OpenAI Earned $10 Billion in a Quarter but Lost $10.9 Billion on Operations

It sounds like a business joke: sold $10 billion, lost another $10.9 billion. But it is no joke; it is OpenAI's third quarter of 2026. The figures come from The Wall Street Journal, citing financial documents prepared for investors, which means this is not public, audited reporting. Even so, the picture is telling.

What the Numbers Show

OpenAI's revenue in the third quarter was $10 billion. Over the first nine months of the year it added up to about $22.4 billion. Simple arithmetic: the first two quarters account for roughly $12.4 billion, or an average of about $6.2 billion per quarter. The third quarter is noticeably above that mark, so the pace of sales is growing.

At the same time, the adjusted operating loss for the quarter is $10.9 billion, more than the revenue itself. Roughly speaking, adding revenue and loss together suggests operating expenses for the three months of around $20.9 billion. This is an estimate based on two published numbers, not a disclosed cost line. The article does not say what exactly is included in the "adjusted" figure and what has been excluded from it.

One more number: the count of paying ChatGPT users has approached 100 million. The available data contains no breakdown by subscriptions, enterprise plans and API, so you should not divide revenue by users to derive an "average check." Such arithmetic would be neat but wrong.

Why This Matters to Those Building on OpenAI

When a supplier spends more than a dollar for every dollar it earns, its prices are determined not only by cost but also by strategy. As long as investors are willing to fund growth, API and subscription prices can be kept attractive. If sentiment changes, the company will have several levers: raise prices, cut free limits, repackage plans, or tighten terms for large customers.

This is not a prediction that it will happen. It is a scenario worth keeping in mind when calculating your product's unit economics. If your service's margin rests on the current per-token price, you effectively depend on someone else's subsidy.

In practice it looks like this. Say you are building a customer support assistant, and each conversation costs you a few cents in model calls. Today, at your subscription price, that yields a healthy margin. But if the cost of calls rises by one and a half to two times, some plans will go into the red. It is better to calculate this in advance than to learn about it from a pricing-change email.

What Developers and Product Managers Should Do

There is no need to panic and leave OpenAI. You just need to do a few inexpensive things.

• Put model calls behind a thin abstraction layer so that switching providers is a matter of days, not months.
• Set up a regression test suite for answer quality. Without it, comparing alternatives and migrating safely is impossible.
• Run your unit economics with a token price 50–100% higher than the current one and see which plans stop paying off.
• For simple tasks, try smaller models or open alternatives. They often cover 80% of requests at a fraction of the cost.

Limitations

The figures came through the press and investor documents, not from audited reporting. An "adjusted" loss may exclude some expenses, so a direct comparison with other companies is not valid. Moreover, unprofitability during the scaling phase is not unusual for technology companies: the question is how long the period lasts during which the market is willing to pay for it.

The bottom line is simple. OpenAI is growing fast and spending even faster. For a developer this is not a reason to abandon the platform, but it is a good reason not to build a business on the assumption that current prices will stay the same forever.