Ed Zitron explains OpenAI’s leaked financials

[00:00:00]But there's something very strange going on with OpenAI's accounting from what I've seen. Even if you remove the cost of research and development entirely, OpenAI is still losing billions of dollars. Tokenbased billing failed. I think they thought that that would turn on companies would pay unlimited sums and pay them in perpetuity. What actually happened was everyone went, "Whoa, that's really expensive. What are you talking about?" Four years into this, four goddamn years, we should not still be debating whether this has ROI.

[00:00:32]I'm sorry. You don't get to do that anymore. You've burned over a trillion dollars on this crap. They don't have any way of justifying their current valuations beyond referring to the future. And their futures look bleak.

[00:00:55]Hi, I'm Isaac Pound and today on the tech report I'm joined by the writer of Where's Your Edat and the host of the Better Offline podcast, Ed Zitron.

[00:01:02]Thanks for being here.

[00:01:03]>> Thanks for having me.

[00:01:04]>> Very big news this week as you finally got your look at OpenAI's financials.

[00:01:09]And somehow it's worse than most people expected.

[00:01:12]>> Yes.

[00:01:13]>> 38.5 billion dollars of loss in 2025.

[00:01:18]you have in your newsletter uh gone through all the numbers, but you refrained from editorializing them. And so now that I've got you here, can I ask what do you make of the numbers?

[00:01:30]>> So there's a few things to note here. So revenue was 13.07 billion. Interesting to actually kind of close to reporting.

[00:01:38]Then the cost of revenue was 7.5 billion. Then they spent $19.18 billion on research and development and 5.73 billion on sales and marketing, $1.57 billion on people, general administrative. So a general loss from operations about $20.92 billion. That $ 38.5 billion net loss is very weird because it includes the conversion from a nonprofit into a for-profit that loses money. And what's interesting about it is I like I'm focusing on this 20.92 billion loss from operations just because I think it's the cleanest number, but there's something very strange going on with OpenAI's accounting from what I've seen. There's moving and I figure we'll get to this in a bit. There's moving numbers around to non-controlling interests. There's all sorts of weird maths happening. And thus the big thing I think people need to focus on is that even if you remove the cost of research and development entirely, OpenAI is still losing billions of dollars. Like that's the fundamental thing. The myth with this company has always been well if you remove training, if you remove R&D, this company will be profitable. Absolutely not. It seems that these this company the only thing that you can guarantee with OpenAI is that as it grows its revenue, it grows its cost linearly above it, which appears to be the case for every single AI lab.

[00:03:01]>> I mean, yeah, as you hinted out, the numbers from OpenAI write off an additional $21.5 billion of loss in 2025 under net loss attributed to non-controlling members capital. Yes.

[00:03:14]and net loss attributed to redeemable non-controlling interest. You explain it's unclear in your newsletter what that means. I will link that newsletter in the comments, but >> do you have an educated guess you could share of what that does mean because I'm I'm completely lost.

[00:03:31]>> So there is to be clear Open AAI it converted from a nonprofit to a for-profit. There is some amount of money within this that was lo that was moved around because of that. That's fair. That's not a cash cost. That's not something which knocks them. What confuses me though is that none of these numbers really line up in a way that a normal business does because of those non-controlling interests. Because putting aside all the stock for a second, this company lost nearly $21 billion. But in the end, the net loss was $38.5 billion. They told the FT, "Actually, we only lost $8 billion."

[00:04:11]something something very dodgy is going on here. I'm not not accusing anyone of anything, but it's very and you when you look back at 2024, they magic away $3.7 billion worth of costs to a non-controlling interest.

[00:04:27]What? And I've talked to multiple accountants since then, and I intend to do a more a bigger deep dive with one at some point in the future. talked to a couple of accountants and all of them are saying, "Yeah, this seems like they're using a different entity to shoulder some of the costs, but this is only on paper. Open AAI did spend $21 billion, sorry, lose $21 billion. They spent $34 billion." And I think what it is is that there is some sort of SPV joint venture situation. It's not the one with the private equity firms. That hadn't happened yet. But there's some sort of exterior entity that Open AAI is using as a means of moving its costs around to make things look better. And you only do that if things are very, very bad.

[00:05:16]>> I mean, what does it mean for the IPO?

[00:05:18]Cuz I mean, you've got to look ahead at all the numbers, but surely there's going to be people looking at these with much higher levels of education than you or I. They're going to be like, "Wait a minute. You're clearly doing something a little shifty here."

[00:05:33]>> I think that the S1 is going to look a bit different from this if I had to guess. I think that they're going to move stuff around even more. But it's kind of impossible to hide this kind these kinds of losses. It's kind of you can't really just you can say on paper, yeah, $3.7 billion that's just not our problem. That's someone else. But uh your cash flow statements, which I'm not party to right now, uh they will show differently and there's no real hiding that. On top of that, it's very clear that OpenAI's cost just linearly increase with its revenue. There's no getting away from that. Even SpaceX, which is a real dog of a company, looks way better than this. Open AAI just loses a bunch of money to make some money. People look at the 13.07 billion and they say, "Wow, it still made a lot of money." Yeah. as the single most prominent startup in the entire world, mentioned basically a hundred times a day on every media out there. Yeah, if they and to be clear, raised over $40 billion last year. Yeah, I would imagine that they would have this much money.

[00:06:38]Like what what do you expect? They've extremely wellnown. But there's also the really the thing that I keep coming back to is the fact that I reported back in November last year that OpenAI spent 8.6 67 billion on inference through Q1, Q2 and Q3. And this one has a cost of revenue at $7.5 billion. I think OpenAI is categorizing its cost in a different way. I think they're doing something weird with the way they're categorization.

[00:07:08]That was not part was not party to that information and what I saw. But I think what's happening is that OpenAI is moving stuff into different categories as a means of hiding how bad its real costs are. And I think that the most obvious place for that is the sales and marketing cost because to be clear, $5.73 billion in sales and marketing is more than the Coca-Cola Corporation.

[00:07:32]They spent a little over $5 billion last year and I've actually been digging into this sales and marketing thing as well because I was quite curious. OpenAI's first major media campaign was September 2025.

[00:07:44]>> That's the first one. It's the first time they ever did one. They do a ton of Reddit, ton of Facebook, but even then I don't even know what ad agency if they even have one. If they are spending billions, like let's just take this to its face. They're spending billions billions and billions of billions on advertising. They would be one of the largest social advertisers in the world.

[00:08:06]Teimu think they spend two three billion dollars a year. Same thing with Shien.

[00:08:12]Is OpenAI the largest advertiser in the world and we just don't know it? No, no.

[00:08:17]It's blatantly obvious that what's actually happening is that OpenAI is putting inference costs in there. If you look as well within the documents that well didn't publish anything like that, but if you look at the the numbers from the documents, there was a $500 million or so cost from Microsoft under sales and marketing. Do you think that that's because OpenAI is running a bunch of ads through Microsoft? Now, what I think it is is kind of simple, which is that whenever open, this is a guess. It's just a guess. When OpenAI gives away free credits, as they regularly do to startups, and Microsoft does the same thing. Whenever OpenAI is a free customer, they categorize it as sales and marketing. They put that cost inside another category so that people go, the cost of revenue is lower than the revenue. Uh uh they're profitable. No, it's like when you add together the sales and marketing cost and the cost of revenue, it comes out to $13.22 22 billion basically the exact little bit more than the revenue. same deal in 2024 you add together it's a little bit more than revenue. I think that they're just moving numbers around and my reporting from last year kind of validates this.

[00:09:26]Like in the documents viewed in that one, it said open AI inference.

[00:09:32]What do you think? What could the what could that category possibly refer to? I can't imagine.

[00:09:37]>> So having seen the numbers for now and knowing what we do know about this year so far, how do you expect it to be getting worse and if so by how much?

[00:09:48]>> So I can't say exactly. I don't have I'm not party to those numbers yet. Reach out if you have them. Eitchron 76 and Signal. Um I would say things are getting worse because Sam Orman himself said a few months ago that costs have become a huge issue with clients. So I think that they claim that they're at $2 billion a month as of March, I want to say. So revenue is increasing, but based on this costs are likely increasing, too. OpenAI has been giving away thousands of dollars of credit on codecs as well. So very high chance that that that's also going to be ballooning that sales and marketing cost. And indeed it really just seems that the costs increase linearly with revenue for everyone. If you look at the two publicly listed uh AI companies in China, Zoo and Minax, both of them, what a surprise. massive sales and marketing division for a companies that don't appear to do much sales or marketing. I it appears that this is the new slush category where people put the costs. I think that the OpenAI is going to burn even more money this year. Uh during the Elon Musk trial, they said they were going to spend $50 billion on compute alone. I think they're going to burn a ton of money. I think their margins are going to get worse. I think that they're going to play some real wacky moon math as a means of hiding some of these losses. I think that sales and marketing cost is going to increase maybe seven or eight billion this year could be if it's over 10. Come on. Um and I think I think at this point we're going to enter the kind of distress despair phase where they're going to throw everything at the wall which is why they're talking about price cuts already.

[00:11:29]>> Well, yeah. To your point about price cuts, how can they afford them?

[00:11:32]Especially given the numbers that you you've revealed to us, how is that going to work if they're already burning money like smoke is worth more than the cash itself.

[00:11:40]>> Yeah. Uh there it's going to just be really bad. Like if these companies cannot afford to lower their prices, there is no magic trick where they're like, "Oh yeah, actually it's way cheaper than we thought." Unless unless the co unless inference is just the worst deal in the world and they're just upcharging by 10x, which is obviously not the case. These companies are just burning money to burn money. And the only reason they're doing price cuts, I think my personal belief is because they've seen that churn is going to be so severe that just if they don't drop the prices, people aren't going to pay them at all, which I think might happen either way. But anyway, it's going to be just as expensive to run the services even if they do a price cut. So, I don't know. I think they're just trying anything they can at this point. Last year, they did a lot of key jingling.

[00:12:28]They said, "We're going to do a device.

[00:12:30]We're going to do a browser. We're going to do a social network. We're going to do Sora. We're going to do this." Now they're Now that hasn't worked. They're like, "Okay, we're going to make everyone pay the tokenbased billing and then um well, people didn't like that, so now they're going to do price cuts."

[00:12:45]It's unclear what the actual strategy of this company is. What I do know is though, if they lower their prices, depending on how much they lower them, they're just going to increase their cost. I only I think it will just cut their revenue as well because might mentioned this before.

[00:13:02]>> You can't measure the ROI of AI. You can't do that. It's not possible. You can barely measure the cost of a task.

[00:13:08]So let's say someone spends a million dollars a month. Okay, cut that to 500 grand. Why is that 500 grand worth it if you can't measure the ROI? It's not like they're saying, "Oh, I've not heard anyone say if it was half the price it would be worth it." I've not heard anyone give any kind of scale to that which suggests that no one really knows what the price should be just that it's too expensive right now. So I don't think that anyone really has answers. I think that they're just kind of throwing stuff around. The price cuts have yet to happen. Those are still a reported by the Wall Street Journal but still a rumor. I guess once those happen, I think the reason they haven't happened yet is because they know once they do it, everyone does it. Once everyone does it, it's just a race to the bottom. And I don't know, I don't I don't think any of them can afford it.

[00:14:02]>> Off the back of those numbers, OpenAI also got the news from Sensor Tower that as of late May, Open AAI market share dropped below 50% for the first time.

[00:14:13]And I was just wondering what you think of that because my read is OpenAI isn't really sea selling anything unique anymore. Apple and Android Google are filling in that consumer space.

[00:14:25]Alultman's so so-called super app you mentioned would have would have taken up and then Anthropic on the enterprise side is offering I mean generally reviewed better product and then that's assuming people don't move to open source.

[00:14:41]>> Yeah, that's the thing. It's unclear. I think Anthropic has this problem too though. It's like these companies offer basically the same product. Everyone offers the same thing. You got a coding artist, you got a coding platform. Also, I just want to take a second to say I think they've announced that super app three times and this time it was on the front page of the Financial Times and nothing happened. Where's my super app?

[00:15:05]They keep promising it. But yeah, I think the problem is something that everyone's going to face, which is there's not really much defined differentiation.

[00:15:16]There is the vague vibes of this one's better at coding. But I've watched people I follow the AI people. I've watched people go from saying that anthropic claude code can marry their wife if they if it would like to to saying that codeex is the best thing ever. And it just seems to be like kind of okay, they both do something, some of them do something better. So yeah, of course, OpenAI's market share is dropping in an industry that's effectively commoditized. And I think it's GLM 5.2, new open- source model that came out scored very high up there with like actually competitive with Fable and and Mythos and all that. Not Mythos, um, Fable and Opus 4.8.

[00:15:57]So yeah, I mean at some point I think it's going to come down to most people not really knowing why they're paying for it and those that do saying why would I pay for the most expensive one get basically the same thing on an open-source model. Now this kind of conversation's been happening a while.

[00:16:16]It's started to take up now only because the ROI conversation is happening. And I also just want to say four years into this, four goddamn years, we should not still be debating whether this has ROI.

[00:16:31]I'm sorry. You don't get to do that anymore. You've burned over a trillion dollars on this crap. Well, you can't just sit there and be like every year be like, "Yeah, well, maybe it's useful."

[00:16:43]Or just saying, "It's obviously useful."

[00:16:45]It actually isn't obviously useful. If it was, we would never be having anyone saying AI is real. AI is here. It's real. It's important. You don't need to say that when something's real and important. You say it when you're not sure why something's real or important and you need to con people further. And I just think fundamentally there really isn't that much differentiation between these products anymore. So you just got this situation where again and again people kind of bop up and down. I'm sure Anthropic's market share will drop and Open AIS will go up. And the funny thing is is that market share is also driven by those subsidized subscriptions. So it's like, okay, great. So it's whoever's providing the best subsidy this week, whoever's providing OpenAI allows you to burn $14,000 worth of tokens on Codeex. Yeah, that's probably why probably why people are jumping there. I don't know. I even did just before this, I got an email because I I retain a Clawude a Claude account. I don't use it, but I have I have one that I'm signed up to to get the announcements. And apparently they're doing the same thing as OpenAI where if you refer a friend, you get free tokens and if you refer a friend, they can try Claude Code free for a week. And it's like, I thought that this was the biggest, most hugest thing ever that was growing like hotcakes. Why are you having to do referral programs like a direct to consumer protein powder? Like this is very they're they're not acting like companies that have the mandate of heaven. They're acting like actually extremely boring startups except they have the worst unit economics I think in tech history. Maybe Juicer No, Juicero was better. I think Juicero Juicero had them beat on margins, but I would have to check my my numbers on that one.

[00:18:37]>> I do want to there's a couple more questions I have on open source in a bit, but we've spoken before about how open air needs to achieve incomprehensible growth essentially just to meet its own commitments and its and its own projections.

[00:18:51]>> Yeah. So just quickly missing missing user and revenue targets failing falling market share rather and worse losses than anyone expected. You once said to me that coreweave would probably be the first to go but now looking at all of this do you think it could be open AI? I think that there is still a chance that OpenAI's S1 really freaks people outwork style, especially if they try some Weiwork stuff, especially if they try and come up with margins using wacky maths.

[00:19:23]And I think that I think that it's going to be I my actual bet is it's going to be a few no-name startups that die first. It's going to be a few little fires that that break out that start making people go, "Oh, I don't know about this. But I think that in the end OpenAI is a good target. I think that Corewave Coreweave has a lot of money. It's raised through debt. So they have some cash cushion. I think OpenAI has a cash cushion as well.

[00:19:51]What it's going to take is a loss of confidence in the fundamental business model. And I think the S1 could do that.

[00:19:57]I also think if Anthropic rushes their S1 and the numbers look bad there too.

[00:20:02]Yeah, I think that that could freak people out because right now the kayfabe of this, the pretend everyone pretending that this is real. All the markets, it's all based on vibes. It's based on the continuing value of stock. It's based on the semiconductor rally. It's based on the continued issuance of data center debt. If any of that breaks, if any of that confidence shifts, if a big data center project dies because the money wouldn't come through, like a big prominent one, I think that that could start causing trouble. There was a story of the information a few weeks ago that Stargates billions of dollars more than they expected in Abalene. And you think happens to that project as well. It's a confidence game. It's not based on anyone's actual revenues other than the semiconductor companies selling stuff to data centers. It's based on the sense that this will never stop growing. And that's what needs to shift. Once the belief really sets in that this is like this is not going to keep growing, there's no way they meet their $1.1 trillion worth of compute commitments through 2030. It's just not going to happen.

[00:21:07]>> I was reading earlier this week that Microsoft is looking to deepsek to power its co-pilot co-work because of the unbearable cost of paying anthropic and open AI for their for their models. And in fact that an open source model GLM 5.2 I think it is performed better on front-end coding than Anthropic's best accessible models. Anyway, what's the impact going to be on AI markets when the business idiot as you call them realizes they can get AI at a fraction of the cost and which use which usage based pricing is going to push them towards anyway if nothing else?

[00:21:45]>> I don't know. I my whole thing is with open source is I'm yet it would take Microsoft actually doing that versus talking about doing that to genuinely make me think something's changing. You had Brian Chesy from Airbnb claiming he'd moved over to open source models as well. I think if that starts happening, I think it like it would need to be a big corporation like Microsoft actually doing it to make me think something was changing because we've had these conversations for ye like over a year now since the first Deep Seek. We've been to hearing about these open source models allegedly being better. I hope that that's the case. I think if this crap has to keep continuing, at least it could be cheaper. But then then comes the problem of it's not as simple for open source models as just paying for them. Like you can't obviously because they're developed out of China. You can't do a thing where oh we'll just use a Chinesebased API because don't know where that data is going. I think that if Microsoft makes this change, it is a C change. It will be something that you see across the board. I heard from someone at Microsoft the other day that they just could not access Claude Opus.

[00:22:51]Who knows what's happening there.

[00:22:54]But I think open source is going to be a way that this could wind down as well.

[00:22:59]It could be the get out of jail free card for people because I think that what companies could do is that go you can use just open source models. You can use these as much as you want and maybe that's enough to make people use them and if maybe that's enough to make people kind of excited about AI but I don't know if it will continue the excitement. I'm not sure how you build much hype beyond behind that. I don't feel I don't feel like anyone's going to be particularly excited for GLM. But maybe that's what AI becomes just this side thing that people sometimes use.

[00:23:30]How much of this keeps going without without the hype behind it. There's also a cost thing as well. Microsoft, they have direct access, I believe, to the weights of OpenAI's models, even their closed source ones. They can run them on their own hardware. They can run them at cost.

[00:23:49]that now those models must be really unprofitable to run if Microsoft's doing this. What how unprofitable? That's a good question. So really the answer is until someone actually commits to doing this at scale, I'm yet to be sold. But Microsoft rumor rumor mill there is interesting, but it's kind of like the price cuts. It's like once it happens, it means something. The fact it's on the border almost makes me wonder if it's a deliberate leak. Nevertheless, Mustafa Sullean, head of Microsoft AI, did say a few weeks ago that anthropics models are too expensive and Microsoft wants to move away from them. To be fair, Mustafa Sullean has also said that we'd have super intelligence within 12 to 18 months for years. So, it's kind of who do you believe here? Nevertheless, if open source keeps doing this, I'm not sure what it is that anthropic and open AI will be left with. because I'm pretty sure you can use clawed code and use codecs with these harnesses. You can use those harnesses with open- source models. So, I mean that they would say people would then say, "Oh, well then they have a product, do they? Is selling access to a harness going to make them enough money to pay $1.1 trillion worth of compute commitments?" I don't think that's true. But I think it comes back to the thing I've been saying, which is just we're this close. Everyone's everyone is clearly tired of this.

[00:25:18]Microsoft's tired of it. The markets are tired of it. I saw a Goldman analyst the other day saying that uh the first hyperscaler to drop their capex will be rewarded for them from by the markets. I think the moment that happens, ooh, or if the markets reward someone for cutting capex, I'm not sure how any of this works anymore. I think it'll be delightful to watch for us haters, but I think the I think it could get really weird and messy very soon.

[00:25:46]>> You mentioned Enthropic and it it does get me wondering because they they've also paused I think their token based billing for the Clawude agent SDK which is kind of the the AI that developers use which burns sort of 10 or 100 times more tokens than the everyday kind of LM. It's it's so it's being able to use your Claude subscription with another harness. I think it is.

[00:26:10]>> How do you imagine Anthropic's financials compared to Open AI's? And I suppose if they're similar, how are they affording to pause raising or applying actual cost rates to something which is must be burning an insane amount of money.

[00:26:26]>> I think Anthropic is doing exactly the same thing as Open AI. I don't think that there is much nuance between them.

[00:26:33]I'm sure that their sales and marketing costs are very high, too. And the other thing is is that they are doing like they're giving away $1,000 of claude code credits on an on enterprise plans.

[00:26:44]They just they're doing this because they realize that people are going to stop paying them. Even though they're paying them for an unprofitable service, people are going to stop paying them.

[00:26:54]And when they stop paying them, that's bad. So, they got to give them more free stuff. I I keep coming back to one thing with these companies. They don't have a plan. I think people think they have a plan or a strategy or a thing that they're working towards and that all of this is part of a STR. No, I think they thought that if they raised enough money and kept training these things, it would get cheaper. It didn't get cheaper. Or they thought that if they came up with a good coding thing, people would pay any amount of money for it. People won't pay any amount of money for it. They're getting upset that they're paying so much money for it. So they're like, "Okay, well, we'll charge people to use their use their coding harnesses with our subscriptions." Oh, no. People didn't like that either, so we're not going to do that. It's just It's very weird, but it's This isn't what a company with a master plan does. It's what a company does when it's just jingling keys like, "All right, do you like this? Do what do you think of this?

[00:27:50]Is this good?" And I think this desperation and confusion is coming from the fact that tokenbased billing failed.

[00:27:56]I think they thought that that would turn on. Companies would pay unlimited sums and pay them in perpetuity. What actually happened was everyone went, "Whoa, that's really expensive. What are you talking about? What do you mean I spent $2 million in a month? What do you mean I spent $500 million because I forgot to set cost controls. I can't wait to find out who spent half a billion dollars in a single month." I was talking to economist Paul Kadroski about this the other day. He said it will show up on someone's earnings.

[00:28:24]Someone will have to declare that because it's a big enough cost. Can't wait. It's not Amazon. I've confirmed that, but it's someone else.

[00:28:32]I I had a source tell me it's not Amazon, but who knows? May maybe that's within Amazon's interest to say otherwise. But someone spent that much.

[00:28:39]So, I think these companies are just going like, "Uh, do you like that? Do do you want the $1,000? I guess I please pay me more even though I lose money on that. And when you pay me, it would be really good if you could just keep paying me so I can raise money so I can lose it. So you can keep paying me, but when you pay me, I lose the money. It's just a very nonsensical industry.

[00:29:03]And it's unclear what the path to the future is. It It's not really obvious what it is that they're going towards.

[00:29:11]They are all jingling the keys of claiming they're working on recursive self-improvement now, which is AI that teaches itself, which by the way is a copout. It's a copout. What you inventing? I don't know.

[00:29:25]I'm inventing something that invents things. What are your ideas? I don't have any. I want an idea machine. Can't the machine do it? I' I've been annoyed at this because back in 2019, Sam Alman said in a speech that he would ask the AI what was profitable, how to make the company profitable, which was like at the time I was like that was 2019. It's before chat GPT. No, they they're doing that again. They're just they're trying to work out how to make the AI do the work cuz they can't work it out. None of them have a plan. I think I saw these are I mean off the top of my head so these might not be entirely accurate but that most of the plans out there if you prompt between two and five times a day you're already breaking even with pretty much what they cost.

[00:30:11]>> Yes.

[00:30:11]>> And I suppose do you think we've hit uh an enterprise pricing wall that we're just not going to be able to pass?

[00:30:19]>> Yeah. I think that we've reached the point where it's obvious that there is a ceiling to how much people will pay for AI. It's not obvious. The thing is the the conflation people are making is the fact that we hit a ceiling means that the actual amount people will pay is here. We actually don't know if it's much lower because you can't measure the ROI of AI. But fundamentally, yeah, we've hit a cost wall. that thing you were talking about with the subscriptions. It was semi analysis assuming that if you had 75% gross margins on influence, no proof this.

[00:30:54]None. Zero proof.

[00:30:57]Nevertheless, if you had those, all it takes is someone using 25% of their utilization in a month on average to make the service unprofitable at any price point. And then I think at the top end if someone uses $14,000 I think it's like negative600% gross margins. But on the tokenbased billing side, yeah, we've hit a wall.

[00:31:23]There's there's no real way to put this other than yeah, we've we have smacked it. We have smacked head first into a wall. Blood streaming down our face.

[00:31:34]Because think of it like this. Let's actually think like the AI companies for a second. And all I can see is blue and yellow. Uh, okay. They need to they need Let's What scenario would make people pay this much? Okay, it's clearly not the current models. Just we know that for a fact. It's not this much. So, the models would have to be so good in a way that was so impossible to argue with, just completely impossible. You would have to just It would be impossible to argue with. They would have to basically be what they've been promising this whole time. They would have to be just type in build me Salesforce and it would crap it out in 2 minutes. It needs to be impossible to argue like going from a flip phone or even a a chocolate bar phone to an iPhone. It would have to be that meteorically and it would have to be in a way that would be measurable by a CEO that does not use it and also the engineers using it. It would in fact it would have to leave engineering. That's the big thing that everyone keeps avoiding talking about. Where are the use cases outside of software? Where are they? Someone will respond to this in the in the YouTube comments and they'll say, "Well, I used it I use it as a research tool. I wouldn't like to read your research, mate." Don't don't think that's very good. But point I'm making is for the costs to grow. Because by the way, just what's being spent right now is nowhere near what they need. They need like 10 times this. For that to happen, it would have to get more expensive and more useful in a way that was impossible to argue with. We wouldn't be, in fact, we wouldn't be having any debate about AI. In fact, the fact we're having a debate about AI being real or good is a kind of a sign that they're in a bad place. The fact that there's anything happening that's not just, "Wow, this is really this is really good. I can't stop using it. I would have to be impressed. Like I just there's nothing that's happened so far that's impressed me. But it would have to be an obvious sea change. It would not just be I spat out an open- source website. I type build a website and a open source website came out. It would have to be something that has taste. The new thing they're mentioning it's like taste of a human. Just complete alien talk. They're like AI that has taste.

[00:33:54]It's like what what are you talking?

[00:33:56]It's like guys online who refer to women as females. It's just like just like people that don't go outside. Long and short of it is they need to make way more money right now and they need to make it fast. They're currently planning to make less money because people won't pay enough. They don't have any way of justifying their current valuations beyond referring to the future and their futures look bleak.

[00:34:23]>> Well, Ezron, thanks for taking the time.

[00:34:26]Thanks for having me. If you enjoyed today's episode and you want to hear more of the tech report, please consider liking and subscribing. Also, you can get episodes of the Tech Report wherever you get your podcasts.