A purported internal document from OpenAI is casting serious doubts on the financial health of the AI giant and the broader generative AI industry's business model.

The document reveals that OpenAI's actual operating costs may far exceed external estimates, while its revenue appears significantly overstated, creating a staggering gap between its high expenses and income.

The controversy stems from tech blogger Ed Zitron, author of the newsletter *Where's Your Ed At* and a vocal critic of the AI industry. Despite his sharp critiques predicting an AI bubble burst, Zitron's PR firm, EZPR, has represented multiple AI startups—a dual role that makes him a polarizing figure. As an industry insider, he argues that the current AI boom is a capital-fueled bubble and warns of an impending "reckoning."

In his blog, Zitron cites leaked data indicating OpenAI's model "inference" costs on Microsoft Azure alone reached alarming levels. For instance, in H1 2025, these expenses neared $5 billion, while media reports pegged OpenAI's "cost of revenue" at just $2.5 billion—suggesting a burn rate nearly triple public figures.

The files also disclose OpenAI's 20% revenue share paid to Microsoft, its primary investor. Back-calculating, OpenAI's actual revenue falls far below reported numbers. For example, Microsoft's $454.7 million share in H1 2025 implies OpenAI's revenue was about $2.27 billion, starkly lower than the $4.3 billion media reports claimed. If accurate, this not only jeopardizes OpenAI's financial viability but also raises existential questions for the entire large-model AI sector.

Microsoft and OpenAI offered vague responses to the data. A Microsoft spokesperson told the *Financial Times* the figures were "not fully accurate" but declined to elaborate. OpenAI deferred queries to Microsoft, while a source close to the company called the data "incomplete." Neither party provided substantive counterevidence.

**Shocking Cost Overruns: Inference Expenses Dwarf Revenue** Zitron's data shows OpenAI's inference costs—expenses tied to generating responses via models like ChatGPT—far outpace revenue, with the gap wider than previously known. From Q1 2024 to Q3 2025, OpenAI spent over $12.4 billion on Azure inference computing, including $8.67 billion in the first nine months of 2025. By contrast, *The Information* reported OpenAI's 2024 inference costs at ~$2 billion and H1 2025 "cost of revenue" at $2.5 billion—meaning actual costs may be triple earlier estimates.

Worse, these ballooning costs aren’t matched by revenue growth. Data suggests inference expenses are rising linearly, eclipsing income—a pattern casting doubt on large-model profitability under current tech and pricing.

**Overstated Revenue? Discrepancies with Public Claims** Beyond costs, Zitron’s files challenge OpenAI’s revenue figures. Microsoft’s 20% share implies:

- **2024**: Microsoft received $493.8 million, indicating OpenAI’s revenue was at least $2.469 billion—well below CNBC/*The Information*’s $3.7–4 billion projections. - **H1 2025**: Microsoft’s $454.7 million share suggests OpenAI earned ≥$2.273 billion, versus reported $4.3 billion.

These gaps are irreconcilable. Even CEO Sam Altman’s claim of "well over" $13 billion annualized revenue clashes with the document’s figures. Zitron speculates inflated public numbers may stem from creative "annualized" or ARR metrics, which OpenAI hasn’t clarified.

**Complex Partnerships and Murky Responses** OpenAI and Microsoft’s financial ties are intricate, featuring reciprocal revenue-sharing. Beyond taking 20% of OpenAI’s income, Microsoft pays OpenAI 20% of Azure OpenAI service sales, plus Bing-related shares and potential royalties. This complexity means Microsoft’s share alone understates total revenue—yet the cost-revenue chasm remains unexplained.

When pressed, neither firm firmly denied the data. Microsoft’s "not fully accurate" remark and a source’s "incomplete picture" defense leave room for concern. Without concrete rebuttals, the leaks gain credibility.

**Industry Wake-Up Call: AI’s Sustainability in Question** If even partially accurate, Zitron’s data signals dire straits for generative AI. *Financial Times* projections show OpenAI’s revenue might only cover inference costs by ~2033 at current growth rates—and after Microsoft’s cut, it could "never" break even.

This poses a brutal question: If OpenAI, the best-funded market leader, faces such pressure, how can other general-model vendors survive? Analysts see two paths: either model costs plummet, or customer fees soar. Neither trend is visible, darkening generative AI’s commercial outlook.