· Enterprise AI · 3 min read
Will Targeted Ads Pull OpenAI Out of the Red?
Targeted ads seem inevitable for ChatGPT, but high inference costs and uneven CPMs may blunt the payoff.
Recently, OpenAI signaled a lot of moves to prepare for launching targeted ads on their products:
- Adding shopping and payment right inside the ChatGPT app
- Releasing Atlas, an AI browser
- Hiring an army of ex-Meta staff on the product side
- Pushing on the “memory” feature to record info about user preferences
According to The Information, OpenAI is ready for its Facebook Era. OpenAI is estimated to make $4.3B in H1 2025 and have $13.5B in losses. They currently have 800M weekly active users, so targeted ads in Facebook style seem to be the best way to go. Will that work out for them?
The biggest difference between Facebook and ChatGPT is the cost to run. Facebook has virtually 0 cost for each free user (a bit of storage and bandwidth). OpenAI incurs cost every time a free user uses the app. Now, OpenAI needs to make sure that the revenue from ads is significantly higher than the serving costs.
There are 2 types of ads: search ads (like when you search on Google) and display ads (like when you scroll your Facebook feed).
Search ads are highly targeted, so they usually command a higher CPM (cost per thousand impressions). ChatGPT mainly falls into this case, so that is good for OpenAI. However, search ads are limited. If users don’t search for shoes, it would be weird for ChatGPT to recommend a pair of shoes. So there is a cap on how much ChatGPT can make from search ads.
Display ads don’t have the limitation of search ads, but ChatGPT is not suitable for showing display ads. OpenAI has to resort to Sora for this type of ads. Whether Sora will take off is still unclear. And Sora videos even cost way more than ChatGPT queries.
One more challenge for OpenAI is the cost of serving LLMs is the same across countries, but the ad price must differ substantially. They can’t charge a high price for businesses in developing countries, because businesses only buy ads when they can make more than they spend.
Besides, a tag “ads” right next to the recommendation of a pair of shoes doesn’t look good. Now, users trust ChatGPT, but that high trust may erode quickly when they see these tags.
One positive point, which could counter all the challenges above, is that the cost of serving will go down and down even more, while the quality goes up and up. At some point, say GPT-7-nano is as good as GPT-5 now, then serving costs become insignificant compared to ad revenue. At that point, OpenAI’s business will be very healthy.
The average cost of ChatGPT per session is not published, but based on OpenAI loses and the average CPM of Facebook and Google, my wild guess is ad revenue would make up around the serving cost now.
It’s unclear whether the targeted ads model of OpenAI will pay off. But one thing is for sure, that is the clearest way to move OpenAI’s bottom line out of the red.
