TheChek News Flow

Because OpenAI’s for-profit business is controlled by a nonprofit, investors currently get stock that entitles them to a share of future profits. The company is trying to convert the shares to traditional equity so its for-profit unit can go public someday, though it faces legal and contractual tussles with the likes of Elon Musk and Microsoft, which could affect the timing and structure of that proposed conversion.

Being publicly traded could make it easier for OpenAI to borrow money and pursue some of its data center ambitions, which could include renting out its servers to other AI developers, according to recent comments from its executives.

Cost Breakdown

Until then, it has to find ways of paying the almost unfathomable costs of developing its technology and serving it to what it hopes will be billions of customers. At the same time, OpenAI is grappling with increasing competition for talent that has already led to key departures and increased salaries and stock compensation.

For instance, OpenAI has projected about $20 billion in additional stock compensation charges between now and 2030 compared to its earlier projection.

But the biggest change emerging from OpenAI’s latest projections was to its cash flows. The company projected it will burn more than $8 billion this year, or roughly $1.5 billion higher than its prior projection from earlier this year. Cash burn will more than double to more than $17 billion next year—$10 billion higher than what the company earlier projected.

And in 2027 and 2028, the company projects to burn roughly $35 billion and $45 billion, respectively. In the prior projection, the company said its 2028 cash burn would be $11 billion, meaning the new estimate is more than four times higher.

One factor pushing up the company’s cash burn is its plan to spend nearly $100 billion later this decade on costs related to servers it would control, the new data suggest. CFO Sarah Friar has said OpenAI could offer servers to other companies the way Amazon Web Services does.

Another large driver of the increased costs is computing costs. While the cost of running its AI models, known as inference compute expenses, is close to what it projected earlier this year, the small differences add up to more than $11 billion in extra spending by the end of the decade. The company expects to spend more than $150 billion on such costs from 2025 through 2030.

And its cost to develop AI models, also called training or research compute, is far higher than it previously projected.

The company projected a steady increase in AI training costs through 2030, which suggests this type of expense could grow even after then.

The company projected a steady increase in these costs through 2030, which suggests this type of expense could grow even after then. As the company’s struggles to develop GPT-5 this year showed, training AI can yield unpredictable results, including false starts and do-overs that can raise costs.

On the bright side, OpenAI’s revenue outlook is improving somewhat.

Overall, the company has projected $13 billion in total revenue this year, up three and a half times from last year and about $300 million higher than its earlier projections for 2025, the new data show. Its revenue projection for 2030 rose about 15% to roughly $200 billion compared to the prior projection.

ChatGPT is a large driver of the company’s increased projections. OpenAI expects the chatbot to generate nearly $70 billion in additional revenue over the next six years, compared to earlier projections. Millions of people and thousands of businesses pay subscription fees to use ChatGPT.

OpenAI projected nearly $10 billion in revenue from ChatGPT this year, an increase of roughly $2 billion from projections earlier this year, and nearly $90 billion in revenue from the chatbot in 2030, a roughly 40% increase from the earlier projections.

OpenAI’s projections also raised expectations for how it will generate revenue from users who don’t pay for ChatGPT. It’s unclear how OpenAI plans to make money off that portion of its user base, although it could include shopping-related services or some form of advertising. The data show the company expects to generate around $110 billion in revenue between 2026 and 2030 from such services.

Altman had previously discussed the possibility of charging shopping-related affiliate fees–or a cut of a sale initiated from a user’s search on ChatGPT. Earlier this year, Friar had also discussed the possibility of selling ads but said the company has no active plans to do so.

Facebook-Like Margins

In its earlier projections, OpenAI projected that the average revenue per user from “free user monetization” products would be $2 per year starting next year and $15 per user per year by the end of the decade, by which time it expected to have two billion weekly active users. OpenAI also told investors these products to have gross profits margins similar to those of Meta Platforms’ Facebook, between 80% and 85%. It isn’t clear if these expectations have changed lately.

Notably, OpenAI lowered its projections for generating revenue from its application programming interface, which companies such as Anysphere and Salesforce use to power their AI applications, by $5 billion over the next five years. And it projected about $26 billion less revenue from “agents,” or products that involve AI that handles tasks involving multiple steps. The reasons for that couldn’t be learned, but it’s possible that more such technology will be built into ChatGPT rather than offered as a separate product.

In any case, the increase in ChatGPT revenue projections more than made up for those decreases.

Microsoft, which is OpenAI’s biggest outside shareholder, is entitled to 20% of OpenAI’s revenue, and the startup’s cash flow projections account for that payout.