内容来源：https://techcrunch.com/2025/11/14/leaked-documents-shed-light-into-how-much-openai-pays-microsoft/

内容总结：

【独家披露】OpenAI财务数据疑云：年化收入或超200亿美元，但运营成本高企恐陷亏损泥潭

据科技博主埃德·齐特隆最新获得的内部文件显示，在经历一年疯狂交易与IPO传闻后，人工智能巨头OpenAI的财务状况正引发市场高度关注。泄露文件首度揭开了这家当红AI企业收入与算力成本的关键细节。

文件显示，作为130亿美元投资协议的一部分，OpenAI需将20%收入分给微软。按此计算，2024年微软获得4.938亿美元分成，而2025年前三季度该数字跃升至8.658亿美元。若以20%分成比例倒推，OpenAI在2024年收入至少达25亿美元，2025年前三季度至少达43.3亿美元。但据《信息报》此前报道，其2024年实际收入约40亿美元，2025年上半年已达43亿美元。更引人注目的是，CEO萨姆·奥特曼近期透露公司年化收入运行率将突破200亿美元，并设定了2027年实现1000亿美元收入的宏伟目标。

然而亮眼收入背后隐藏着惊人成本。齐特隆分析指出，2024年OpenAI在模型推理运算上投入约38亿美元，而2025年前九个月该项支出已飙升至86.5亿美元。尽管该公司通过微软Azure获得大量训练算力抵扣，但主要需现金支付的推理成本仍持续攀升。此前有报告称其2024年总算力支出达56亿美元，2025年上半年“收入成本”为25亿美元。

值得关注的是，消息人士透露微软与OpenAI存在双向分成机制——除获得OpenAI收入分红外，微软还需将必应搜索和Azure OpenAI服务收入的约20%返还给OpenAI。但泄露数据显示的仅为微软净收入分成，已扣除这部分返还金额。由于微软未在财报中披露相关业务具体收入，双方实际资金往来仍存迷雾。

业内人士分析，若推理成本持续高于营收，即便如奥特曼所言实现200亿美元年化收入，OpenAI仍可能处于亏损状态。这种“增收不增利”的现状，为当前炙手可热的AI投资热潮敲响警钟——当行业领军企业尚难实现盈利，整个领域惊人的估值泡沫或将面临严峻考验。

截至目前，OpenAI与微软均未对泄露数据置评。

中文翻译：

在经历一年的疯狂交易和IPO传闻后，针对OpenAI的财务审查正日益严格。科技博主埃德·齐特龙获取的泄露文件进一步揭示了这家公司的财务状况——特别是其近两年的营收与算力成本。

齐特龙本周披露，2024年微软从OpenAI获得4.938亿美元收入分成。根据其查阅的文件，2025年前三季度这一数字跃升至8.658亿美元。据称，按照此前软件巨头向这家强势AI初创公司投资超130亿美元的协议，OpenAI会将20%营收分给微软（双方均未公开证实该比例）。

但情况在此处变得微妙：知情人士向TechCrunch透露，微软同样会向OpenAI返还约20%来自必应和Azure OpenAI服务的收入。必应由OpenAI技术驱动，而Azure OpenAI服务则向开发者和企业提供OpenAI模型的云端接入。

该消息源还指出，泄露的支付金额指向微软的净收入分成，而非总收入分成。这意味着其中已扣除微软向OpenAI支付的必应及Azure OpenAI服务版权费。由于微软财报未单独披露这两项业务的具体收入，实际返还款项难以估算。

尽管如此，泄露文件仍为观察这家当下最炙手可热的非上市公司提供了窗口——不仅揭示其营收规模，更展现其相对于收入的支出情况。

根据广泛流传的20%分成比例推算，OpenAI2024年营收至少达25亿美元，2025年前三季度至少43.3亿美元，但实际数字很可能更高。《信息报》此前报道称，OpenAI2024年营收约40亿美元，2025年上半年即达43亿美元。奥特曼近期更表示公司年化营收运转率将超200亿美元（此为预测值而非实际营收指引），甚至可能在2027年突破千亿美元。

齐特龙分析显示，OpenAI2024年推理计算支出约38亿美元，2025年前九个月飙升至86.5亿美元。推理计算指运行已训练AI模型生成回应所需的算力。虽然OpenAI历史上几乎完全依赖微软Azure获取算力，但其已与CoreWeave、Oracle达成合作，近期更拓展至AWS和谷歌云。

此前报告显示，OpenAI2024年整体算力支出约56亿美元，2025年上半年“营收成本”为25亿美元。知情人士透露，虽然OpenAI的训练支出大多以微软投资授予的积分抵扣（非现金形式），但其推理支出主要需现金支付。

这些数字虽不完整，却暗示OpenAI的推理成本可能已超过其营收。这一趋势势必加剧从纽约到硅谷无处不在的AI泡沫论调：若行业巨头OpenAI运行模型仍处于亏损状态，那些以惊人估值获得巨额投资的AI企业又将面临怎样的未来？

OpenAI拒绝置评，微软未回应TechCrunch的采访请求。

（注：文末联系信息及会议推广内容已按中文阅读习惯精简处理）

英文来源：

After a year of frenzied dealmaking and rumors of an upcoming IPO, the financial scrutiny into OpenAI is intensifying. Leaked documents obtained by tech blogger Ed Zitron provide more of a glimpse into OpenAI’s financials — specifically its revenue and compute costs over the past couple of years.
Zitron reported this week that in 2024, Microsoft received $493.8 million in revenue share payments from OpenAI. In the first three quarters of 2025, that number jumped to $865.8 million, according to documents he viewed.
OpenAI reportedly shares 20% of its revenue with Microsoft as part of a previous deal where the software giant invested over $13 billion in the powerful AI startup. (Neither the startup nor the people in Redmond have publicly confirmed this percentage.)
However, this is where things get a little sticky, because Microsoft also shares revenue with OpenAI, kicking back about 20% of the revenues from Bing and Azure OpenAI Service, a source familiar with the matter told TechCrunch. Bing is powered by OpenAI, and the OpenAI Service sells cloud access to OpenAI’s models to developers and businesses.
The source also told TechCrunch that the leaked payments refer to Microsoft’s net revenue share, not the gross revenue share. In other words, they don’t include whatever Microsoft paid to OpenAI from Bing and Azure OpenAI royalties. Microsoft deducts those figures from its internally reported revenue share numbers, according to this person.
Microsoft doesn’t break out how much it makes from Bing and Azure OpenAI in its financial statements, so it’s difficult to estimate how much the tech giant is kicking back.
Nevertheless, the leaked documents provide a window into the hottest company on the private markets today — and not just how much it makes in revenue, but also how much it’s spending in comparison to that revenue.
Join the Disrupt 2026 Waitlist
Add yourself to the Disrupt 2026 waitlist to be first in line when Early Bird tickets drop. Past Disrupts have brought Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla to the stages — part of 250+ industry leaders driving 200+ sessions built to fuel your growth and sharpen your edge. Plus, meet the hundreds of startups innovating across every sector.
Join the Disrupt 2026 Waitlist
Add yourself to the Disrupt 2026 waitlist to be first in line when Early Bird tickets drop. Past Disrupts have brought Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla to the stages — part of 250+ industry leaders driving 200+ sessions built to fuel your growth and sharpen your edge. Plus, meet the hundreds of startups innovating across every sector.
So, based on that widely reported 20% revenue-share statistic, we can infer that OpenAI’s revenue was at least $2.5 billion in 2024 and $4.33 billion in the first three quarters of 2025 — but very likely to be more. Previous reports from The Information put OpenAI’s 2024 revenue at around $4 billion, and its revenue from the first half of 2025 at $4.3 billion.
Altman also recently said OpenAI’s revenue is “well more” than reports of $13 billion a year, will end the year above $20 billion in annualized revenue run rate (which is a projection, not guidance on actual revenue), and that the company could even hit $100 billion by 2027.
Per Zitron’s analysis, OpenAI may have spent roughly $3.8 billion on inference in 2024. That spend increased to roughly $8.65 billion in the first nine months of 2025. Inference is the compute used to run a trained AI model to generate responses.
OpenAI has historically almost exclusively relied on Microsoft Azure to provide compute access, though it has also struck deals with CoreWeave and Oracle, and more recently with AWS and Google Cloud.
Previous reports put OpenAI’s entire compute spend at roughly $5.6 billion for 2024 and its “cost of revenue” at $2.5 billion for the first half of 2025.
A source familiar with the matter told TechCrunch that while OpenAI’s training spend is mostly non-cash — meaning, paid by credits Microsoft awarded OpenAI as part of its investment — the firm’s inference spend is largely cash. (Training refers to the compute resources needed to initially train a model.)
While not a complete picture, these numbers imply that OpenAI could be spending more on inference costs than it is earning in revenue.
And those implications promise to add to the incessant AI bubble chatter that has seeped into every conversation from New York City to Silicon Valley. If model giant OpenAI really still is in the red running its models, what might this mean for the massive investments at jaw-dropping valuations for the rest of the AI world?
OpenAI declined to comment. Microsoft did not respond to TechCrunch’s request for comment.
Got a sensitive tip or confidential documents? We’re reporting on the inner workings of the AI industry — from the companies shaping its future to the people impacted by their decisions. Reach out to Rebecca Bellan at rebecca.bellan@techcrunch.com or Russell Brandom at russell.brandom@techcrunch.com. For secure communication, you can contact them via Signal at @rebeccabellan.491 and russellbrandom.49.