Is the AI business model about to crash? Tech bloggers delve into OpenAI's "financial black hole": the burn rate is three times that of publicly available data, and revenue is exaggerated and cannot cover costs!

A document allegedly from within OpenAI poses a severe challenge to the financial status of this artificial intelligence giant and the business model of the entire generative AI industry.

The document indicates that OpenAI's actual operating costs may far exceed external expectations, while its revenues are significantly exaggerated, creating a shocking gap between its high operating costs and revenues.

The source of this upheaval is well-known tech blogger, author of the newsletter "Where's Your Ed At," and a sharp critic of the tech industry—Ed Zitron. On one hand, he vehemently criticizes the AI industry through his personal media platform, predicting that its bubble is about to burst; on the other hand, his PR firm EZPR has provided public relations services for several AI startups. This dual role of both criticizing AI and representing AI companies makes him a controversial voice in the tech world. As an "insider" well-versed in the industry's operational rules, he firmly believes that the current AI craze is essentially a bubble driven by capital, and he predicts that the "judgment day" for the entire industry is imminent.

Zitron published an article on his blog, citing data from the document he reviewed, claiming that OpenAI's "inference" costs on Microsoft's Azure cloud platform have reached astonishing levels. For example, in the first half of 2025, this expenditure is close to $5 billion, while the previously reported "cost of revenue" for the same period was only $2.5 billion. This indicates that OpenAI's burn rate may be nearly three times that of publicly available data.

At the same time, these documents also reveal that OpenAI pays a 20% revenue share to its main investor, Microsoft. By reverse calculating this data, OpenAI's actual revenue is far lower than previously reported figures. For instance, in the first half of 2025, the revenue share received by Microsoft corresponds to OpenAI's revenue of about $2.27 billion, which is significantly different from the $4.3 billion reported by the media. If this data is accurate, not only is OpenAI's financial health concerning, but the profitability prospects of the entire large AI model industry will also be cast into serious doubt.

Regarding the accuracy of this data, the responses from Microsoft and OpenAI have been vague. A female spokesperson from Microsoft told the Financial Times that "the numbers are not entirely correct," but refused to provide further details. OpenAI did not comment, only suggesting that the media verify with Microsoft. A person familiar with OpenAI stated that these numbers do not provide a "complete picture." Although official confirmation could not be obtained, the parties involved have also failed to provide strong evidence to substantively refute this data.

A Shocking Cost Black Hole: Inference Expenses Far Exceed Revenue

According to the data disclosed by Ed Zitron, there is a huge gap between OpenAI's operating costs (specifically, model inference costs) and its revenues, and this gap is larger than any previous reports. Inference is the process of calling large language models to generate responses in applications like ChatGPT, and it is one of the core operating costs Data shows that from the first quarter of 2024 to the third quarter of 2025, OpenAI's spending on inference computing on Azure alone exceeded $12.4 billion. In the first nine months of 2025, its inference costs reached $8.67 billion. In contrast, The Information previously reported that OpenAI's inference costs for the entire year of 2024 were about $2 billion, and the "cost of revenue" for the first half of 2025 was $2.5 billion. Zitron's data indicates that the actual costs are nearly three times these reported figures.

Even more concerning is that these astronomical costs do not seem to align with revenue growth. Data shows that even when considering revenue growth, inference spending appears to be rising linearly at a faster rate, completely consuming revenue. This pattern raises questions about whether large model businesses can achieve profitability under current technology and pricing.

Exaggerated Revenue? Significant Discrepancies Between Revenue Data and Public Reports

In addition to high costs, documents disclosed by Zitron raise serious doubts about OpenAI's revenue. By reverse calculating from the 20% revenue share received from Microsoft, the minimum revenue level for OpenAI can be inferred, and this figure is far from the optimistic data cited by the media or disclosed by OpenAI itself.

Fiscal Year 2024: The documents show that Microsoft received $493.8 million in revenue sharing that year, implying that OpenAI's revenue was at least $2.469 billion. However, reports from CNBC and The Information at the time predicted that OpenAI's revenue for 2024 would be between $3.7 billion and $4 billion.

First Half of 2025: Microsoft received $454.7 million in revenue sharing, suggesting that OpenAI's revenue during the same period was at least $2.273 billion. Yet, The Information reported that OpenAI generated $4.3 billion in revenue during that time.

These significant discrepancies are hard to reconcile. Even OpenAI CEO Sam Altman's claim that the company's annualized revenue "far exceeds" $13 billion is inconsistent with the financial situation revealed in the documents. Zitron speculates that the high revenue figures reported publicly may stem from a creative calculation method for "annualized revenue" or "annual recurring revenue" (ARR), but OpenAI has never clarified its definition.

Complex Partnership and Ambiguous Responses

It is noteworthy that the financial relationship between OpenAI and Microsoft is extremely complex and not simply an investment and cost relationship. Reports indicate that there is a two-way revenue-sharing agreement between the two parties. In addition to taking 20% from OpenAI's revenue, Microsoft also needs to pay OpenAI 20% of the revenue generated from selling models through Azure OpenAI services. Additionally, there are revenue shares related to Bing and potential royalty payments.

**This complex structure means that inferring OpenAI's total revenue solely from the revenue share received from Microsoft will inevitably lead to an "undervaluation." However, even considering this, the significant gap between costs and revenues revealed by Zitron's data remains difficult to explain **

According to reports, when verifying with the two companies, the Financial Times did not receive a clear denial. Microsoft's response that "the numbers are not entirely accurate" and a source's statement that "the complete picture could not be provided" leave a significant space for market imagination (or concern). In the absence of more specific rebuttals, the credibility of the leaked data is increasing.

Industry Alarm: The Sustainability of AI Business Models Under Scrutiny

If the data disclosed by Zitron is even partially accurate, it will sound the alarm for the entire generative AI industry. This means that OpenAI, as the industry leader, may find its business model fundamentally unsustainable.

The Financial Times' projections based on this data indicate that at the current growth rate, OpenAI's minimum estimated revenue may not cover inference costs until around 2033. If we further deduct Microsoft's 20% revenue share, it may "never" be able to cover inference costs through its own revenue.

This finding poses a sharp question for investors and the market: If even the most well-funded and market-leading OpenAI faces such significant financial pressure, what will be the situation for other general large model providers? Analysts believe that there are only two possible futures: either the operating costs of the models must drop dramatically, or the charging standards for customers must be significantly increased. However, currently, there are no signs of either trend, which casts a thick shadow over the commercialization prospects of generative AI