My takeaway from this is that it's incredibly validating as a business model. Inference is _highly_ profitable. Of course, like any company that has ever tried to grow at breakneck pace, you run at a loss until you "win."
It's more like once you figure out how to make a really good lamp then producing lots of lamps will be profitable. But the lamps are currently suboptimal so we'll be in the red until that time.
And then someone will come up with lamp pro max and you’ll be out of business. You realize why R&D exists in tech companies even though it’s a cost center right?
Watch them flare out like a star… but there is lots of questions re the the return on RnD. Is it worth spending another order of magnitude for only marginal frontier gains?
People keep overlooking the fact that costs for these providers scale along with customer acquisition. Most startups don't have that linear expense. Also, training costs are accelerating to get new models out faster. You don't just "get rid of R&D" costs. I can't actually imagine R&D goes down anytime soon unless you're willing to play third fiddle.
Unless these frontier providers feel some type of squeeze or constraint the Chinese are well positioned to leave the US bag holders of an NVidia bound system. And if you have to wonder how one provider for a critical piece of infrastructure will go, well...
Even if they keep the R&D costs, more efficient inference and 0 Marketing spend also gets you there. Inference is honestly super inefficient at this point, we can do far better than GPUs, push utilisation up, build more efficient datacentres.
I'm really curious about something: how far will you go to support AI? Clearly they'll need to monetize things further, would you still use [whatever AI you are paying for] if the price was doubled? Tripled? Where would you stop and would you stop using AI altogether or would you look at competitors?
I will do nothing to “support” AI. Either it has utility or it doesn’t. I feel no loyalty or duty to help make it work if it doesn’t.
Anyway: Zero, as of right now.
I fully expect to be able to run useful LLMs on a machine I can justify buying for other reasons. I already can on the secondhand kit I own, and I don’t expect the cost-benefit analysis of local LLMs to ever really get worse.
If I ever need to pay for it, it will likely be to shift some of the capacity into the cloud for either business or pragmatic personal reasons (so I can just carry an iPad etc.)
I fully intend my expenditure to be negligible. Because once one realises that outspending others is impossible, only spending minimisation makes sense.
I foresee it potentially making sense for me to move some mature tools off a local LLM to openrouter, maybe. But probably to the same or similar models.
I don’t and won’t support AI. For a while I paid 200€ a month and would have been happy to pay up to maybe 600€. However I don’t want to participate anymore in using such an anti-human technology and industry
I've spent a grand total of $25 on AI ever, so apparently my answer is $25. But I'm not a big time software dev like the rest of you.
When I bought my last GPU, running AI models locally was a consideration though not the only one, and I have it set up but haven't used it much yet. I mostly use the free tiers of ChatGPT or Google to write the occasional script for me. I guess they're going to have to inject a truly unfathomable number of ads to get their money's worth.
I have a feeling my experience is closer to an average persons' than a dev, but it doesn't seem like they'll be able to monetize just from devs even if each one is spending thousands a month.
I'm not a coder but now work way faster than the coder I pay, stuff breaks but it's tenable and it's easier to get things to completion as the harnesses get better.
Don't give up just keep trying you can truly build personally life changing things. Don't look at it purely from a how do I sell this lense, just empower yourself with these tools while the getting is good
For work, it depends, but if I have to spend more than a few hundreds bucks probably I'll start looking for alternatives (local models, Chinese providers, ecc)
PS: I'm in Italy, I guess in several parts of the world these figures are even smaller.
Stretching the analogy, something that gets you from point A to point B for a fraction of the price without the same level of comfort is totally fine for me. For some of my tasks, that means using local models. For others it might mean a frontier-last-year kind of model. That's totally acceptable most of the time. For anything else I guess it's like renting a truck to move; just get the right vehicle as needed and pay the premium.
Agreed. For personal use it's already easily worth $100 a month (to me personally). More probably. For work, it's entirely based on its financial impact for a given role, and for some people/companies it will be worth the cost even at $X thousand per month per seat.
I had codex write a CAN driver for a motor controller in Ardupilot in cpp. It took two fixes that it found and also helped me set the parameters once I had it compiled and installed in the board. I was considering getting an experienced Ardupilot dev to help me because I’m unfamiliar with CAN and cpp, which surely would have been $1000+ and lots of back and forth etc. . It’s such great technology.
6bn seems excessive but despite GPT 5.5 arguably being better than Claude I don't see a lot of adoption of Codex yet.
Some of my coworkers even use Sonnet (the default in Claude Code for the 20 USD subscription) and see no reason to change even though that model is definitely "outdated" compared to current SOTA.
Marketing might help at some workplaces, presumably that are dedicated to Microsoft, for example our network blocks Claude (and DeepSeek) and is slowly rolling out Codex team by team. They should encourage Amazon/AWS to market for them.
You mean the lack of pro-Anthropic/OpenAI comments, who are gambling tokens at their casinos and won't admit that they are very expensive.
This is because people here are quietly realizing that they fell for the "token-maxxing" marketing drive which was complete BS for you to gamble more money on tokens as the big AI labs gave heavily subsidized token prices they cannot afford.
Jevon's paradox does not exist at those companies, but it certainly exists at the Chinese AI Labs at Deepseek, Alibaba, z.AI and Xiaomi.
>This is because people here are quietly realizing that they fell for the "token-maxxing" marketing drive which was complete BS for you to gamble more money on tokens as the big AI labs gave heavily subsidized token prices they cannot afford.
Good callout. All these "trends" in AI were definitely from the AI companies themselves in order to push the sales of more tokens. What's after agent orchestration? Whatever it is, it will involve a big spend.
The scale of the numbers is exceptional, but the shape is pretty typical for a high-growth, scale startup with a big TAM where a winner can take most. And compute, supply constrained as it is for the foreseeable future, is absolutely a moat. I come away from this thinking OpenAI is actually in very good shape given that revenue is growing fast enough that break-even has a clear path without doing anything draconian.
Look, for coding and a lot of other things, AI is awesome.
But the here's the killer. I have a dinky 16gb VRAM card, and that's kind of the sweet spot for the level of AI I actually want. I don't want it doing too much, I'd rather create slowly than have it one shot something that I have to then pore over later.
Feels like a company investing kazillions in, i don't know, air-conditioning or building wi-fi. Yes, it's going to be around, and also no one's gonna need THAT MUCH.
During the internet bubble collapse in the 00s quite some companies went bankrupt. But that's actually a good thing. It doesn't stop progress. It creates new opportunities and new baselines. Same will happen here. AI will not be less or gone or reduced to useless. It will become better , bigger and faster.
If anything this is MORE evidence that the infinite money printer will be coming online any second now! Yep aaaaany second now... OH THERE IT- awww one of you guys wasn't praying hard enough.
I'm a simple guy and I don't understand the "sales and marketing" cost.
I don't like these products. I have several negative opinions on them. To the extent they work and there is a customer base what marketing could you /possibly/ be engaged in? Doesn't the product sort of market itself? Or another way is this a product that you can market to expand your MAUs?
It's so polarizing I can't imagine how that $5.7B is being spent.
I didn't look at the financials but the subscription product is heavily discounted relative to the API pricing and that difference could well be booked as a marketing expense. They also have a string of grant and similar initiatives (like $50M each) that could be marketing. There's a lot of stuff they could assign at least partially to marketing, and it sounds like they spend money pretty freely.
I've seen physical billboards in the Portland, OR area for OpenAI, so I guess that accounts for at least part of it. Not really sure what kind of return they're getting on those but apparently they can just do whatever they want, even if they're losing money.
I cannot consume any content anywhere without being slapped in the face with an unending stream of OpenAI ads and paid plugs. I'd guess most of that money is going directly to Google and Facebook.
They need marketing because they have competition that essentially offers an identical product. Why should a consumer choose openai over anthropic or whatever else there is? The answer is not obvious.
They have a large and rapidly growing enterprise sales organization. If you want to sell to enterprises you need account executives, solutions engineers, forward deployed engineers, etc.
I've seen lots of ads saying I should use chatgpt to plan a workout or give me recipes. Thats apparently the killer app for 95% of the population at this point.
”The company reports over 900 million weekly active users of ChatGPT, though only about 50 million of those are paid subscribers.”
With so many free models available the ai companies are going to struggle to convert active free users to paid.
Discussed yesterday: https://news.ycombinator.com/item?id=48550465
My takeaway from this is that it's incredibly validating as a business model. Inference is _highly_ profitable. Of course, like any company that has ever tried to grow at breakneck pace, you run at a loss until you "win."
Yes, it is like a new era - the startups have huge direct revenue instead of "users" which yet to be monetized.
If these numbers are right, it's actually not that bad. Cut r&d costs and they are mostly profitable.
Yes if you ignore all the reasons why they’re horribly unprofitable, they’re profitable.
R&D costs are hurting profit side and while you can cut that one just becomes irrelevant overnight in this space if you do, hence the problem.
So you’re saying if you cut all the cost centers a company would only have profit centers? If you ignore all the losses you’ll only have profits?
It's more like once you figure out how to make a really good lamp then producing lots of lamps will be profitable. But the lamps are currently suboptimal so we'll be in the red until that time.
And then someone will come up with lamp pro max and you’ll be out of business. You realize why R&D exists in tech companies even though it’s a cost center right?
OpenAI can easily cut R&D costs by replacing engineers with Claude Code
Cut down on the one thing they need to keep themselves relevant in this space?
Watch them flare out like a star… but there is lots of questions re the the return on RnD. Is it worth spending another order of magnitude for only marginal frontier gains?
While you cant discount 100% R&D they are close, agreed
I bet any FAANG spend is mostly R&D.
If it's not materials, not energy or taxes, not manufacturing, not licensing or rental fees, then I can only think of R&D.
People keep overlooking the fact that costs for these providers scale along with customer acquisition. Most startups don't have that linear expense. Also, training costs are accelerating to get new models out faster. You don't just "get rid of R&D" costs. I can't actually imagine R&D goes down anytime soon unless you're willing to play third fiddle.
Unless these frontier providers feel some type of squeeze or constraint the Chinese are well positioned to leave the US bag holders of an NVidia bound system. And if you have to wonder how one provider for a critical piece of infrastructure will go, well...
Even if they keep the R&D costs, more efficient inference and 0 Marketing spend also gets you there. Inference is honestly super inefficient at this point, we can do far better than GPUs, push utilisation up, build more efficient datacentres.
If they cut down on R&D they will be no better than the open source models you can run at cost yourself.
I'm really curious about something: how far will you go to support AI? Clearly they'll need to monetize things further, would you still use [whatever AI you are paying for] if the price was doubled? Tripled? Where would you stop and would you stop using AI altogether or would you look at competitors?
I will do nothing to “support” AI. Either it has utility or it doesn’t. I feel no loyalty or duty to help make it work if it doesn’t.
Anyway: Zero, as of right now.
I fully expect to be able to run useful LLMs on a machine I can justify buying for other reasons. I already can on the secondhand kit I own, and I don’t expect the cost-benefit analysis of local LLMs to ever really get worse.
If I ever need to pay for it, it will likely be to shift some of the capacity into the cloud for either business or pragmatic personal reasons (so I can just carry an iPad etc.)
I fully intend my expenditure to be negligible. Because once one realises that outspending others is impossible, only spending minimisation makes sense.
I foresee it potentially making sense for me to move some mature tools off a local LLM to openrouter, maybe. But probably to the same or similar models.
I don’t and won’t support AI. For a while I paid 200€ a month and would have been happy to pay up to maybe 600€. However I don’t want to participate anymore in using such an anti-human technology and industry
I've spent a grand total of $25 on AI ever, so apparently my answer is $25. But I'm not a big time software dev like the rest of you.
When I bought my last GPU, running AI models locally was a consideration though not the only one, and I have it set up but haven't used it much yet. I mostly use the free tiers of ChatGPT or Google to write the occasional script for me. I guess they're going to have to inject a truly unfathomable number of ads to get their money's worth.
I have a feeling my experience is closer to an average persons' than a dev, but it doesn't seem like they'll be able to monetize just from devs even if each one is spending thousands a month.
I'm not a coder but now work way faster than the coder I pay, stuff breaks but it's tenable and it's easier to get things to completion as the harnesses get better.
Don't give up just keep trying you can truly build personally life changing things. Don't look at it purely from a how do I sell this lense, just empower yourself with these tools while the getting is good
I have made life changing things with it, just not anything so life changing I'd consider paying more than $25. Stingy bastard, I am.
For personal use not more than 30$/month.
For work, it depends, but if I have to spend more than a few hundreds bucks probably I'll start looking for alternatives (local models, Chinese providers, ecc)
PS: I'm in Italy, I guess in several parts of the world these figures are even smaller.
Max 60 bucks a month. More than that and I'd just move to local qwen 35b or some other cheaper model on openrouter.
I’d easily pay multiple hundreds. Possibly a thousand a month.
If I were really forced to.
LLMs provide me about the same value as a car does.
Paying a thousand a month for a car is also very stupid.
Stretching the analogy, something that gets you from point A to point B for a fraction of the price without the same level of comfort is totally fine for me. For some of my tasks, that means using local models. For others it might mean a frontier-last-year kind of model. That's totally acceptable most of the time. For anything else I guess it's like renting a truck to move; just get the right vehicle as needed and pay the premium.
I’d pay thousands a month, if I had no cheaper choices, my productivity is now limited by the intelligence of AI, I’m basically a PM now.
Agreed. For personal use it's already easily worth $100 a month (to me personally). More probably. For work, it's entirely based on its financial impact for a given role, and for some people/companies it will be worth the cost even at $X thousand per month per seat.
That's crazy. Can you provide some examples?
I had codex write a CAN driver for a motor controller in Ardupilot in cpp. It took two fixes that it found and also helped me set the parameters once I had it compiled and installed in the board. I was considering getting an experienced Ardupilot dev to help me because I’m unfamiliar with CAN and cpp, which surely would have been $1000+ and lots of back and forth etc. . It’s such great technology.
I would probably still pay if the cost doubled, but I would also look at competitors, offline solutions, etc
We have benchmarks on our domain and it does there are models that are 2x to 10x cheaper for a small drop in percentage points in accuracy
Never paid a cent, never will pay a cent. I have my principles.
It may put me at a disadvantage when it comes to quickly slop something together? But so far the free-to-use chat bots do as well for my needs.
Zero. It provides no value to me.
Almost 6 bln in sales in marketing? It looks an enormous amount given that they used to have the best models and used to give-aways tokens.
6bn seems excessive but despite GPT 5.5 arguably being better than Claude I don't see a lot of adoption of Codex yet.
Some of my coworkers even use Sonnet (the default in Claude Code for the 20 USD subscription) and see no reason to change even though that model is definitely "outdated" compared to current SOTA.
Marketing might help at some workplaces, presumably that are dedicated to Microsoft, for example our network blocks Claude (and DeepSeek) and is slowly rolling out Codex team by team. They should encourage Amazon/AWS to market for them.
Suspicious lack of pro-AI comments here
their PR department is probably still trying to figure out what narrative the bots should follow for this one.
You mean the lack of pro-Anthropic/OpenAI comments, who are gambling tokens at their casinos and won't admit that they are very expensive.
This is because people here are quietly realizing that they fell for the "token-maxxing" marketing drive which was complete BS for you to gamble more money on tokens as the big AI labs gave heavily subsidized token prices they cannot afford.
Jevon's paradox does not exist at those companies, but it certainly exists at the Chinese AI Labs at Deepseek, Alibaba, z.AI and Xiaomi.
>This is because people here are quietly realizing that they fell for the "token-maxxing" marketing drive which was complete BS for you to gamble more money on tokens as the big AI labs gave heavily subsidized token prices they cannot afford.
Good callout. All these "trends" in AI were definitely from the AI companies themselves in order to push the sales of more tokens. What's after agent orchestration? Whatever it is, it will involve a big spend.
The scale of the numbers is exceptional, but the shape is pretty typical for a high-growth, scale startup with a big TAM where a winner can take most. And compute, supply constrained as it is for the foreseeable future, is absolutely a moat. I come away from this thinking OpenAI is actually in very good shape given that revenue is growing fast enough that break-even has a clear path without doing anything draconian.
Ha, not a problem.
Look, for coding and a lot of other things, AI is awesome.
But the here's the killer. I have a dinky 16gb VRAM card, and that's kind of the sweet spot for the level of AI I actually want. I don't want it doing too much, I'd rather create slowly than have it one shot something that I have to then pore over later.
Feels like a company investing kazillions in, i don't know, air-conditioning or building wi-fi. Yes, it's going to be around, and also no one's gonna need THAT MUCH.
During the internet bubble collapse in the 00s quite some companies went bankrupt. But that's actually a good thing. It doesn't stop progress. It creates new opportunities and new baselines. Same will happen here. AI will not be less or gone or reduced to useless. It will become better , bigger and faster.
Beginning to see why he needed seven trillion dollars.
I'm not surprised
Leaked: OpenAI is a rapidly scaling startup, has economics similar to other startups
If anything this is MORE evidence that the infinite money printer will be coming online any second now! Yep aaaaany second now... OH THERE IT- awww one of you guys wasn't praying hard enough.
I want to see the person who thought they were losing only hundreds of millions
I'm a simple guy and I don't understand the "sales and marketing" cost.
I don't like these products. I have several negative opinions on them. To the extent they work and there is a customer base what marketing could you /possibly/ be engaged in? Doesn't the product sort of market itself? Or another way is this a product that you can market to expand your MAUs?
It's so polarizing I can't imagine how that $5.7B is being spent.
I didn't look at the financials but the subscription product is heavily discounted relative to the API pricing and that difference could well be booked as a marketing expense. They also have a string of grant and similar initiatives (like $50M each) that could be marketing. There's a lot of stuff they could assign at least partially to marketing, and it sounds like they spend money pretty freely.
I've seen physical billboards in the Portland, OR area for OpenAI, so I guess that accounts for at least part of it. Not really sure what kind of return they're getting on those but apparently they can just do whatever they want, even if they're losing money.
I cannot consume any content anywhere without being slapped in the face with an unending stream of OpenAI ads and paid plugs. I'd guess most of that money is going directly to Google and Facebook.
They need marketing because they have competition that essentially offers an identical product. Why should a consumer choose openai over anthropic or whatever else there is? The answer is not obvious.
OpenAI will make fully autonomous killing machines while Anthropic wont.
They have a large and rapidly growing enterprise sales organization. If you want to sell to enterprises you need account executives, solutions engineers, forward deployed engineers, etc.
They are paying influencers to pretend they use LLMs, and discredit Chinese models: https://www.wired.com/story/super-pac-backed-by-openai-and-p...
I've seen lots of ads saying I should use chatgpt to plan a workout or give me recipes. Thats apparently the killer app for 95% of the population at this point.
Don’t forget changing the background of a picture. This alone can triple the GDP.
Practically printing money!!!
That aligns pretty well with a past job. Those two areas were very popular user interests. Third one was cosmetics like skincare routine.
It costs money to get influencers to set up kool-aid stands on their platforms.