For most of the past three years, the question of how frontier AI actually pays for itself sat politely in the background while the models got bigger and the funding rounds got larger. This week it moved to the foreground, and it did so from two directions at once. OpenAI began turning its most popular product into advertising space. Anthropic, meanwhile, let it be known that its case for a record public offering rests on a revenue figure two years out that is roughly four times its current pace of business.

Neither move is a surprise on its own. Together they sketch the two routes the industry is now betting on to close the gap between what these systems cost to run and what anyone is willing to pay for them.

Ads arrive in ChatGPT

OpenAI has started showing ads inside ChatGPT. They appear as a "sponsored" block beneath the assistant's main answer, tied to what you asked about, and they are limited to the Free and Go tiers. Anyone paying for Plus, Pro, Business, Enterprise or Education keeps an ad-free product.

The rollout has been geographic and quick. Ads are already running for Free and Go users across the United States, Canada, Australia and New Zealand, and OpenAI has switched them on in the United Kingdom, Mexico, Brazil, Japan and South Korea. In the EU the company says personalised ads will only reach users who explicitly opt in, a concession to European privacy rules rather than a change of direction.

It is worth being precise about what this is. ChatGPT is, by a wide margin, the most-used consumer AI product in the world, and the free tier is where the overwhelming majority of that use happens. Putting a sponsored slot under the answer converts all of that attention into inventory. It is the oldest business model on the consumer internet, arriving in the newest consumer software, and it tells you that OpenAI has decided the free tier cannot simply be a loss leader forever.

Anthropic asks to be priced on 2028

The same week, Reuters reported that Anthropic's pitch to public-market investors hinges on a projection it has not disclosed before: revenue of roughly 190 to 200 billion dollars in 2028. The company is preparing what could be one of the largest listings on record, and bankers are being asked to value it not on what it earns today but on what it might earn two years from now.

The distance between those two numbers is the story. As recently as May, Anthropic pointed to a revenue run-rate of about 47 billion dollars. The IPO case asks investors to underwrite a jump to four times that inside three years. Pricing a company on forward revenue multiples is normal for fast-growing software firms that have not yet settled into steady profits; pricing it on a forecast this far out, at this scale, is Wall Street reaching further into the future than it usually cares to.

Two answers to one question

Strip away the detail and both companies are answering the same question, and answering it differently.

OpenAI is monetising attention. Its lever is a vast free user base, and the plan is to sell access to it the way search and social platforms always have. That is a high-volume, low-margin-per-user model, and it comes with the familiar tensions: an assistant people trust for answers now has a commercial reason to shape what sits next to those answers.

Anthropic is monetising revenue growth. Its lever is enterprise contracts, and its case to investors is that business demand will compound fast enough to justify a valuation most companies never reach. That is a lower-volume, higher-value model, and its risk is simpler to state: the forecast has to come true.

Both routes exist because the underlying arithmetic has not gone away. These systems are expensive to train and expensive to serve, and the era in which that could be waved off as an investment phase is closing. Ads in ChatGPT and a nine-figure revenue forecast at Anthropic are the two most concrete signals yet that the leading labs are being asked, finally, to show their working.

What to watch

Two things. First, whether ChatGPT's ads stay in that quiet sponsored slot or start to colour the answers themselves — the line between the two is thinner in a conversational product than it ever was in a list of search results. Second, whether Anthropic's 2028 number survives contact with the prospectus. A figure briefed to reporters is one thing; a figure a company has to stand behind in a public filing is another. Both will tell us more about the real economics of frontier AI than any benchmark released this month.