Monologue: AI's Compute Demand Story Is A Lie

of them seemed to be losing money. I know, I know, though some of you are going to say, but these businesses had great quarters. But you'll nose is that said quarterly growth does not actually seem to be coming from AI when you read their results where they actually share

the AI roic return on investing capital. So, while Meta and Google refused to actually explain their AI returns because that would require admitting how bad they were, Microsoft revealed that he had thirty seven billion dollars in AI revenue run rate about three point oh eight billion dollars a month or so, and Amazon have fifteen billion dollars annualized or around one point two five billion dollars a month.

Do not be fooled. These are bad numbers based on my own reporting, Open ai spent three point six four eight billion dollars on inference in the third quarter of twenty twenty five on Microsoft Azure, or around fourteen point

four billion dollars on an annualized basis. While I only add half of the fourth quarter numbers, I estimate based on them that open AI's annualized spent hit over eighteen point five billion dollars or around four point six billion dollars a quarter by the end of twenty twenty five, and that's not accounting for things like Sorra two or

a launch of its code ex coding platform. In total, this puts open ai and estimated thirteen billion dollars of spend on Azure just on inference in twenty twenty five,

with billions of dollars more on training models. Now, Microsoft also accounted for sixty seven percent of core we's five point one five billion dollars in twenty twenty five revenue, or around three point four five billion dollars, and as all of that is used by open Ai, I really think you need to think very deeply about well, how much AI demand there really is outside of two very

large fail suns. I'm also confident that Corewave's computer is used for training open AI's models, as Corewave's announcement related to its direct deal with open ai specifically said it was contracted and I quote to power the training of open AI's most advanced next generation models, and said capacity per semaphore was only available because Microsoft declined to extend its current agreement with core Weave to use that compute

for open Ai altogether. That puts open ai spend on Microsoft services at over eighteen billion dollars in twenty twenty five, and it's easy to see how that could grow to over twenty four billion dollars on an analyzed basis in the last quarter, or around two billion dollars a month. Microsoft is open AI's primary cloud provider, and I estimate the open Ai represents around seventy percent of its AI revenue if not more, while taking up the majority of

its AI infrastructure. Otherwise, Microsoft's twenty million Copilot three sixty five subscribers likely pay no more than seven billion dollars a year. I'll get to my feelings about that in a bit now. As far as Amazon goes, things get

a lot grimmar. In early April, per Reuter's Andy Jesse, their CEO admitted that its cloud business's AI revenue run rate was more than fifteen billion dollars in the first quart of twenty twenty six, which translates to about one point twenty five billion dollars in monthly revenue, or roughly zero point four to one nine percent of the two hundred and ninety eight point three billion dollars in capex Amazon spent so far on the AI bubble, or around

twenty five percent of the five billion dollars Amazon just invested in Anthropic last week. I think it's also reasonable to assume that a large part, if not a majority,

of that revenue comes from Anthropic. Per my reporting last year, Anthropic spent five hundred and eighteen point nine million dollars on Amazon Web services in September twenty twenty five, at a time when it had around seven billion dollars in annualized revenue, a figure that increased by five hundred percent if you believe Anthropic, which I don't know if I do, to thirty billion dollars in annualized revenue since as of April, five hundred and eighteen point nine million dollars is about

six point two billion dollars in annualized spend, and I think it's fair to assume that its spend will have at least doubled to twelve billion dollars in annualized AI revenue for Amazon, or around eighty percent of the AI revenue. I also want to address this continual claim that capacity

constraints in AI are proof of some incredible demand. In reality, the reason that there's so little available compute is that the majority of it is taken up by either open ai Andthropic or hyperscalers running their own services with the dregs of what's left behind. Think of it like seeing a very very large man on the bus taking up four seats. There are absolutely capacity issues on this bus,

but they're not a result of everybody wanting to ride it. Now. Epokai, who I have had my differences with, estimate and I'm using these numbers because there's fuck all out there as far as compute capacity goes. They estimate the Microsoft had around two point oh two gigawatts of compute capacity available at the end of twenty twenty five, a time when open ai said it had access to around one point

nine gigawats of compute. As Microsoft is open AI's primary cloud provider, and Oracle had yet to stand up more than one hundred megawatts of stargate abeline. I think it's safe to assume that open Ai takes up eighty to ninety percent of Microsoft's available capacity. Similarly, Epoch estimates that by end of twenty twenty five, Amazon had access to an estimated one point sixty seven gigawatts of capacity, and I estimate over seventy five percent it's taken up by anthropic.

Amazon's anthropic dedicated project Rainier Data Center accounted for five hundred megawats by the end of that year, around a third of that capacity, and considering anthropic spend, I think it's reasonable to believe it's taking up the vast majority of what's available. Of course, there are capacity issues. You're

selling all the capacity to one fucking company. EPOC also estimates the Google had around two point ninety five gigawats uh remark figure that also realized on literally doubling its

capacity since the beginning of twenty twenty five. Nevertheless, it likely uses a lot of its GPUs for anthropic too, but it's hard to estimate how much as we have very little insight into anthropic spent on Google Cloud outside of its promise to use and I quote after one million TPUs with over a gigawad of capacity coming online in twenty twenty six, from the end of last year and a month ago when it said it would use and I quote multiple gigawats at next generation TPU capacity

starting in twenty twenty seven. I also think it's fair to assume that large swards of Google's compute infrastructure is taken up by Google Gemini being shoved into literally every single part of Google's products. It's everywhere. It's like a fucking weed. It's horrible. Google also refuses to break out its AI revenues, but considering it owns around thirteen percent of anthropic I also think it's fair to assume a large capacity amount goes to them. All Right, you know what,

I tried to be nice. I tried to keep my cool, but this whole situation is beginning to piss me off. I'm tired out of hearing people online may do this victory lap thing about how AI demand is there when it isn't. And when I started these monologues, I was told I'd be allowed to rent, and I haven't really let myself off the chain, so I think it's time to do so. The story of massive AI demand is a lie and nearly a trillion dollars, if not more, annihilated.

To create the largest circle joke of all time. Venture capitalists and hyperscalers feed money to open ai and Anthropic so that vengre capitalists can feed money to startups to feed money back to open Ai and Anthropic to feed back into hyperscalers for GPU compute, at which point they

turn to Jensen, Hogovinvidia and buy more GPUs. While it might seem tempting to credit these men as geniuses for creating companies specifically to feed themselves revenue, well, but to keep up with this k fabe of doing AI to substantiate the build out means that they've had to massively over commit to the bit even though the only two meaningful businesses in AI appear to be Anthropic and Open Ai, and that's only because they're effectively intellectual honeypots for the

entire industry. Outside of those two, the only other competitive AI businesses are those of Amazon, Microsoft, and Google, two of which have an annualized revenue of around six percent of their capital expenditures. So far, Google's AI business is so booming that it refuses to break it out. And while it nebulously claims that AI is creating growth, it's not really clear how, and they're vague about it because analysts and the media are ready to swallow the narrative

as long as number go up. That's why Google doesn't break out their AI revenues by the fucking way. Amazon and Microsoft had their hands forced by the markets after their stocks tumbled and fucked up by sharing their AI revenues. Amazon's two hundred and ninety eight point three billion dollars in capex has successfully created an AI business that and more than a quarter of a way to a trillion dollars, has successfully managed to make one point two five billion

dollars a month. That's fucking pathetic. If we had investors with IQs above room temperature, they'd run Anty Jassey out of Arlington like fucking But this is the AI bubble where everybody is wrong, because once everybody admits what's actually happening, they're gonna have to admit they all sounded insane for years. One point two five billion dollars a month after nearly three hundred billion dollars of Capex. Unbelievable, what a joke.

And like I said, more than seventy five percent, if not eighty percent or more of that revenue is from fucking anthropic who Amazon just agreed to fund up to another twenty five billion dollars. At some point, it's just Microsoft, Google, and Amazon handing themselves money or handing it to Nvidia to justify what they've burned over eight hundred billion fucking dollars creating some of the most wretched businesses of all time.

I'm sorry, Wow, Satcha. You managed to get up to twenty million paying Microsoft three sixty five copilot subscriptions six hundred million dollars a month in revenue, not profit, and all it took was you investing more than thirteen billion dollars in open forcing large language models into every one of your products in a way that brought us on harassment and about what two hundred and forty something billion dollars in Capex, as well as laying off tens of

thousands of people and savaging the Xbox brand. What the fucking shit, man, You should be ashamed of yourself. Open AI is in and of itself a kind of psychosis generator. It was the first thing in a long time that felt like a new thing since the iPhone for the

people that entirely obsess over growth. It was the panacea for the entire tech industry, creating a new way for business idiots to spend money on infrastructure, a new thing for consultants to scam people, with a new series of things to be an expert in as a journalist, the blogger, or just a scam artist, all wrapped up in something that could also be a consumer product and enterprise software products, and a new kind of API to attach to other

enterprise software products and then arch more money. In theory, open AI's success would lift everything at once, hardware, software and even adjacent fields. It promised to both democratize access to creating software while also heavily reinforcing existing power structures to the point that every dollar inevitably ended up in the Magnificent seven's pocket or asshole. The problem is that the system needed to actually work one day. It needed

to eventually make more money than it cost. Every single one of these companies is talking about AI NonStop, and not one of them can show you a profit. The only thing they can do is tell Liz of omission by saying AI helped boost everything, and when you ask for specifics, well, the results are either tepid or so

secretive you'd think they were hiding a dead body. The only reason Google, Amazon and Microsoft are being tolerated that their current excess is because their non AI segments continue to grow through endless price increases and in Sitification and its other business units, by which I mean Open AI and Anthropic are yet to run out of money to feed them. Sorry, By the way, I just I don't know what Meta's doing. I haven't talked about very much because I don't think Meta knows what Meta is doing.

Every so often they bury effect in one of their blogs about they saw a three percent increase in something related to AI. Then they promised to burn another one hundred and seventy billion dollars in a year, and again we can't really tell why. They also lost another four billion dollars on reality labs in the quarter. By the way, there should be a legitimate criminal inquiry into where this money is going. This is very strange. I think they're at like eighty eighty six billion dollars and all we

have to show for it is the metaverse and pervert glasses. Meanwhile, SpaceX and mister Musk is rushing to have the strangest and largest IPO of all time. All those daily stories leak about billions of dollars of losses and whatever the fuck that deal with Cursor is. Apparently SpaceX will buy it for sixty billion dollars or pay it ten billion dollars to fuck off. I think what actually happens is

a secret third thing. SpaceX five it for a bit, then there's a falling out between Musk and Michael Truell of Cursor, and then the company either rushes another acquisition or just dies. Remember Elon Musk killed Curs's funding round that was in place by doing this acquisition, and he can't buy this company before SpaceX goes public. Elon Musk took fucking open Ai to court. Do you think he'll care about killing Cursor? Who's gonna sue them the remains?

It's just a fucking rapper company. I'm so fucking tired of it. Anyway, that lawsuit with open Ai. It's a real alien versus predator situation, And if I'm honest, I found the whole thing a little boring a duo of Dullard's shoulder, budging each other to see who can run a company that neither of them can really describe because neither of them do anything other than pontificate and take credit for other people's work. Two dumb bitches going exactly

if this breaks open AI, I'll be surprised. But if it does, it would be extremely fitting that fucking Elon Musk would accidentally destroy the AI industry, like mister Bean sitting down on a button that launches a nuke here would be very funny. I'm just not giving it much hope. Nevertheless, this entire industry is only made possible by the global kfabe circular economy of taking every single sign is good for AI and ignoring every single possible glaring warning sign

and the hopes they'll go away. That is the entire strategy. That's all it is. It's just people being like, well, the bad things don't matter. They do matter. Last week, Microsoft said it's shifting gehub copilot to token based billing, meaning that people will have to pay the actual cost of their models. This is effectively killing the product as people know it and invalidates every single story about its growth, revenue or otherwise ever written. I don't see anyone on

writing those. By the way, no one gives a fuck. And to give you some context about the scale of how big GitHub copilot was or is until they stop subsidizing people's compute because right now when you use GitHub Copilot, Microsoft's just paying the models for you with them anyway, gehub Copilot is the second largest customer Ofanthropics models and was only that large because it was subsidized in the compute spend of its customers. A lot of that money

spend Onlanthropic. This is bad, and really that's the only way to build any kind of AI business. Google and Amazon realize that AI revenues are contingent on the continued survival of Anthropic, and Amazon and Microsoft's revenues are contingent on open AI and Anthropic. They know that if these companies die, they're going to lose billions of dollars in revenue. But also they have to compete with them for fear that they'll be seen as falling behind their horrible progeny.

And thus they're incinerating their brands and endlessly pontificating about the power of AI or spending nearly a trillion dollars on Capex, almost entirely to make sure their competition, which is also their customer, which is also the welfare recipient, doesn't die because if they die, they lose a bunch of revenue. But they need to spend way more money than they'll ever make to justify that. Yeah, it's a mess and a mistake, and eventually one of these companies

is going to grow tired of it. Microsoft was already billions under the analyst estimates for Capex. They're moving to token based billing on GitHub Copilot. They claim to invest an Anthropic in February, but didn't mention it in their ten Q their earnings in any way, shape or form, which means Nomani was sent. At some point, these fuck nuts are going to be forced to reckon with what

they're doing. Until then we'll have increasingly more frenzied and ejaculatory statements about AI demand that fail to match with reality. I truly think that it's going to be like this, if not crazier, until one day when the music suddenly stops. Somebody's going to blink. Somebody is going to take a step back and give everybody else permission to stop too. Maybe a perplexity lovable rip or cognition dies. Maybe Microsoft shifting GitHub copilots to token based billing on June first

inspires others like Anthropic to follow suit. Maybe AI tooken austerity begins at Microsoft, Meta or any other large company burning ten plus percent of their head count on AI tokens. Maybe in video fails to inspire the markets in just the right way, Or the fact that the data centers are not opening fast enough to have fully digested twenty twenty five's GPUs finally catches up with the economic mismatch. The Jensen wog always beats and raises expectations, and that

really is the strangest thing. At the current rate of sales, it's taking more than six months to install the single quarter's worth of GPU sales. At this point, it's obvious

that there are warehouses of these fucking things. It just isn't obvious whether they're in ones owned by Hyperscalers or the Taiwanese odm's original design manufacturers like Quantum Computing and fox con that build their servers, and last quarter it looked like Quantu Computing inventories were growing by billions of dollars. Is that good anyway? None of this makes sense. It

hasn't from the beginning. It's the largest bubble in history and has reached such an intellectual and financial scale that many have taken signs on it in a way that will be completely impossible to walk back if they're wrong. As things deteriorate, expect them to cling to their mythologies tighter and become more and more agitated and make vulgar threats and weird comments that don't really make sense but are nevertheless quite verbose. And really, we've never seen anything

like this in our lives. You realize that Open Eye doesn't make any sense right. It can't survive without eternal subsidies. None of these businesses can. Every single AI business you see is unprofitable, and none of them have a path to break even, let alone sustainability. And to make matters worse, it is the only way to create demand was to literally fund it yourself. These men believe they've created perpetual energy. What they've actually done is shit their pants and set

their houses on fire. Anyway, catch you next week.