The curtain is pulling back on OpenAI’s finances, and the picture emerging is both fascinating and concerning. Thanks to leaked documents analyzed by tech blogger Ed Zitron, we’re getting a rare look at what it actually costs to run the world’s most talked-about AI company, and the numbers are eye-opening.
According to the leaked information, Microsoft pocketed $493.8 million in revenue share payments from OpenAI in 2024. That figure nearly doubled to $865.8 million in just the first three quarters of 2025. This stems from a deal where Microsoft, which has invested over $13 billion in OpenAI, receives approximately 20% of the AI startup’s revenue.
But here’s where things get complicated. Microsoft isn’t just taking money from OpenAI, it’s also paying money back. The tech giant reportedly kicks back around 20% of revenues from Bing and Azure OpenAI Service to OpenAI, according to sources familiar with the arrangement.
Bing’s search results are powered by OpenAI’s technology, while Azure OpenAI Service sells cloud access to OpenAI’s models for developers and businesses.
OpenAI Projected to Surpass $20 Billion Annual Revenue While Battling Massive Compute Costs
A source told TechCrunch that the leaked payment figures represent Microsoft’s net revenue share, not the gross amount. This means they don’t include what Microsoft paid OpenAI from Bing and Azure royalties. Microsoft subtracts those payments before reporting its internal revenue share numbers, making it difficult to calculate the complete financial picture.
Using the 20% revenue-share figure as a baseline, we can estimate that OpenAI brought in at least $2.5 billion in revenue during 2024, with that number jumping to $4.33 billion in the first nine months of 2025. However, these are likely conservative estimates. Previous reporting from The Information suggested OpenAI’s 2024 revenue was closer to $4 billion, with the first half of 2025 alone generating $4.3 billion.
OpenAI CEO Sam Altman has been even more bullish about the company’s trajectory. He recently claimed OpenAI’s revenue is “well more” than the reported $13 billion annual figure and projects the company will close out the year above $20 billion in annualized revenue run rate. He’s even suggested the company could hit $100 billion by 2027.
Revenue is one thing, but spending is where the story gets really interesting. Zitron’s analysis suggests OpenAI spent roughly $3.8 billion on inference in 2024—the compute power needed to run trained AI models and generate responses. That spending exploded to approximately $8.65 billion in just the first nine months of 2025.
Is OpenAI’s Inference Spending Signaling a Bubble?
For years, OpenAI has relied almost exclusively on Microsoft Azure for its computing needs, though it has recently diversified with deals involving CoreWeave, Oracle, AWS, and Google Cloud. Earlier reports estimated OpenAI’s total compute spending at $5.6 billion for 2024, with cost of revenue hitting $2.5 billion in the first half of 2025.
Here’s a crucial detail: while OpenAI’s training spend, the resources needed to initially train models, is mostly covered by non-cash credits from Microsoft’s investment, the inference spending is largely paid in actual cash. This distinction matters because inference costs represent real money flowing out the door.
When you crunch the numbers, a troubling pattern emerges. OpenAI appears to be spending more on inference costs than it’s generating in revenue. This isn’t just an academic concern, it raises fundamental questions about the sustainability of the entire AI industry.
If OpenAI, the most prominent and well-funded AI company on the planet, is still operating in the red just to keep its models running, what does that signal for the hundreds of other AI startups valued at astronomical figures? The implications are reverberating through investment circles from Wall Street to Silicon Valley, adding fuel to ongoing debates about whether we’re witnessing an AI bubble.
Neither OpenAI nor Microsoft responded to requests for comment on these financial revelations, leaving the tech world to draw its own conclusions about the true cost of the AI revolution.
