OpenAI is burning cash

OpenAI is printing money. I'm spending

thousands a month myself. I'm sure many of you are spending crazy amounts as well, which is why they're making$20 billion a year. When I looked at these

numbers, I felt kind of insane. They

have 800 million users, but only 5% are paying. They do 13 billion in recurring

paying. They do 13 billion in recurring revenue every year. That's about $325 per annual paying user, or $27 a month per paying user. 70% of their revenue is coming from subscriptions. The rest is

API, which means that if you're using a service like T3 Chat, you're being counted in the 30%. The vast majority of their money is coming from people using chat GBT directly, which is kind of

crazy if you think about it. But the

crazy numbers are these ones. They've

lost $8 billion for the first half of 2025, and they're looking like they'll have a 20 billion a year run rate. That

means they're spending $3 for every dollar they make. What the hell is going on at OpenAI? You'd think that a company worth $400 billion or whatever their most recent valuation was would be

making a lot of money and would be very very profitable. Right. Right. Nothing

very profitable. Right. Right. Nothing

is that simple, especially in the modern world of startups, AI, and just the chaos that is modern economics. These

numbers should seem insane because they are. But if you look at these numbers in

are. But if you look at these numbers in the framing of this year, you're not looking at them right. The way we have to look at these numbers is a good bit different. We have to think about this

different. We have to think about this in terms of not how much money they're making and losing this year, but how much money they're spending to make way more next year and the year after and the year after that. All of this is an

investment into future potential profits or I guess in this case nonprofits.

Good pun. I remember the days where OpenAI was a proper nonprofit. It's been

a while, but you get the idea.

Regardless, when I saw this post, I had a crazy gut reaction of, "Holy [ __ ] those numbers are insane." But I realized that most people will. And

after I realized why these numbers could be insane and thought it through a bit more, I realized that a lot of people probably don't have the context I have.

Maybe you haven't seen how others talk about these things. Maybe you haven't seen this surprisingly good interview with Daario where he goes in depth on how this stuff works. Maybe you're just not familiar with the dynamics of how

these earlier stage businesses think about their long-term profits. There's a

lot of reasons why these numbers should scare you, but I want to try and break down why they don't scare Sam or the team over at OpenAI or most importantly the investors that just handed them like

$40 billion more. Kind of insane. All of

that said, I don't got 20 bill to spare.

So, we're going to have a quick break for today's sponsor and then dive in.

When I work with a new brand, I'll often ask them for examples to use in my ads with them because I want to show you what they actually look like to use in production. Grepile is an AI code review

production. Grepile is an AI code review app and it turns out they're being used by Post Hog, another one of my sponsors that happens to be mostly open source. I

figured this is a great opportunity. So,

I asked the Grapile guys, hey, can you send me some examples of Post Hog getting value out of Grapile? I figured

they'd send like two or three. I did not think they would send over 150 of them.

Turns out AI code review can actually be really really helpful even to super experienced fast-moving teams like the Post Hog team. Greile will leave comments on GitHub just like a normal code reviewer would both summarizing the

PR which is really nice. But my favorite little thing is they leave a confidence score telling you how they think the PR is overall like is this safe to merge or not and why. Surprised how much value I've already gotten out of this feature.

But this wouldn't be useful if the suggestions weren't. Thankfully the

suggestions weren't. Thankfully the suggestions are really really good.

Here's an example where it caught the fact that this Reddis connection was never being cleaned up. This can cause your Reddus connection pool to just fail outright. That's really bad. This

outright. That's really bad. This

probably prevented an outage. That's

super cool. The CTO of Brex has come out and said pretty much the same. We've

tried more code review tools than I can count. Grappile outperforms them all by

count. Grappile outperforms them all by a mile. It's honestly the only AI

a mile. It's honestly the only AI reviewer that doesn't annoy the [ __ ] out of me. That's kind of been my

of me. That's kind of been my experience, too. Grappile doesn't feel

experience, too. Grappile doesn't feel annoying. It just feels useful. It's

annoying. It just feels useful. It's

super easy to set up. It's super fairly priced. And it's also free for open

priced. And it's also free for open source and half off for startups. If you

never used AI code review before, they're the one to try. And if you have and didn't like it, they're still worth a shot because the quality of what you get is significantly better than anyone else. Check them out now at

else. Check them out now at soyv.link/grapile.

soyv.link/grapile.

So the first piece I want to talk about here is the $3 to $1 ratio because that sounds like a business that will never

succeed. So if I don't know, GPT5

succeed. So if I don't know, GPT5 costs $1.25 25 per mill tokens in and then $10 per mill tokens out. Does that

mean if I do a million output tokens, it costs them $30? No. Does not mean that at all. Not just because employees cost

at all. Not just because employees cost so much money, not just because training costs so much money. Not just because the GPUs cost so much money. It's some

combination of these things plus the fact that the training is just such an egregious upfront cost. In an interview in August, Sam talked about this. Thank

you, Simon, for posting it on HN. And to

Bot Cooper for finding this before I talked about it. Most of what we're building out at this point is the inference. We're profitable on

inference. We're profitable on inference. If we didn't pay for

inference. If we didn't pay for training, we'd be a very profitable company. I think this will make even

company. I think this will make even more sense when you hear the quotes from Daario because right now the research cost is so much higher than the income.

It just sounds insane. There's kind of like two different ways you could describe what's happening in the model business right now. So let's say in 2023 you train a model that costs $und00

million and then you deploy it in 2024 and it makes $200 million of revenue.

Meanwhile, because of the scaling laws in 2024, you also train a model that costs a billion dollars. And then in 2025, you get $2 billion of revenue from

that $1 billion and and you spend $10 billion to uh train the model. So if you if you look in a conventional way at the profit and loss of the company, um you

know, you've you've lost $100 million the first year, you've lost $800 million the second year, and you've lost $8 billion in the in the third year. So it

looks like it's getting worse and worse.

If you consider each model to be a company, the the model that was trained in in in in 2023 um was was profitable.

You you paid a hundred million and then it made 200 million of revenue. there's

some um you know cost to inference with the model. Um but you know let's just

the model. Um but you know let's just assume in this cartoonish cartoon example that even if you add those two up you're you know you're kind of in a good state. So every model was a company

good state. So every model was a company the the model is actually you know in in this in this example is actually profitable. What's going on is that at

profitable. What's going on is that at the same time as you're reaping the benefits from one company, you're founding another company that's like much more expensive and and requires

much more upfront R&D investment. And

and so the way that it's going to shake out is, you know, this will keep going up until the numbers go very large, the models can't get can't get larger and, you know, then it'll be a large very

profitable profitable business or at some point um you know, the models will stop getting better, right? the, you

know, march to AGI will be halted for some reason. Uh, and then perhaps it'll

some reason. Uh, and then perhaps it'll be some overhang. So, there'll be a one-time, oh man, we spent a lot of money and we didn't get anything for it and then the business returns to, you know, whatever scale, whatever scale it was at.

>> You get the idea. I think a lot about this particular post, mostly because the replies were so dumb that they made me lose a lot of faith in like humanity and

the intelligence of the average person.

bought a cow for $900, sold it for 1,200, bought it back for 1,300, and then sold it for 1,600. How much did you earn? Do the math in your head quick so

earn? Do the math in your head quick so you can know if you're in the stupid side of the world. Thankfully, my chat is smarter than average, apparently, cuz

you all got the number right. It's 600.

The number is very obvious. Plus 300,

1200, 1300US 100, 1300, 1600 plus 300.

So 500. They apparently only made500 because they did numbers that don't make sense. Another person saying 500.

sense. Another person saying 500.

Where'd you get the $100 from? Where'd

you get the $100 from? People just being [ __ ] stupid. People are doing 1,200 - 900 plus 600 - 3000 which is completely wrong. You earn $300, but you're losing

wrong. You earn $300, but you're losing $100 because you're buying, not selling.

That's not how it works. You're not

losing the money when you get the asset.

You're transferring the money in trade for an asset. You don't lose money because the difference is there. And

this is what I think is happening when people are looking at these numbers with these models. When they see a thing cost

these models. When they see a thing cost more, that means you lost money.

They don't understand what profit is. If

Anthropic spends 100 mil on a model and then the next year they make 200 mil from that model, they profited on that model. My poor editor is in chat raging.

model. My poor editor is in chat raging.

It's okay. I just have to breathe. I'm

glad that these things exist in the internet. Otherwise, LM would be

internet. Otherwise, LM would be smarter. Yeah, this this stuff drives me

smarter. Yeah, this this stuff drives me up a wall, but it's I meant to bring out this kind of silly example to emphasize how bad people are at understanding

these things, but also why our gut feel when we see the numbers that started this whole video is so bad. Because when

you think of it this way, it is pretty bad. But if you think of it as different

bad. But if you think of it as different cows that cost different amounts of money that sell for more money later, it makes much more sense. I can't believe I actually feel the need to diagram this

out, but that's where my IQ is at, I guess. So, if in 2021 you spend $900 on

guess. So, if in 2021 you spend $900 on a cow, so 900 for cow one, and then in 2022 you spend another $1,300 for cow 2.

Now you're at -2,100. So, you've lost a lot of money at that point. Like, your

business sucks. You're really bad at what you're doing. if you're at that position, right? So then 2023, you make

position, right? So then 2023, you make 1,200 selling cow 1. You're still

negative 100 right now. But then in 2024, you sell cow 2 for 1,600.

Now your business is doing quite well.

God damn it, CJ. Just got in. Why are we talking about Moo Tools? Please add a cow Tools reference. Very important.

Appreciate you guys. The point I'm trying to make here is that these businesses are investing in next year's profits in next five years profits. So

the cost right now relative to the revenue right now makes no sense because they're spending right now to make more profit next year. And this is why it makes so much sense to buy the third

cow. Actually though, the core point I'm

cow. Actually though, the core point I'm trying to make here I think makes sense.

But the only way this makes sense is if your profit is growing relative to the spend. And OpenAI's revenue growth has

spend. And OpenAI's revenue growth has been kind of [ __ ] crazy. For the

first half of 2025, OpenAI generated 4.3 billion in revenue, which is 16% more than they made all of 2024. That's an insane rate of growth.

2024. That's an insane rate of growth.

Take your salary that you made last year, however much money you made, add 16%. And imagine you made that in 6

16%. And imagine you made that in 6 months. That's kind of crazy, right?

months. That's kind of crazy, right?

Like that's just unfathomable. That's

like greater than 200% year-over-year growth for their income. And it's

continued to grow even crazier since.

It's also fun to look at the ratio here cuz they made 4.3 bill in rev and it cost them 6.7 bill to do research. So

those numbers aren't actually that far off. But they've continued to ramp up

off. But they've continued to ramp up their spend significantly, which is why this is getting so crazy. They've also

made some pretty crazy commitments. The

big one that everyone's been talking about, as they probably should, is that they plan to spend $1 trillion over the next five years and have already made the commitments to do that.

Really beautiful website you guys have here, Yahoo Finance. OpenAI is printing money right now. Kind of bold considering how much they're losing, but they're pulling in roughly 13 billion in annual revenue with 70% coming from

everyday people paying 20 bucks a month to chat with AI. Pretty well when you consider they have 800 million regular users and only 5% are paying. Breaking

in billions, though it may be, OpenAI is also committed to spending over 1 trillion over the next decade. The

company has recently locked in deals with more than 26 gawatt of computing capacity from Oracle, Nvidia, AMD, and Broadcom. Infrastructure that'll cost

Broadcom. Infrastructure that'll cost vastly more than what's coming in. Going

from 13 bill a year income to somehow canceling out that trillion spend is crazy cuz that's like even over the 10

years that's 100 bill a year average.

That's brutal. That's another 10x away from being even close to profitable. So

to bridge the gap they're getting creative. A five-year plan includes

creative. A five-year plan includes exploring government contracts, shopping tools, video services, oh, Sora, consumer hardware, and even becoming a computing supplier itself through

Stargate as the data center project.

Growing number of these businesses need the map to work out. So, of America's most valuable companies are now leaning on OpenAI to fulfill major contracts. If

OpenAI falters, it could potentially destabilize the broader US market. Yeah.

If OpenAI struggles immediately, Nvidia collapses and the S&P 500 is only making money right now because of the top 10 companies. If you axe the top 10 out of

companies. If you axe the top 10 out of S&P 500, it's down right now. I'm not

finding the numbers I'm looking for, but I've seen enough to roughly do the math here. I would say very very generously

here. I would say very very generously that every year OpenAI's costs go up by around 3x and their profit goes up by

around 4x.

If this maintains, if they continue to have their cost 3x and their rev 4x and the current numbers that we're looking

at are 13 bill in and 20 bill lost. So

that's 20 bill as their cost. So you

have to add these two numbers to get the actual value. So right now they're going

actual value. So right now they're going to make 13 bill in the year and they're going to spend 33 bill in the year.

Assuming these numbers continue, they're going to profit 42 bill. No, it's 52 bill in 2026 and they're going to spend

99 bill. Like oh my god, they went from

99 bill. Like oh my god, they went from losing 20 bill a year to losing over 40 bill a year. That's terrible. They're so

[ __ ] They're going to burn. This is

not going to work. 2027 comes, they're going to profit. Assuming that the growth maintains, profit 28 billion,

spend 297 billion. Oh my god, their cost is more than double how much money they made last year. There's no world where this company's ever going to profit. We

can keep going through this. You know

what? I'm going to be super lazy. We're

going to use a good chat app. Keep

calculating until the company is profitable.

2028 profit 812 billion spend 8.91 2029 profit 3.2 trillion spend 2.6 trillion.

Suddenly we're profitable.

Kind of crazy when you think about it this way.

Okay. I should have said revenue, not profit. That's fair. I screwed up.

profit. That's fair. I screwed up.

Thinking's hard. Okay. The growth won't be linear. Theo, no [ __ ] It's not

be linear. Theo, no [ __ ] It's not linear.

3x is not linear growth.

Someone should This is a form of exponential growth. Like, I'm sorry. I

exponential growth. Like, I'm sorry. I

don't know how to like more simply explain that. The point I'm trying to

explain that. The point I'm trying to make here is simply that the numbers work out really well in not that long of a timeline. We'll quickly get to the

a timeline. We'll quickly get to the point where they're spending a trillion year quite possibly, which is insane if you think about it. But it also makes sense when they have these multipliers.

The goal isn't for OpenAI to cut spending and make way more money right now. It is to spend enough to have

now. It is to spend enough to have bigger multipliers later. Cuz as Chad is saying now, this is what the investors like. Exactly. The goal of an investor

like. Exactly. The goal of an investor isn't to double their investment in a few years. It's to potentially 100x

few years. It's to potentially 100x their investment in many years. I've

invested in over a 100 early stage startups and thus far I've exited zero of them. even though I've been investing

of them. even though I've been investing in them for over three years. The hope

is that around 5% of them will do so well that it makes the whole portfolio profitable. Obviously, for these later

profitable. Obviously, for these later stage companies, you want a much better ratio and you're expecting a smaller multiple, but you're still looking for a 5 to 10x for this. And when a company's

currently spending 33 bill a year and they have the potential to make a trillion a year in the future, that's an investment that you would want to make.

So that is why their current cost doesn't make a lot of sense because their current cost is investing in next year's profit and the year after that's profit. And when you combine that with

profit. And when you combine that with the multiplier they have on their revenue versus their costs, it makes a lot of sense, especially when you consider the fact that their profit was

zero until somewhat recently. They were

making no money at all until they released Chat GBT at the end of 2022.

The idea of them making money is still a relatively new thing. I would have more concern about where their revenue would be over time if competitors can have

better like spend ratio to profit ratio, but their competition can't even keep their websites up right now. I'm not

particularly concerned. I think OpenAI is set to do very well. If I could invest, I would have. I haven't. I have

no financial return from them doing well. That all said, seeing that 70% of

well. That all said, seeing that 70% of the rev comes from subscriptions, I do have more hope for our own subscription platform. If you would like to subscribe

platform. If you would like to subscribe to T3 Chat, it's only eight bucks a month. You get access to every model.

month. You get access to every model.

And in your first month, you can get for $1 if you use code rate exceeded at checkout. It's been genuinely fun diving

checkout. It's been genuinely fun diving into the finances of how all of this works. And I'm curious how y'all feel.

works. And I'm curious how y'all feel.

Do you want more videos like this or want me to stick with tech? Let me know.

And until next time, he's nerds.