How Does OpenAI Make Money? (2026 Revenue Breakdown)

But the real story is volume. ChatGPT reached around 700 million weekly active users by mid-2026, and yet only about 20 million of them pay. That is a conversion rate near 3%, which sounds tiny until you multiply 20 million seats by $20-plus a month.

Here is something most articles skip. OpenAI launched ChatGPT Go at $8, partly to capture price-sensitive markets, first in India and then worldwide. Cheaper plans in price-sensitive regions widen the funnel, and a wider funnel feeds the whole revenue model.

Pro Tip: When you read a headline about OpenAI’s user count, check whether it says “weekly active” or “paying.”

The gap between the two, hundreds of millions of free users versus around 20 million payers, is the exact problem the ads program is trying to solve.

How does OpenAI make money from its API?

The API earns roughly 25% of revenue by charging developers a small fee for every chunk of text a model processes. This is the quiet workhorse of the business, and it scales automatically as more apps get built.

Pricing is based on tokens, with about 1,000 tokens roughly equal to 750 words. You pay for the words you send in and the words the model generates back, so a chatty app costs more than a simple one.

Honestly, I expected the API to be a bigger slice than it is. So many startups run on OpenAI’s models that you’d think it would rival subscriptions, but consumer ChatGPT simply dwarfs it on raw dollars.

There’s a second layer here, too. OpenAI offers fine-tuning, where a company pays extra to train a model on its own data. That premium service matters most in specialized fields like medicine, law, and engineering, where a generic model just won’t cut it.

And there’s a risk worth naming. A price war with rivals like Anthropic could force OpenAI to cut token prices, squeezing already-thin margins on this stream. So API revenue is growing, but it is not a comfortable, protected cash cow.

What does OpenAI earn from its Microsoft partnership?

The Microsoft relationship cuts both ways: it brings in licensing revenue but also costs OpenAI a large share of its income. Microsoft invested around $13 billion over the years and hosts OpenAI’s models on Azure.

Most coverage treats this deal as pure upside for OpenAI. The picture is more tangled than that. Under terms renegotiated in October 2025, OpenAI agreed to pay Microsoft 20% of its revenue through 2032, according to Sacra’s reporting.

So Microsoft is both a customer and a tollbooth. It resells OpenAI’s models through the Azure OpenAI Service to enterprises, which generates licensing income for OpenAI. But that 20% cut means a fifth of the top line flows straight back to Redmond.

Compared to a normal cloud vendor relationship, this one is unusually deep. The revised 2025 terms also removed Microsoft’s right of first refusal on new cloud workloads, which freed OpenAI to sign massive deals with Amazon and Oracle. That shift tells you OpenAI wants to depend on Microsoft less, not more.

Are ads now a revenue stream for OpenAI?

Yes, OpenAI began testing ads inside ChatGPT in February 2026, and they appear only on the Free and $8 Go tiers in the US. Paid plans like Plus, Pro, Business, and Enterprise stay completely ad-free.

This was a genuine philosophy shift. Sam Altman once called mixing AI and advertising “uniquely unsettling” back in 2024, and treated ads as a last resort. The math eventually won the argument, because only about 5% of users pay, while infrastructure bills keep climbing.

Here is how the ads actually work, since few explainers get into the mechanics:

1. Ads show at the bottom of an answer when a relevant sponsored product or service matches your conversation.
   
   
2. They are clearly labeled as sponsored and visually separated from the organic reply.
   
   
3. OpenAI says ads never influence the model’s actual answer.
   
   
4. No ads appear for users under 18 or near sensitive topics like health, mental health, or politics.
   
   
5. Microsoft Advertising handles the ad sales through its existing Bing infrastructure.

The early pricing was steep. Reports pointed to a $60 CPM and a $200,000 minimum spend at launch, which made it an enterprise-only play at first. Some ad networks have since slashed their minimums to attract more advertisers.

If you run a business or a blog, this matters because it signals where AI traffic is heading. The moment a 700-million-user assistant starts serving sponsored answers, the entire discovery layer of the web begins shifting.

Do Sora and licensing add to revenue?

Sora and other licensing deals make up the smallest slice, roughly 5% of revenue, but they point to where OpenAI wants to grow next. Sora is OpenAI’s short-form AI video product, and it briefly attracted big media interest.

One vivid example: Disney announced a $1 billion investment in OpenAI in December 2025 and signed a licensing deal that lets users generate videos featuring hundreds of Disney, Marvel, Star Wars, and Pixar characters.

Disney later exited that deal in March 2026, after Sora was discontinued, showing how quickly these experimental lines can rise and fall. So this bucket is real money, but it is volatile. Treat it as the experimental wing of the business rather than a stable pillar.

OpenAI runs a hybrid model: a freemium consumer product on top, a usage-based developer platform underneath, and a for-profit company now occupying the space once occupied by a non-profit.

The structure has changed more than once. It started in 2015 as a non-profit research lab. In 2019, it adopted a “capped-profit” structure, enabling it to raise venture capital while limiting investors’ returns.

Then, in 2025, it restructured again into a for-profit public benefit corporation, clearing the path to raise huge sums and eventually go public.

The platform logic is what gives the business leverage. Free users improve the models through feedback and word of mouth. Some convert to paid plans.

Developers build on the API, which spreads OpenAI’s models into thousands of other products. Each layer feeds the next.

In My Experience

After tracking OpenAI’s numbers and using the paid tiers for over a year, the thing that surprised me most was how quietly the business shifted under everyone’s feet.

I paid for Plus mainly for faster responses, never thinking about where my $20 actually went.

When the ads rollout hit in early 2026, I finally saw the strategy click into place. On a free account, the sponsored block sat right at the bottom of an answer on mobile, and it ate up a noticeable chunk of a small screen.

It didn’t change the answer, but it absolutely changed the feel. What caught me off guard was the gap between perception and reality. Friends assumed OpenAI was swimming in profit because ChatGPT is everywhere.

The actual financials tell a story of enormous revenue and even more enormous spending. Using the product daily, you’d never guess the company behind it loses billions.

One limitation I kept bumping into: because OpenAI is private, you can’t fully verify any single figure.

You triangulate between Sacra, Value Add VC, The Information’s leaks, and OpenAI’s own posts, and you settle on a range rather than a clean number. That uncertainty is part of the honest answer here.