OpenAI Losing Money on Every Query Due to Microsoft Hosting  ## Metadata - Author: [[Hard Fork AI]] - Full Title: OpenAI Losing Money on Every Query Due to Microsoft Hosting - Category: #podcasts - URL: https://share.snipd.com/episode/fe339954-35b3-4473-a865-0d045870698f ## Document Notes ## Highlights - **Microsoft Revenue Share Revealed** - Leaked documents show OpenAI paid Microsoft hundreds of millions in revenue share and compute-related payments. - The payments reflect a 20% revenue-share tied to Microsoft's large early investment in OpenAI. Transcript: Jaeden Schafer And he gave a little bit of a glimpse into OpenAI's revenue, their financials, and all of that. And specifically, their compute costs, which a lot of people have been looking at, which is an eye-watering number. So according to his report, he said that in 2024, Microsoft received $493.8 million in revenue share payments from OpenAI. What's interesting is that that was last year. But if you fast forward to this year, that number jumped to 865.8 million. And that's just in the first three quarters of this year. So we have a whole, you know, another third quarter or fourth quarter that hasn't been reported in those numbers yet. And so at something like, you know, $865 million, you can assume it's going to be over a billion dollars that they're going to be paying Microsoft for this year alone. Now, all of this is part of the 20% revenue share that OpenAI agreed to pay Microsoft as part of their previous investment. When OpenAI first came out with Chai GPT, they needed a lot of money. It was getting insanely popular. And Microsoft came in and invested $10 billion at the time. They've since invested a little more. So I think total, or actually, I think it was prior to that, they'd invested $1 billion and a little bit more. So total, I think Microsoft all in is at $13 billion. And with that, I think, especially with that $10 billion that they gave OpenAI, they agreed to a 20% revenue share. So 20% of the revenue OpenAI made, they were going to pay to Microsoft. Neither people have also publicly confirmed that percentage of the 20%, by the way, but this has been reported and leaked by a lot of different people. So this is where things get a little bit more, I guess, tricky. Microsoft already shares revenue with OpenAI. They give them about, and again, it's hard to verify numbers, but from leaked documents, about 20% of the revenue from Bing and from Azure OpenAI service. Now, these are obviously partnerships that they've made with OpenAI because OpenAI's ChatGPT powers Bing right now. And also ([Time 0:00:28](https://share.snipd.com/snip/853e1910-2662-45bc-b74c-466a1acc5d03)) - **Money Flows Are Two-Way And Opaque** - Microsoft also shares revenue back to OpenAI for Bing and Azure OpenAI integrations, complicating net flows. - Public financial statements do not break out these amounts, so leaked internal figures matter for estimates. Transcript: Jaeden Schafer And so it's been said that OpenAI is giving 20% of their revenue because of the big investment back to Microsoft. And then Microsoft is giving 20% of revenue from things like Azure OpenAI Service, which is basically using OpenAI on Azure. So companies that go to the Azure Microsoft's cloud program can do that. Also, in regards to all of this money that they're making, there was a source that was recently talking to TechCrunch about this, and they said that these leaked payments that we see Are talking about Microsoft's net revenue share, not their gross revenue share. So basically what that means is that they're not including whatever Microsoft paid to OpenAI for being an Azure, for their OpenAI royalties, for having OpenAI on their platforms and Charging their customers for it. Microsoft deducts those numbers from its internally reported revenue share numbers, according to the person who leaked some of this info. So what's interesting here is Microsoft never breaks down how much money they're actually making from Bing and Azure OpenAI in their financial statement. So it's pretty hard to get a good estimate on how much they're currently kicking back in all of this. But regardless, all these leaked documents that we see give a pretty clear view into, you know, basically the biggest AI company on the market today and also one of the biggest private Companies, you know, private market companies today that's not public. ([Time 0:02:38](https://share.snipd.com/snip/aea8ab22-b074-407a-bf6e-4edd3dd46f02)) - **Inference Costs Are Eye-Watering** - OpenAI's inference (serving) costs are extremely large and rising rapidly, per the leak. - Reports suggest inference spend rose from ~$3.8B in 2024 to $8.65B in the first nine months of this year. Transcript: Jaeden Schafer Because based off of the really wildly, widely reported 20% revenue share that everyone talks about between OpenAI and Microsoft, we can basically deduct that OpenAI's revenue was At least 2.5 billion dollars last year and this year we can say that it was about 4.33 billion dollars just in the first three quarters of 2025 so it's likely to be more but I mean based off Of those numbers that's what we can see. There are some other reports that came out of the information that said that OpenAI's 2024 revenue was about 4 billion and that its revenue from the first half of this year was 4.3 billion. So we like slightly conflicting, but they're kind of all in the same ballpark. Sam Altman was actually asked about this. And of course, when a CEO is asked about their company, they always want to talk it up and make it seem like it's doing more. When he was asked about their revenue, he said that is quote, well more when he was referring to reports of $13 billion a year. So he was like, oh, we're doing way more than $13 billion a year. But he doesn't say what it is, of course. He said that by the end of this year, they're going to end it above $20 billion in annualized revenue run rate. So what's interesting here, though, he's not saying how much money they're making. He's just giving a projection, an annualized revenue run rate is just a projection on what he believes they'll be able to do in the following year. Right. So saying like, yeah, we're going to end it. We're making way more money than 13 billion dollars. Yeah, we're going to end the year with, you know, over 20 billion dollars in annualized revenue run rates. Like, OK, well, if all of your plans pan out all of your growth strategies work then you could do that so anyways i do think that's interesting and worth taking note of um they also he also Says that the company is going to hit a hundred billion dollars by 2026 in just two years so apparently by the end of the year they'll be set to be making 20 billion and in two years they're Gonna be set to making a hundred billion or they'll be making 100 billion i don't know if that's actually going to happen but this is what he says um according to zytron who leaked a lot Of this stuff and his uh analysis open ai might have well he so he didn't leak it he received the information from the leaker so he said that open ai may have spent about 3.8 billion in inference Last year, and that they've increased that to $8.65 billion in the first nine months, in the first three quarters of this year. ([Time 0:04:12](https://share.snipd.com/snip/244af22c-57ef-4970-8414-379f69299735)) - **Altman's Ambitious Revenue Projections** - Jaeden cites Sam Altman's public claims that OpenAI's revenue run rate will exceed $20B and could hit $100B by 2026. - Altman used projections rather than disclosing precise current revenues. Transcript: Jaeden Schafer Sam Altman was actually asked about this. And of course, when a CEO is asked about their company, they always want to talk it up and make it seem like it's doing more. When he was asked about their revenue, he said that is quote, well more when he was referring to reports of $13 billion a year. So he was like, oh, we're doing way more than $13 billion a year. But he doesn't say what it is, of course. He said that by the end of this year, they're going to end it above $20 billion in annualized revenue run rate. So what's interesting here, though, he's not saying how much money they're making. He's just giving a projection, an annualized revenue run rate is just a projection on what he believes they'll be able to do in the following year. Right. So saying like, yeah, we're going to end it. We're making way more money than 13 billion dollars. Yeah, we're going to end the year with, you know, over 20 billion dollars in annualized revenue run rates. ([Time 0:04:53](https://share.snipd.com/snip/7e902672-ddf8-463b-8e95-692502aef054)) - **OpenAI Is Diversifying Cloud Providers** - Historically OpenAI relied almost exclusively on Microsoft Azure for compute due to large investments and credits. - Recently it diversified to CoreWeave, Oracle, AWS, and Google Cloud, reducing single-vendor dependency. Transcript: Jaeden Schafer We're seeing CoreWeave, Oracle. They just made a huge one with AWS for like $38 billion over the next seven years. They have one with Google Cloud. So they're making a lot more of these. But historically, it's mostly been Microsoft Azure just due to the nature of the $10 billion that Microsoft put in 13 billion total, but 10 billion right after Chattachie launched. And so there's some other reports that put OpenAI's entire compute spend at about $5.6 billion last year. And its cost of revenue at $2.5 billion for the first half of this year. So with all of that being said, there's some sources that are familiar with the matter names, not names. So take them with a grain of salt, but they're talking to TechCrunch about all of this. And they said that while opening eyes training spend is mostly non cash, right? So that means that a lot of it is credits that they're paying Microsoft, like Microsoft gave them a bunch of credits. They're like, here's $10 billion. But a bunch of this is just credits to spend on Microsoft Azure for compute. ([Time 0:06:51](https://share.snipd.com/snip/fcdcc1a9-f293-4ba6-9c3e-915f7d2ef766)) - **Credits Cover Training, Cash Funds Inference** - Training spend may be covered largely by Microsoft credits, while inference costs are paid in cash. - That split means day-to-day serving of users generates significant real cash outflows for OpenAI. Transcript: Jaeden Schafer And they said that while opening eyes training spend is mostly non cash, right? So that means that a lot of it is credits that they're paying Microsoft, like Microsoft gave them a bunch of credits. They're like, here's $10 billion. But a bunch of this is just credits to spend on Microsoft Azure for compute. That's kind of how the deal went. And so most of their spend is, you know, kind of these like tokens or credits that Microsoft already gave them. But there is a lot of spend that is largely cash when it comes to inference. So their training spend is tokens, but their inference, right, what they're using to service all of our requests when we're chatting with chat GPT all day long, those are actually cash They got to pay. ([Time 0:07:33](https://share.snipd.com/snip/3264c888-1a87-4a49-811c-274727018004)) - **Losses Would Rattle The AI Market** - If OpenAI runs models at a loss after these costs, it challenges valuations across the AI industry. - Widespread investor concern may rise if the largest private AI firm isn't yet profitable on serving users. Transcript: Jaeden Schafer And I think the implications of this say that, you know, we're going to hear a lot more people talking about this quote unquote AI bubble. Because if OpenAI is really still running their models at a loss, that this is like a massive deal for the investment world who is spending tons of money on these AI models, the valuations Are absolutely insane. But if OpenAI is not running at a profit, that means a lot of other people are not as well. Of course, OpenAI and Microsoft have both declined to comment on this story. ([Time 0:08:22](https://share.snipd.com/snip/b77e7895-5de1-4cd2-8d65-e0ccc41c75f7))