths into regulated surfaces, and the 6% ceiling is calculated on global turnover — meaning European risk is priced against American and enterprise revenue.

The third is attribution. Meta’s advertisers can measure a click into a purchase across a pixel graph built over fifteen years. OpenAI states that ads are clearly labeled, do not influence answers, and do not give advertisers access to private conversations — the correct policy, and also a deliberate reduction in the targeting and measurement signal that makes feed advertising expensive. CNBC reported in March that some advertisers were already frustrated by the pilot’s conservative pace. A channel with a dollar of yield and thin measurement competes for budget against channels with sixty-eight dollars of yield and mature attribution.

What would change the verdict: a disclosed revenue-per-user figure from OpenAI rather than one derived from a run rate; a published auction mechanism with pricing; or evidence that conversational placements convert at multiples of feed placements, which would let a lower impression count carry a higher yield. Absent those, the $1.00 is the planning number.

What to do before the ad budget moves

The signal to watch is not the next run-rate headline. It is whether OpenAI begins reporting advertising as a segment with its own user and pricing metrics ahead of an IPO, because that is the moment the channel becomes measurable rather than announced.

  • Buy for reach measurement, not performance, this quarter. At roughly $1.00 of annual revenue per weekly user, ChatGPT’s ad inventory is priced and scaled like an emerging channel. Run a test budget sized to what you would spend learning a new surface, and benchmark it against the $67.56-per-person machine you already know, not against its promise.
  • Assume European delivery changes by January 2027. Any campaign plan that depends on current ChatGPT ad behavior in the EU should carry a contingency: the systemic-risk mitigations due under the DSA designation can alter placement, targeting, and minors-related eligibility without notice to buyers.
  • Separate the answer from the sponsorship in your own products. OpenAI’s constraint — labeled ads that do not influence responses — is the standard your users will apply to your assistant too, and the boundary is far cheaper to design in now than to retrofit under regulatory pressure.
  • Price the compliance overhead into any assistant with EU scale. The 45-million-user threshold is not theoretical for consumer AI products, and the obligations attach to the service, not the model vendor.
  • Watch the denominator, not the numerator. A run rate divided by a growing user base can rise while revenue per user falls. Ask any vendor quoting ad reach for both numbers, the same discipline the paper applied to Alphabet’s paper-gain-heavy pretax profit and to Meta’s newest training silicon, where a headline ratio hides the base it was computed on.

OpenAI has proven that an assistant can sell advertising quickly. It has not yet proven that an assistant can sell advertising densely, and the difference between those two sentences is sixty-seven dollars a user.

Sources