Media. Hello, and welcome to a very special episode of Better Offline. I'm, of course, your host ed ze trunk. For years, I've been hunting down the core details behind open aiyes costs and revenues, and today I'm going to bring you some of them. A lot of what I say today is going to be reflected in my newsletter,
which I'll link to in the notes. Based on documents viewed by my newsletter, I'm able to report open aiy's inference spend on Microsoft asure in addition to its payments to Microsoft as part of its twenty percent revenue share agreement, which was reported in October twenty twenty four by The Information. In simple terms, that last bit means that Microsoft receives twenty percent of open eyes revenue in addition to whatever
it spends on GPUs and servers. As a reminder, influence is the process through which a model creates an output, which I'll be reminding you of a few times because it's actually kind of important. Now a few notes. I don't have open AI's training spend, nor do I have information on the entire extent of open AI's revenues, as it appears that Microsoft shares some percentage of its revenue from being as well as twenty percent of the revenue Microsoft receives from selling open AI's models on as you.
What I do have, as I've mentioned, is its inference spend. And if you're new to this, like I said, this means all the computation's open ai does when processing requests sent to its services like chat, GPT, and Sora. Now, before publishing, I asked a financial time supporter to help corroborate some of the data in the documents. They reached out to Microsoft and open Ai, who both declined to comment. Now, the following will be a lot of numbers, and it
might be easier for you to read them. However, I'm going to try and make things as easy and clear as possible, because the documents I've seen call into question what we actually knew about open AI's business and the sustainability of said business. To keep things simple, all the years in this piece are calendar years. Microsoft has fiscal years. I'm not going to play that game. It's impossible to follow along with nobody thinks this way. Now we've done that,
let's get to him. According to the document's view by my newsletter, open Ai spent five point oh two billion dollars on inference alone with Microsoft Azure in the first half of calendar year twenty twenty five. This is a pattern that has continued through the end of September twenty twenty five, by which point open ai had spent eight point six seven billion dollars just on inference. Open AI's inference costs have risen consistently over the past eighteen months too.
For example, open ai spent three point seven six billion dollars on inference in twenty twenty four, meaning that open ai has already more than doubled its inference costs in just the first nine months of twenty twenty five. These costs are dramatic and significantly higher than has been previously reported. According to the Information, open AI's computer run models, which I understand to mean inference, was two billion dollars in
twenty twenty four. Additionally, an April twenty twenty five piece from the Information stated that open AI's inference costs for twenty twenty five would be around six billion dollars, or roughly two billion dollars less than open ai appears to have spent through the end of September. I want to be clear as well, I'm just reporting what these documents have said. This is not a statement about the information. They do great reporting. But then there's the issue of
the revenue share. As I've previously stated, the following numbers are based on the revenue share paid to Microsoft as part of its deal with open Ai, where it gives
Microsoft twenty percent of its revenues. According to the documents, Microsoft received four hundred and ninety three point eight million dollars in revenue share payments in twenty twenty four from open Ai, implying revenues for twenty twenty four for open Ai of at least two point four to six nine billion dollars, or around one point twenty three billion dollars less than the three point seven billion dollar number that's
been previously reported in multiple outlets. Similarly, for the first half of twenty twenty five, Microsoft received four hundred and fifty four point seven million dollars as part of its revenue share agreement, implying open AI's revenues for that six month period or at least two point twenty seven three billion dollars or around two billion dollars less than the four point three billion dollars previously reported for that period.
Through September, Microsoft's revenue share payments total eight hundred and sixty five point eight million dollars, implying open AI's revenues are at least four point three to two nine billion dollars through the end of Q three twenty twenty five. To be clear, and I'm going to say this, Microsoft also pays open Ai a cut of bing's revenues under certain circumstances I could not confirm, as well as a cut of about twenty percent of all open ai models
sold through a zore. Just to be clear, Microsoft is the only party that can sell open AI's models other than open Ai. I don't have the details on those payments, like I said, but I am skeptical that they can account for the massive difference between those numbers that have been leaked and the ones in the documents in question. I do not know, nor will I speculate on why
these differences are so distinct. What was important about today was getting you these numbers and shedding light on the differences I see between the story told about open ai and the reality of its spend and potential revenues. You've also I probably noticed that this podcast is a bit of a different tone to the usual no insults, no jokes, haven't called anyone clammy, haven't even said a swear word for the first time in maybe one hundred episodes. The
reason simple. These numbers are serious and seriously different to those reported. Open AI's costs are dramatically higher than previously reported in thought, and based on the extrapolations from Microsoft's revenue share, its implied revenues are also seemingly dramatically lower than we knew. The ramifications of these numbers are severe. Open AI's influence costs are incredibly high, absorbing any and all revenues and seemingly scaling with every increase in chat
GPT's user numbers. As revenue goes up, so does their inference costs. Conversely, if these implied revenues are indicative of the larger financial picture, open ai is not as successful as company as we previously believed. In any case, the reality of the AI bubble is becoming clearer. Inference, the process of creating outputs for a model appears to be
an incredibly burdensome cost. And if these implied revenue is there any indicator the actual business of selling generative AI services and models doesn't really seem to be as good a business as we thought either. It's all looking a little bleak out there. I don't want to editorialize too much because I want this information to sit on its own own self. But it's it's strange being here. It's strange getting these numbers and seeing them myself, and I
have to wonder how things work out from here. I truthfully have no idea, but I do know I'll be happy to do this every week, and I will tell you what happens now. These numbers allow us to kind of see the real picture of the AI bubble, and I have to wonder what other companies look like now that I've seen these numbers. Email me contact me Esitron dots seventy six on signal. If you ever want to tell me anything, do you ever want to show me anything.
I'm always interested to hear, and I'm honored to do this