Ed Zitron has been relentlessly pursuing the questionable economics of AI and has tentatively identified a bombshell in his latest post, Exclusive: Here\u2019s How Much OpenAI Spends On Inference and Its Revenue Share With Microsoft<\/a>. If his finding is valid, large language models like ChapGPT are much further from ever becoming economically viable than even optimists imagine. No wonder OpenAI chief Sam Altman has been talking up a bailout.<\/p>\n By way of background, over a series of typically very long and relentlessly documented articles, Zitron has demonstrated (among many other things) the absolutely enormous capital expenditures of the major AI incumbents versus comparatively thin revenues, let alone profits. Zitron\u2019s articles on the enormous cash burn and massive capital misallocation that AI represents have the work of Gary Marcus on fundamental performance shortcomings as de facto companion pieces. A sampling of Marcus\u2019 badly needed sobriety:<\/p>\n 5 recent, ominous signs for Generative AI<\/a><\/p>\n Five signs that Generative AI is losing traction<\/a><\/p>\n Could China devastate the US without firing a shot?<\/a><\/p>\n<\/blockquote>\n For a quick verification of how unsustainable OpenAI\u2019s economics are, see the opening paragraph from Marcus\u2019 November 4 article, OpenAI probably can\u2019t make ends meet. That\u2019s where you come in:<\/a><\/p>\n A few days ago, Sam Altman got seriously pissed off when Brad Gerstner had the temerity to ask how OpenAI was going to pay the $1.4 trillion in obligations he was taking on, given a mere $13 billion in revenue.<\/p>\n<\/blockquote>\n By way of reference, most estimates of the size of the subprime mortgage market centered on $1.3 trillion. And the AAA tranches of the bonds on mortgage pools of AAA bonds were money good in the end, although they did fall in value during the crisis when that was in doubt. And in foreclosures, the homes nearly always had some liquidation value.<\/p>\n Now to Zitron\u2019s latest.<\/p>\n Many, particularly AI advocates in the business press, contend that even if the AI behemoths go bankrupt or are otherwise duds, they will still leave something of considerable value, as the building of the railroads (which spawned many bankruptcies) or the dot-com bubble did.<\/p>\n But those assumptions seem to be often based on a naive view of AI economics, that having made a huge expenditure on training, the ongoing costs of running queries is not high and will drop to bupkis. This was the case with railroads, which had high fixed costs and negligible variable costs. The network effects of Internet businesses produce similar results, with scale increases producing both considerable user benefits and lowering per-customer costs.<\/p>\n That is not the case with AI. Not only are there very large training costs, there are also \u201cinference\u201d costs. And they aren\u2019t just considerable; they have vastly exceeded training cost. The viability of AI depends on inference costs dropping to a comparatively low level.<\/p>\n Zitron\u2019s potentially devastating find is breadcrumbs that suggest that OpenAI\u2019s inference costs are considerably higher than they pretend. Zitron further posits that the user prices for ChatGPT greatly subsidize the inference expenditures. Because the reporting on AI economics by all the big players is so abjectly awful, Zitron\u2019s allegations may well pan out.<\/p>\n First, a detour to explain more about inference. From Primitiva Substacks\u2019 All You Need to Know about Inference Cost<\/a> from the end of 2024. Emphasis original:<\/p>\n Over the first 16 months after the launch of Gpt-3.5, the market\u2019s attention was fixated on training costs, often making headlines for their staggering scale. However, following the wave of API price cuts in mid-2024, the spotlight has shifted to inference costs\u2014revealing that while training is expensive, inference, even more.<\/strong><\/p>\n According to Barclays, training the GPT-4 series required approximately $150 million <\/strong>in compute resources. Yet, by the end of 2024, GPT-4\u2019s cumulative inference costs are projected to reach $2.3 billion<\/strong>\u201415x<\/strong> the cost of training.<\/p>\n<\/blockquote>\n As an aside, Gary Marcus pointed out in October that GPT-5 didn\u2019t arrive in 2024 as had been predicted and has been disappointing<\/a>. Back to Primitiva:<\/p>\n The September 2024 release of GPT-o1 further accelerated compute demand to shift from training towards inference. GPT-o1 generates 50%<\/strong> more tokens per prompt compared to GPT-4o and its enhanced reasoning capabilities result in the generation of inference tokens at 4x<\/strong> output tokens of GPT-4o.<\/p>\n Tokens, the smallest units of textual data processed by models, are central to inference compute. Typically, one word corresponds to about 1.4 tokens. Each token interacts with every parameter in a model, requiring two floating-point operations (FLOPs) per token-parameter pair. Inference compute can be summarized as:<\/p>\n Total FLOPs \u2248 Number of Tokens \u00d7 Model Parameters \u00d7 2 FLOPs.<\/strong><\/em><\/p>\n<\/blockquote>\n Compounding this volume expansion, the price per token for GPT o1 is 6x<\/strong> that for GPT-4o\u2019s, resulting in a 30-fold increase in total API costs to perform the same task with the new model<\/strong>. Research from Arizona State University shows that, in practical applications, this cost can soar to as much as 70x<\/strong>. Understandably, GPT-o1 has been available only to paid subscribers, with usage capped at 50 prompts per week\u2026.<\/p>\n The cost surge of GPT-o1 highlights the trade-off between compute costs and model capabilities, as theorized by the Bermuda Triangle of GenAI: everything else equal, it is impossible to make simultaneous improvements on inference costs, model performance, and latency; improvement in one will necessarily come at sacrifice of another.<\/p>\n However, advancements in models, systems, and hardware can expand this \u201ctriangle,\u201d enabling applications to lower costs, enhance capabilities, or reduce latency. Consequently, the pace of these cost reductions will ultimately dictate the speed of value creation in GenAI\u2026.<\/p>\n James Watt\u2019s steam engine was such an example. It was invented in 1776, but took 30 years of innovations, such as the double-acting design and centrifugal governor, to raise thermal efficiency from 2%<\/strong> to 10%<\/strong>\u2014making steam engines a viable power source for factories\u2026<\/p>\n For GenAI, inference costs are the equivalent barrier.<\/strong> Unlike pre-generative AI software products that were regarded as a superior business model than \u201ctraditional businesses\u201d largely because of its near-zero marginal cost, GenAI applications need to pay for GPUs for real-time compute.<\/p>\n<\/blockquote>\n Zitron is suitably cautious about his findings; perhaps some heated denials from OpenAI will clear matters up. Do read the entire post<\/a>; I have excised many key details as well as some qualifiers to highlight the central concern. From Zitron:<\/p>\n Based on documents viewed by this publication, I am able to report OpenAI\u2019s inference spend on Microsoft Azure, in addition to its payments to Microsoft as part of its 20% revenue share agreement, which was reported in October 2024 by The Information<\/u><\/a>. In simpler terms, Microsoft receives 20% of OpenAI\u2019s revenue\u2026.<\/p>\n These numbers in this post differ to those that have been reported publicly. For example, previous reports had said that OpenAI had spent $2.5 billion on \u201ccost of revenue\u201d \u2013 which I believe are OpenAI\u2019s inference costs \u2013 in the first half of CY2025<\/u><\/a>.<\/p>\n According to the documents viewed by this newsletter, OpenAI spent $5.02 billion on inference alone with Microsoft Azure in the first half of Calendar Year CY2025.<\/p>\n As a reminder: <\/strong>inference is the process through which a model creates an output.<\/p>\n<\/blockquote>\n This is a pattern that has continued through the end of September. By that point in CY2025 \u2014 three months later \u2014 OpenAI had spent $8.67 billion on inference.<\/p>\n OpenAI\u2019s inference costs have risen consistently over the last 18 months, too. For example, OpenAI spent $3.76 billion on inference in CY2024, meaning that OpenAI has already doubled its inference costs in CY2025 through September.<\/p>\n Based on its reported revenues of $3.7 billion in CY2024 and $4.3 billion in revenue for the first half of CY2025<\/u><\/a>, it seems that OpenAI\u2019s inference costs easily eclipsed its revenues.<\/p>\n Yet, as mentioned previously, I am also able to shed light on OpenAI\u2019s revenues, as these documents also reveal the amounts that Microsoft takes as part of its 20% revenue share with OpenAI.<\/p>\n Concerningly, extrapolating OpenAI\u2019s revenues from this revenue share does not produce numbers that match those previously reported.<\/p>\n According to the documents, Microsoft received $493.8 million in revenue share payments in CY2024 from OpenAI \u2014 implying revenues for CY2024 of at least $2.469 billion, or around $1.23 billion less than the $3.7 billion that has been previously reported<\/u><\/a>.<\/p>\n Similarly, for the first half of CY2025, Microsoft received $454.7 million as part of its revenue share agreement, implying OpenAI\u2019s revenues for that six-month period were at least $2.273 billion, or around $2 billion less than the $4.3 billion previously reported<\/u><\/a>. Through September, Microsoft\u2019s revenue share payments totalled $865.9 million, implying OpenAI\u2019s revenues are at least $4.329 billion.<\/p>\n According to Sam Altman, OpenAI\u2019s revenue is \u201cwell more\u201d than $13 billion<\/u><\/a>. I am not sure how to reconcile that statement with the documents I have viewed\u2026.<\/p>\n Due to the sensitivity and significance of this information, I am taking a far more blunt approach with this piece.<\/p>\n Based on the information in this piece, OpenAI\u2019s costs and revenues are potentially dramatically different to what we believed. The Information reported in October 2024<\/u><\/a> that OpenAI\u2019s revenue could be $4 billion, and inference costs $2 billion based on documents \u201cwhich include financial statements and forecasts,\u201d and specifically added the following:<\/p>\n OpenAI appears to be burning far less cash than previously thought. The company burned through about $340 million in the first half of this year, leaving it with $1 billion in cash on the balance sheet before the fundraising effort. But the cash burn could accelerate sharply in the next couple of years, the documents suggest.<\/p>\n<\/blockquote>\n I do not know how to reconcile this with what I am reporting today. In the first half of CY2024, based on the information in the documents, OpenAI\u2019s inference costs were $1.295 billion, and its revenues at least $934 million.<\/p>\n Indeed, it is tough to reconcile what I am reporting with much of what has been reported about OpenAI\u2019s costs and revenues.<\/p>\n<\/blockquote>\n So this is quite a gauntlet to have thrown down. Not only is he saying that OpenAI may still have business-potential-wrecking compute costs., but his evidence indicates that OpenAI has also been making serious misrepresentations about costs and revenues. Because OpenAI is not public, OpenAI has not necessarily engaged in fraud; one presumes it have accurate with those to whom it has financial reporting obligations about money matters. But if Zitron has this right, \u00a0OpenAI has been telling howlers to other important stakeholders.<\/p>\n\n
\n
\n
\n
\n
![\"\"](\"https:\/\/www.nakedcapitalism.com\/wp-content\/uploads\/2025\/11\/00-berumda-triangle-scaled.jpg\")
<\/p>\n\n
\n
\n
![\"\"](\"https:\/\/www.nakedcapitalism.com\/wp-content\/uploads\/2025\/11\/Screenshot-2025-11-13-at-2.42.53\u202fPM.png\")
<\/p>\n<\/blockquote>\n![\"Print](\"https:\/\/cdn.printfriendly.com\/buttons\/print-button-gray.png\")
<\/a><\/div>\n<\/div>\n