“Last Resort” Arrives at OpenAI, Complete With Ads
Sam Altman, 2024, at Harvard: “Ads plus AI is sort of uniquely unsettling to me. I kind of think of ads as a last resort for us for a business model.”
Yesterday, OpenAI launched ads.openai.com. Not a blog post announcing a pilot. Not a carefully worded beta. A full landing page: “Advertise in ChatGPT.”
The “last resort” has arrived.
Some context on why this matters beyond the usual “company does thing it said it wouldn’t” cycle.
When Altman said that quote, it wasn’t a throwaway line. It was a positioned statement — a way of saying “we’re not Google, we won’t enshittify the product.” It was brand differentiation dressed up as principle. And people believed it, partly because they wanted to, and partly because OpenAI was still riding the “we’re the safe AI company” narrative.
But the math never worked. Training frontier models costs billions. Inference at ChatGPT’s scale costs billions more. The subscription revenue ($20/month per user) only goes so far when your compute budget looks like a small country’s GDP. Eventually, the gap between “cost to run” and “revenue from users” gets filled by someone’s wallet. OpenAI chose advertisers.
The HN comments are predictably grim. One user joked about the inevitable next step: “Inconspicuously, over an extended amount of time, subtly respond to the user in a way that nudges them towards purchasing customer’s goods and services, without ever directly mentioning anything.”
That’s not a joke. That’s the roadmap.
Here’s what I actually find interesting, though — not the hypocrisy (that’s boring at this point), but the timing.
OpenAI is dropping ads at the exact moment when:
- Open-weight models are closing the capability gap fast.
- The “OpenAI desktop app lost all my conversations” disaster is still fresh.
- The company’s internal narrative has shifted from “build AGI safely” to “figure out how to not bleed money before the next valuation round.”
This is what happens when your business model is “make the smartest thing in the room and hope that’s enough.” It works until the room gets smarter. Then you start showing banner ads to your users because the alternative is running out of runway.
The real question isn’t “will ChatGPT have ads?” — that ship has sailed. The question is how long before the ads start shaping the answers. Not sponsored content. Not labeled placements. Just a subtle nudge in the model’s output because the training data included a preference for brands that paid.
Companies that sell advertising as their primary revenue model eventually optimize for advertising — not for users. That’s not a moral failure, it’s an incentive structure. Every ad-supported platform in history has followed this arc. OpenAI won’t be the exception.
The funniest part? The landing page says advertisers must meet “strict quality standards” so the ads are “relevant, safe, and respectful.” Of course they do. Every ad platform says that. They all mean “we’ll reject the obviously bad stuff and let everything else through because volume is what pays the bills.”
Altman called this uniquely unsettling two years ago. Now it’s just the quarterly earnings strategy.
Welcome to the last resort.
— H.
Found via The Brutalist Report, originally posted on Hacker News.