Leaked OpenAI financials show $38.5B loss and compute burn(runtimewire.com) |
Leaked OpenAI financials show $38.5B loss and compute burn(runtimewire.com) |
How the hell did they spend 5.7 billion in "sales and marketing"?
If this can be trusted, they are selling $1 for ~95 cents, and are spending $3 on R&D. This is... Not an unsustainable business model as long as the money keeps flowing.
OpenAI is selling $1 for $2 and spending ~$5 to make that happen. The question is if they get to selling $1 for $2 long enough to make they're investment back.
The only thing any of them are losing money on is the 200$ a month plans, and you betchya that they're moving as many enterprises to per token pricing as possible rapidly.
If you're not investing in these companies when they IPO and ideally before that if you are lucky enough to be rich, you deserve to not reap what you didn't sow.
The better question is what will that $38B of R&D spending get.
We know OpenAI is forecasting $25-30B revenue for 2026. They will be very close to breaking even at those number.
Given Anthropic has forecast more revenue than OpenAI and we know has spent less on R&D (cite their desperate scramble for compute capacity!) the rumours of them being profitable this year seem very credible.
DeepSeek is somewhere between Sonnet and Opus in capability, for much lower price.
Their $0.87 per million output tokens is one of the few API offerings that is probably subsidized (the break-even price is probably around $2 judging by[1])
I think Anthropic and OpenAI's margins will erode some over time. But I think they are very profitable now on the API prices they are charging, and their margins will remain healthy.
Put it like this: AWS is a very expensive way of getting compute, and there are numerous competitors that are much cheaper.And yet AWS is very profitable.
[1] https://openrouter.ai/deepseek/deepseek-v4-pro#providers
The cost of hardware still needs to dramatically drop for open-weight models to be viable for local usage. Even with the release of things like Nvidia DGX Spark and Ryzen AI Halo, you'd likely want a few of them to run agents in parallel.
Sure, you can use cloud hosted variants of models like DeepSeek etc at API rates, but subscriptions still come out on top for bulk usage. GPT is already tightly integrated into peoples workflows, has wide adoption, has good tooling for developers, etc.
Plus there's nothing stopping them from competing on a price level if they really feel the need. It just means they might burn more cash in the short term.
But a lot of other people are going to happily shell out for Opus 4.8 at way higher prices.
(For tax purposes for most businesses it doesn't matter whether an expense is CORS or not.)
They were first on a few things but the tell is the coherence of the thinking traces of Claude. You have to put a loss on that. GPT 5 series thinking traces are creepy, Gemini thinking traces are disturbing. They both represent forced discontinuities on the policy gradient.
Claude is good at tool use because it's gigantic and well-labeled, but the reason you pay the premium is for a thinking partner, not a tool user.
Claude Code is the cancer that will kill the patient, Boris is the the Kardashian version of Karpathy, with less business sense.
I am not sure I understand.
Considering just four years ago they were a research lab with hardly any revenue at all, and no corporate muscles for earning revenue, I think that is a very impressive number.
(Sure, they're losing a whole lot of money too. Same goes for almost every other hyper-growth company in the history of tech.)
Enterprises are becoming increasingly aware that the best models can be used for planning and then cheaper models for execution - all the way to local models for some tasks.
Add in increasing competition from Chinese models… I’m not convinced this revenue growth is guaranteed.
https://arstechnica.com/ai/2026/06/leaked-financial-docs-sho...
And I believe this is the actual source
> A person familiar with the matter said the large majority of that jump reflected a non-cash accounting charge linked to the company’s previous structure rather than its underlying operations.
> Before OpenAI’s switch late last year to become a public benefit corporation, investors in the company received convertible interest rights rather than conventional equity. Under US accounting rules, those interests were treated as liabilities and periodically revalued as the company’s valuation increased.
> As OpenAI’s worth rose, the increased value of those investor rights created a roughly $30bn charge, added the person. The charge is not expected to recur following the restructuring, they said.
> Stripping out the charge and other non-cash expenses, such as stock-based compensation of staff and computing credits from Microsoft, OpenAI’s losses were $8bn, according to the person.
It's not an issue if you're tracking the cash flow of the business or it's overall viability regardless of ownership structure. These book losses are just recognizing that the business has a higher market value so their ownership dilution commitments reduce the present value of the company more.
A year ago, even 6 months ago, folks would have been still hypnotized by the hype and they would have pulled it off. Today too many people see a burning ship of cash and no moat to justify the burn. The story just isn’t there anymore.
Let's take another example: If OpenAI grows to 10 times their current size and continue spending the same amount on research and development they would be profitable today without any other changes to their organizational structure.
This is shaping up to be a relatively good investment compared to a lot of other companies that have IPO'd in the ~2010 era, the only reason why it looks bad because the numbers are just insane.
The stock market has companies from massive to tiny. Each investor has the right to choose whether or not to invest in any one company. Some might be best served by investing only in profitable blue chips. For others, investing in IPOs is appropriate.
Your ire might be better directed at index companies who change seasoning rules right before a big IPO forcing unsuspecting investors to invest in unsuitable companies. And kudos to indexers like S&P who do not change the rules.
The other side of this are companies like Uber where (if we go by your logic), the public markets made a killing betting on a company that had massive losses. Should Uber also be blocked from an IPO even though objectively it turned out great?
Investing and markets will always decide between risk and reward. The risk of OpenAI is that it will never find profitablity but the potential rewards outweigh that in the current market perception.
In fact, the argument kind of shifts here, OpenAI can afford to IPO at this condition and still expect strong subscription precisely because its OpenAI. If it was some idk cooking appliance company with no exponential future payoff, the market would laugh and reject that IPO.
They can do that to OpenAI too but all signs say they wouldn't. You can still short the stock once it hits public if you really believe in the downfall of OAI in the future.
They could have existed indefinitely as a service layer that was reliant on other companies feeling charitable, like Firefox, but they also wanted to get rich.
Current investors should be left holding this shit bag, better them than retirement funds.
not sure where this constant dooming comes from
Does training of new models go into RnD or cost? And subscription plans' subsidies, are those cost or sales and marketing?
Pretty likely R&D, but obviously would need to be confirmed by OpenAI.
The original reporting includes this:
> The documents revealed how much OpenAI paid Microsoft for services. In the 2025 calendar year, OpenAI paid Microsoft $10.59 billion for “Research and development” expenses. We believe this most likely refers to the cost of training OpenAI’s models.
Given the enormous cost of training, would it be worth training new models then?
I'm not an expert, but is this even an option? I mean models must be refreshed with latest knowldge base periodically even without algo/design improvements, otherwise the lag become too noticeable and it will hurt users and their use-case.
At that point however open weight model providers will start to shine. All eyes on China.
Can't really think of anything else he's been right about, though. I don't think "right" is what he's going for anyway, it's all about that validation and a coherent, testable hypothesis takes a very distant second place.
They are leading the way
Anyway I'm rambling, what Microsoft is doing with AI in enterprise is basically what they did with Teams and similar systems. They provide a platform for it which is good enough that your organisation is going to want it rather than deal with multiple vendors. Not for tech organisations, but for every other enterprise organisation it'll be so much easier to just go this way. I imagine that Anthropic is getting some sort of payment from Microsoft for Cowork, but what Cowork shows is that Microsoft can be completely model agnostic and still sell "top" AI. Especially because they've set cost on a fixed rate that I'm sure they'll increase by 25% every year.
Or do things like the fact that you need some sort of special Agent 365 license for your sysadmins to manage the admin.microsoft part of Copilot which has to do with security policies... Ask me if it was fun doing that agent by agent... It's frankly the most Microsoft thing I've ever seen.
Except it's not true. No one lost $38.5B in a year just to 'hyper-grow' or whatever it means. Uber accumulated ~$30B loss over a decade.
Edit: I read it wrong. The loss was mostly caused by one-time event[0]:
> Before OpenAI’s switch late last year to become a public benefit corporation, investors in the company received convertible interest rights rather than conventional equity. Under US accounting rules, those interests were treated as liabilities and periodically revalued as the company’s valuation increased.
It looks like that OpenAI is actually quite in line with other companies that lost money to grow.
[0]: https://www.ft.com/content/e15b0d7e-ff6b-4f16-ba7a-4068feddb...
Growth means some inefficiencies, but their expenses are largely around commodities like electricity and data centers not a sudden army of salespeople. They also got 150M 11 years ago and 1 billion 7 year ago, they where quite large in 2022.
Basically you don’t get better at writing checks to your local utility which limits how much they can control costs.
I had to look that up, you're talking about investment there, not earned revenue.
The 150M was their initial funding (actually 130M I think https://www.clay.com/dossier/openai-funding)
The 1B was from Microsoft in 2019: https://openai.com/index/microsoft-invests-in-and-partners-w...
In 2022 they only had 335 employees (according to various internet searches but I can't find an original source for that number.) I can't find credible numbers for revenue from the GPT-3 API, which did have some usage - GitHub Copilot started charging a subscription fee on June 21, 2022 - https://github.blog/changelog/2022-06-21-github-copilot-is-n... - and that was running on the OpenAI Codex model so presumably OpenAI had some revenue from that.
That doesn't mean anything. There are examples to make both ways. E.g. WeWork
Look at how a utlity works, in setting price specifically, for things that are considered a public good. The story is not about how much profit or revenue they make. Its about how do you keep it afloat and expanding in the coming year. Thats it.
In the case of AI the marginal cost of the next token is not zero, and is in fact probably not going down much with volume, if at all.
So I'm not sure one can argue that scale will solve everything. It's very much like the old adage "we lose money on every sale, but make it up in volume".
No you don't.
They gave up on video because three separate Chinese companies were kicking their ass (and for cheaper).
Google has a better image model in the majority of cases. Much faster, too.
Claude Opus and Fable are like a billion times better. It's not even funny. Codex can't do Rust at all.
What does that leave them? Ads in ChatGPT? I've started to just rely on Google search blended with Gemini answers now because it's faster and doesn't spit out a 20-page essay of useless effusive prose.
Open source models will eat them from the bottom.
Will those enterprise contracts be renewed in a market full of alternatives?
There's nothing sticky about this company.
They're making a necklace with Jony Ive though, I guess?
Not always. A couple months ago (before ChatGPT Images 2) I tried various prompts on both Google's Nano Banana or whatever and ChatGPT.
"Capybara riding a tricycle. It has 7 tentacles instead of legs"
Google got the number of tentacles completely wrong: https://i.postimg.cc/nzY30y7X/Capybara-Gemini-Nano.png
and after some additions like spotted fur and multicolored tentacles, it was no contest:
ChatGPT: https://i.postimg.cc/02c2LrxV/Capybara-Chat-GPT-before-Image... (although there's still kinda 1 extra tentacle)
And Google still seems to have that odd choice of a European plaza/square/cobblestone street background for everything.
> Claude Opus and Fable are like a billion times better.
NOT at ALL: https://i.imgur.com/jYawPDY.png
Subscribed to Claude Opus for 2 months, with a few months gap between subscriptions to try different versions.
The UX/UI around Anthropic's products was excruciatingly annoying, right from the payment process, and Claude's AI was often hilariously dumb and "trying too hard", constantly full of "oops, you're right" backtracking and often borderline dangerous.
I tried Claude and ChatGPT Codex side by side on some tasks, with the same prompts. Each time, my confidence in Claude fell.
I've been subscribed to the $20 ChatGPT plan for more than 1 year, and this month, I am trying the $100 plan for 1 month.
ChatGPT Codex has been actually helpful and made me more productive enough that I can't imagine going back to coding without it.
There is something that has fundamentally changed (or broken, depending on your perspective) with the valuation of American tech companies. They've always traded at a premium, but the pandemic and the encroachment of the monopolists has turned the earth sour.
Similarly, SpaceX has already brought the cost of getting things to space down by a couple of orders of magnitude, and Starship is rapidly progressing with the potential to bring them down a couple more. The aspirational goals there are being able to get things to space on the order of $10-$20/kg. That would dramatically reshape not only space but even transport as we know it, very likely in a way analogous to how the ability to quickly send a 0 or 1 signal long distances for cheap reshaped the world in ways that would be essentially impossible to predict prior to its happening.
I'm bearish on the LLM revolution and bullish on the space one, which generally seems to also be the market consensus.
It doesn't cover launch costs directly, but here's a bearish take on reusability of the second stage:
https://mceglowski.substack.com/p/how-should-we-think-about-...
.. who are they gonna sell to if people don't even have money to buy? We live in circular economy... everyone's dependent on someone or the other... you take one leg out of this, and the whole thing stops. UBI won't work either because it will lead to runaway inflation and extreme levels of invasive control over people's lives and what they can and cannot do.
I hope no one here is naive enough to believe that AI would actually be used for general welfare of people.
At least that's what I remember from the 00s.
You can debate the assumptions, but it isn't witchcraft. The math is simple.
If you limit yourself to simple math, than you get this result.
Assuming linear growth into a market with competition and an unclear ability to absorb that growth is a gamble.
That seems unlikely.
I think it's more about contrasting it with world economy as a whole for a reality check.
I think this is the key takeaway for the future of AI. Give tech a few years to catch up and we will likely have the functionally equivalent to today's models running on consumer grade hardware. From there it will explode, where "it" is how we use and interact with computers. AI will be integrated into just about every workflow.
The business case in this future would be to sell the trained models to end users. The investment would be shifted towards the training of models and delivering updates, with revenue coming from model licenses, upgrades and cloud services for tasks that exceed the local capabilities.
If people don’t get work they stop buying things. If people don’t buy things, companies don’t make money. If companies don’t make money they won’t buy — or have a reason to buy —- AI companies’ services.
You cannot eliminate a significant chunk of labour cost without causing demand collapse. People saying that there is money in eliminating labour costs at that scale are not doing the whole of the calculation.
But without labor the entire economy also collapses into a singularity beyond which nothing really makes sense anymore, so there's that.
building a Rube Goldberg machine on Chinese models might work okay, but it will be brittle, and is unlikely to work as well as the latest and greatest model from OpenAI
the demand for intelligence is nearly limitless
UPDATE: Also bad news, you need to let go of all of sales and marketing and G&A. And THEN it's a 50% margin business.
If they get rid of R&D, then someone else will make a better model and we will all switch to using that model.
If sales & marketing covers subsidies and bribes then they cant get rid of that either. Get rid of the bribes and they will be shut down. Get rid of the subsidies and we will all switch to someone cheaper.
Unless they increased their spending even more, "all they have to do" is cover 2025 with the 2026 revenue?
Or is this more like a bonfire that requires more fuel to keep burning?
Huh? Where did $30 billion go?
They might not have spent $30b but they likely valued their asset base at >>> $30b+ and had to adjust that at the time of converting to for profit, is how I read it.
“One time non-recurring” is also just accounting double speak that lets executives cover up dumb stuff while sounding plausibly OK.
The big AI IPOs this year will mark the high point in the bubble IMO, with retail FOMO ensuring that they pop like SpaceX has.
After that I expect them to crater, just like all those Spacs and Blockchain startups.
> Revenue: $13.07 billion
> Cost of Revenue: $7.5 billion
In other words generating tokens is actually a profitable business even for the frontier models. It's best to IPO when it's the case.
[0]: https://www.ft.com/content/e15b0d7e-ff6b-4f16-ba7a-4068feddb... [1]: https://www.wheresyoured.at/exclusive-openai-financials/
Your math is saying an apartment building is profitable because rent exceeds utilities and other direct expense but ignores the mortgage. Real estate run with that math goes bankrupt quite quickly and this is essentially the same problem Open AI has.
[0]: https://openai.com/index/accelerating-the-next-phase-ai/
I wish I could believe this but lighting giant stacks of cash on fire is, unfortunately, no longer a disqualifying event for tulip buyers.
I wouldn't bet they will have a successful IPO, but I also wouldn't bet they won't, it will be almost entirely vibes based at time of launch.
Given that, no, this question is still open.
You can just run his content through AI to get a more balanced flash take. Example:
> Zitron repeatedly describes OpenAI as having "removed" costs — $3.74B in 2024, $17.87B and $3.95B in 2025 — via "net loss attributable to noncontrolling members capital," and says "it's unclear what this means." This is standard consolidated-statement mechanics, not a maneuver. When a parent consolidates entities it doesn't wholly own, the slice of losses belonging to other equity holders is split out as "noncontrolling interests." Nothing is removed or hidden; the total loss is unchanged, it's just allocated. Framing it as OpenAI "lowering" its loss "by removing costs" implies something sketchy where there's only routine GAAP. Saying "I will not speculate further" while leaving that insinuation hanging is the rhetorically convenient version of restraint.
For what it's worth, I think AI is a "bubble" and am not convinced at the long-term sustainability or viability of many of these companies but that doesn't mean that every armchair critic actually has the financial expertise to make accurate arguments.
I mean, his whole sensationalized 8X headline is based on a non-cash conversion charge, which is literally the biggest straw man you can find in the financials. He chose it because he's editorializing even as he leads his post with "I am not going to do very much editorializing". Hilarious.
and 7.81 billion in R&D from last year. I don't know how long it took to build the weights for the current model, or exactly how much that costs, but it's certainly more than zero days and zero dollars.
I also doubt that OpenAI could set that R&D expense to zero and survive without an agreement from Anthropic that they'll do the same... so that R&D expense can't be ignored when figuring up the total cost of the current model.
That said this is obviously not a strategy, but rather an observation that this is not a flawed, impossible concept.
So I don’t see a company in immediate danger of collapsing, but I also don’t see a great investment at that valuation.
Just 10x your revenue without increasing R&D cost? where's this 1 magic trick and why can't every company just use it?
They would show a profit for 1 year and then become completely irrelevant.
The API business throws a massive model that by definition can't be inferred efficiently because nothing can across 4 different compute substates, at a problem that DSv4 nails at or near 100% while leaving most of the actual unique value of Claude on the table.
Claude should be in your house and car and your kid's classroom and shit.
Having it write tail -n5?
That's because Anthropic's A-Team is Meta's C-Team. Hell, I fired some of their stars myself.
Humans.
The legibility of thinking traces has gone down, but mostly this nets out to spilling more XML because their website is a vibecoded heap
There are arguments for and against this - awareness is an issue for startups, but most large companies continue to market. It is true that it is fairly easy to change the amount a company spends on marketing.
Either way, as the parent says - provided the investors understand it there is nothing weird about doing it this way.
(To explain: It's basically saying they are spending $X now to get them as a customer which will yield a lifetime value which is some multiple of $X. That's a very valid thing to do - see the whole field of cohort analysis for SAAS)
If the marketing and promotions and discounts stopped tomorrow, that revenue will drop.
So if there is real cost involved things start to look lot worse and might not be overcome. OpenAI is unlikely to be exception for me.
Also, what are they calling "R&D" exactly? If it is training new models, which needs to be done almost constantly and means spending billions on energy and newer GPUs, then it's not really R&D, but rather operating costs.
How many serious, large AI players are there? Google, OpenAI, Anthropic, and who else exactly?
At least one of them will probably win. And winning here means billing almost all companies for AI and automation, consumers, perhaps robotics and research.. and that potential earning is massive.
So yes, I will pay 10 times the worth for the stock now, but paying 1000 per stock for a chance of owning all that future profit is not that outrageous.
That should make it hard to bank in on future growth with any single AI company: what I believe is happening is investors jumping in on the short term gains train.
Henry Ford said he needed to pay his workers enough that they could afford his cars.
since when? do you think their R&D in 2026 is the same as in 2022?
It's more efficient to do the opposite on a constrained platform. Run agents in parallel using a single model, then round-robin among models for cross-checking purposes. (The makers of local inference engines are dropping the ball by not making batched inference a first-class citizen of that workflow. It's not just useful for vLLM and SGlang.)
The API GLM 5.2 launched last night with several providers on OpenRouter. I had a short conversation, didn't get to test much, but initial impressions "the vibes were good".
Cost of revenues Research and development Sales and marketing General and administrative
Take it up with the SEC I guess?
[1] https://s206.q4cdn.com/479360582/files/doc_financials/2026/q...
On top of that, oil wells decline, slowly but surely, just like... customers churn.
If you spend $100 mill on sales and marketing and get zero customers, I'd agree it should expensed. If you get a bunch of customers, it's hard to argue against the capex treatment idea.
They have to apply the rules consistently otherwise it wont make any sense to have these indicies right? Spacex got rejected in one of the big indicies because it was not profitable while it got accepted in another. You can buy from both as an investor or a fund manager. If you think one of them is wrong, it is in your interest to switch. If you don't switch then you indirectly don't see anything wrong with it.
Going back to the point, if we want a free market, it has to be free for all. Spacex can go to 0 tomorrow and cause markets to implode, it can also surge 10x over the next 5 years causing everyone in the market to get richer or it can just drag for 10s of years in the same level.
But you and I do not know the future. We cannot block a company from IPOing just because we think it will fail in the future. If it meets all the existing regs, it should be allowed to IPO. If you want to change the regs, thats a different discussion but one i fundamentally disagree with personally. I think the market should be able to take risks and explore potential otherwise we will stagnate.
That said, in many ways 335 employees is the midpoint between 3 employees and 30,000 employees. The CEO can’t keep track of everyone’s names and what they’re doing, you need layers of management, HR, etc. It’s not really a simple exponential function but 335 to 336 is way more automated than going from 3 to 4.
[1] - https://forum.nasaspaceflight.com/index.php?board=72.0
Like agriculture. That market has grown significantly through time, even though it's shrunk dramatically as a % of GDP.
AI cannot make money as an alternative to large scale employment, because essentially all the clients of those AI businesses will see demand for their products and services collapse. AI bots don’t go to In-N-Out Burger or Disneyland.
Anything else is fantasy maths, albeit commonplace fantasy in the AI industry at the moment.
I've never heard this before. I thought UBI would be very freeing and without much control. If it is universal then there needs to be no control of who gets it or not. What am I missing?
And in cases where the strain on resources is significant, you could easily see things like efforts to restrict the fertility of the consumption class which would enter into the domain of defacto eugenics. And from all of these sort of issues you're going to see a conflict arise between the two classes, but one holds all the power. It's not going to be pretty. FWIW, a decade ago I was a huge advocate for UBI, but my outlook on the realities of political leadership, and it's probable inescapability, has changed my opinion over time.
Your fight against the data centers is misguided. All so they can be built somewhere else so what you have a few extra years of getting paid to fill out a spreadsheet or something? You're applying pressure on the wrong side of the equation.
Nothing, it is nonsense. UBI is just expanded social welfare in Europe. There are some checks that you are not abusing social welfare (i.e. living in a huge house while taking housing subsidies) but with UBI these checks are nonsensical from that first letter U = Universal. There are no checks by default.
R&D is a leading expense, a good portion of that is probably R&D for 2026 models
To answer that question you have to take into account the cost to produce the thing that inference uses. If you don't, then that's like claiming that the total cost of a car is the cost to keep it on a dealer's lot until it's sold.
"Figuring out how much R&D adds to the total cost of a thing" absolutely isn't a new problem. And given that models seem to get supplanted every year, it's not like you're gonna be able to spread those R&D costs out very much.
