rewrite this title ChatGPT Is Getting Ads—Here’s What to Expect – Decrypt

Remember when Sam Altman said ads in AI were “uniquely unsettling” and a “last resort?” Well, a year and a half later, OpenAI is gearing up to roll out advertisements to ChatGPT.

The company announced on Friday that it will start testing ads in the U.S. within weeks. Free users and subscribers to the new $8/month ChatGPT Go tier will see sponsored content at the bottom of responses. Plus, Pro, Business, and Enterprise remain ad-free.

What changed Altman’s mind? Probably money. Or rather, the lack of it.

OpenAI burned through roughly $8 billion in 2025. Internal documents show projected operating losses hitting $74 billion by 2028. The company has 800 million weekly users, but only 5% pay for subscriptions. Meanwhile, it has committed over $1.4 trillion to infrastructure spending.

Also, there is fierce competition in the AI space, and ChatGPT is losing market share. According to Similarweb data, ChatGPT dropped from 87% of the AI chatbot market in January 2025 to around 65% by this month. Google Gemini surged from 5% to over 18% during the same span. That’s not a small shift.

Besides this, there’s the problem of hardware supply. Google has something OpenAI doesn’t: its own chips. Google’s TPUs have been in development for over a decade, and they cost roughly 4-6 times less per unit of compute than the Nvidia GPUs that OpenAI depends on.



When OpenAI pays for compute hardware, a massive chunk goes to Nvidia’s margins—also known as the “Nvidia tax”. When Google runs Gemini, the company is essentially paying itself.

Google also has distribution that OpenAI can’t match. Gemini is baked into Android, Gmail, Chrome, and YouTube—which makes it extremely easy (if not unavoidable) for users to have contact with its AI model. OpenAI, as ubiquitous as it is, still has to convince people to visit a website.

So, ads it is.

What to expect

OpenAI published five “ad principles”: ads must be beneficial, they won’t influence ChatGPT’s answers, conversations stay private from advertisers, you can turn off personalization, and user experience is supposed to be prioritized.

Sound familiar? Google said similar things about search quality. That worked until revenue pressure took over. The difference is that Google had profitable businesses to subsidize search for years. OpenAI doesn’t have that kind of runway.

Whether those promises hold depends on execution. Advertising at scale has historically succeeded, with Google and Meta as clear examples—both of which built precise targeting systems by collecting enormous amounts of user data.

ChatGPT’s memory feature and conversational data offer similar potential. The platform knows what users ask about their jobs, their health, and their relationships—information advertisers prize. So targeting ads in the world’s most popular AI chatbot may prove to be extremely profitable.

For now, the first ads will be basic product placements below responses. OpenAI said it will be “extremely respectful” of user data and refine the format based on feedback. The test starts in the U.S. for adult users in the coming weeks. Whether it’s enough to close a $74 billion hole remains to be seen.