Monologue: Anthropic Slows Down

Hello and welcome to this week's Better Offline Monologue, I'm your host Ed Zittroth. On Wednesday night, Anthropix Q3 numbers dropped through leaks from meetings with investors in preparation for its IPO, and boosters have been packing their lunches like they're Bud Dwyer on the way to a press conference. Why? Why?

because despite hitting $ 16. 8 billion in revenue, sales only grew by 46% quarter over quarter, which might seem like a lot, but Anthropic's entire sales pitch to investors and the general public was that its revenues were eternally parabolic, and based on the oinking and squawking from Twitter, the expectation was that it would hit $ 20 billion or more for the quarter.

Another reason the Hogs are so upset is that Anthropic neglected to include a new annualized run rate figure. A made-up number derived from multiplying, in most cases, a trailing 28-day period's worth of revenues by 13, specifically to make the company's revenues seem bigger and extrapolate growth in what I'd describe as a deeply deceptive way. The reason Anthropic neglected to do so is that it's likely that the number was lower than boosters expected. A few months ago, Bloomberg and others reported that said run rate had hit $ 65 billion at the end of July. which means that the expectations had them set anywhere from $ 75 billion to $ 90 billion, though I should add that these numbers are all complete bullshit, and none of these articles ever explain

how the number was derived and over what period it referred to. In fact, Bloomberg just says a short period, annualized. It's completely fucking useless. In any case, boosters have loved these big, beautiful numbers because they allowed them to express anthropic's growth in terms that made things look far better than they were. A few months ago, multiple outlets had reported investors expected Anthropic's run rate to hit $ 100 billion by the end of the year, and now many of them are posting that's not going to happen with the tone of someone who found out their dog is dying.

The leak also revealed that Anthropic had an adjusted operating margin of 11% and adjusted gross margins of 85%. Adjusted is a term used to describe stuff that's not up to the standards of GAAP, Generally Accepted Accounting Principles. because it leaves out costs like stock-based compensation, R & D, or, well, anything the company wants to, because it's non-GAAP. Nevertheless, some outlets are using this to suggest that Anthropic is approaching profitability, as in it would be profitable if it wasn't for all the expenses. Another wrinkle in the story was that Anthropic noted that 5-7% of its revenues came from distillation.

as in Chinese model developers using its services to train their own models, meaning that anywhere from $ 840 million to $ 1. 17 billion worth of its Q3 revenues came from its competitors building competing products and paying them for the privilege. Anthropic recently reported that it had taken measures to reduce these attacks, as they call them, by more than 90%, which is good news on a competitive level, but also means that its revenues have, for several years... been artificially inflated, probably by billions of dollars. This all puts Anthropic in kind of a bind. To match its current revenue growth, it would need to grow another 53.6% or so quarter over quarter, as its revenue base is now I assume

about 5% lower, meaning it needs about $ 25. 8 billion or so in Q4. Anything less than 20 is a fucking disaster by the way. And per Ramp data, 80% of Anthropic and OpenAI's enterprise revenues come from the top 1% of its customers, with the vast majority of those being unprofitable, unsustainable AI startups, meaning that that growth will mostly have to come from venture-backed money losers in perpetuity. Ramp also reported last week that AI spend is falling despite rising usage at exactly the wrong time for all of this to be happening. And all of this times extremely poorly, with an aggressive sell-off of better-off-line third-of-the-century core-weave. which has now reversed somewhat because of the dunces that bet on

the financial markets, on NVIDIA CEO Jensen Huang saying that Corweave was in good financial condition and that NVIDIA did not need to backstop it, despite the fact that NVIDIA has already backstopped Corweave's leases and sunk billions of dollars into the company. I do not see these factoids in a single fucking story mentioning Jensen Huang's comments, because journalism is apparently too fucking difficult for everyone. And this pablum did very little to calm the bond market, despite the stock bursting a little bit by about 3% as I'm looking at it. Corweave's debt now trades at a yield of at least 14.6%, with its worst one trading at 15.47%, meaning that those closest to the actual money, the actual debt,

the actual, you know, payments of stuff, are pricing Corweave's debt like the junkiest of the junk. I think the worst of the junk is like 17-20%, so some ways to run, but pretty damn junky. And as all of this boiled over, Gartner released a report saying that by 2030, 80% of neoclouds would have ceased operations, adding that forecasts for AI revenues in the next three years were trillions of dollars short of what it would take to have even a 12% return on invested capital. And I must be clear, the assumption that they have is that AI will make $ 3. 5 trillion worth of revenue.

And they're still trillions of dollars off.

Just to be clear, AI is not even making $ 300 billion of annual revenue. People would know this if they fucking looked, including Gartner. Gartner has the resources. Why is no one fucking bothering? I feel like I'm taking crazy pills. And this was mere days after a Jefferies report came out that said that the most likely outcome of the so-called AI boom would be a massive capital destruction in the US. I just need to ask you a question.

Is that good? Anyway, markets in both the US and Asia puked a little on all of this bad news. Anthropic's growth had to be absolutely meteoric to warrant its supposed demands for a $ 2 trillion valuation and to pay for its $ 413 billion worth of non-cancellable contracts. This company had been priced for perfection and made promises contingent on achieving growth that, at some point, comes down to how much venture capital money its customers can raise and continue spending in ever-growing amounts for businesses that only lose money.

You know, much like OpenAI and Anthropix. This has been a lot of fun to record. I will admit, I got a little stuck with the show for a bit. I hadn't been having as much fun as I wanted. I realize it's because I've not been showing my fucking contempt for these companies. I find all of them sickening.

I find the people that inflate their egos and their balance sheets, at least with fake numbers, truly despicable. I find it contemptuous that the so-called journalistic professionals out there are still using run rates, that they're taking them seriously. I see financial analysts taking them seriously as well. I don't come from a finance background. I'm just a regular guy who read a lot of stuff. And let me tell you something. It is fucking disgraceful that the AI bubble has had anything based on these fucking elf numbers, this magical math of run rates. And to see the hogs braying so oinkily about the fact that we didn't get a new number is both vindicating and sickening. You motherfuckers should actually look at the real numbers. You should

want real data. You shouldn't want elf numbers. You shouldn't want imaginary run rates. And any journalist listening to this, if you are still using run rates, you are failing at your job. Make no mistake, I am not mincing words here. If you're using them without a giant goddamn disclaimer saying that you don't have the underlying matrices, the underlying way that they're calculated, I mean in bold and italics, like a warning sign on a pack of fucking cigarettes, you are failing at your job.

That is just the truth.

I know this might hurt your feelings, but your feelings need to be hurt for something to fucking change. Because when this explodes, it's going to be so much more harmful because it got inflated on these flimsy numbers. And you'll notice that everyone seems to freak out the moment real numbers come into the picture. The moment we learn the actual underlying financial conditions of these companies, people begin to panic. In a real industry, in something that actually is worth all this money, the real numbers would be good news. The real numbers would show you how well AI was doing. Yet whenever we get a real number, it sucks. It fucking sucks. And it continues to suck.

It is fucking crazy. I put it in my newsletter this week. It's the annualized Enron rate.

I love that title.

I put it in there, but we don't know the financial profile of the vast majority of AI startups. We now have, what, nearly two years of Anthropix revenues. We don't have their costs for most of it. We have, thanks to me, the auditive financials of OpenAI for 2024 and 2025. We don't have shit on perplexity or cognition.

Cognition's worth $ 48 billion, and we don't know anything about P & L. We just have fucking annualized run rates. One billion annualized run rate. Who gives a fucking shit? I'm sorry for not being more eloquent about this. I know.

But I just can't help myself. I can't help myself. For having.

contempt for these fucking companies and the shitty ways in which journalists help inflate them, inflate their egos, inflate their financial profiles, help them raise money. I have watched multiple financial outlets today run cover for both Corweave and Anthropic. That's not your fucking job. If you feel the instinct to tell a good story when things are going badly, as a journalist, you are no longer participating in journalism unless you are reporting. I mean, actually reporting something. I don't mean I've looked at the adjusted operating margins and I found a good way of looking at it. I've done some napkin maths and I've come up with a new way the run rate could be good. That is not journalism. It's fucking marketing. So forgive me if I'm

a little bit fiery about this stuff. Forgive me if I'm calling people fucking idiots for doing this stuff. Forgive me for saying people are not acting in a professional way when they use these run rates.

Forgive me. Actually, don't. I don't give a fuck.

What I give a fuck about is the truth, and I'm sick of hearing these fucking numbers. I am sick of the ways in which journalism is holding up these companies. It's not going to change a thing. And as I've said, as more real numbers come out, Things will get worse because the actual underlying economics of AI do not work.

And I know so many of you email me agreeing, and I love it. I love you all. I truly do.

And that is a reference to the menu if you're new here. And I'm just, I'm happy to have you, but I'm frustrated because even with this anthropic news, as I look over at my terminal, I thought, oh, maybe this will push it over the edge. No, call weaves back up because Jensen fucking Huang said they're fine. This is how dumb the markets are. And of course, journalism, just copy-pasting that shit. No need to think. No need to consider. No need to have an opinion or even form any thought at all. Just put the story in the bag. That's all it takes.

Well, guess what?

Everyone who participated in that, without exception, don't care where you were, everyone who helped inflate these companies, when this shit explodes, it's on you. If you're listening to this and you're a journalist who's done this stuff, it's on you. You can still change things. You can turn against these companies.

You can actually push back on the underlying rotten economics. You won't. Sorry, I know you won't. And if it's because your editors are telling you not to, please reach out and tell me, because I'd at least like to know those of you who are being muzzled. But if you're not being muzzled, if your reason for using run rates is because you actually think they are a useful measurement, you're either ignorant or corrupt.

You choose. I realized that was a little damning. I realized that was a little brutal.

But I'm going to be honest.

This is the show now.

I really let go of what made Better Off Line great, which was the actual emotional connection with all of this stuff, airing the frustrations I have, ad-libbing a little as I have at the end of this.

I'll be honest.

I fucking love doing it. And you're going to really hear that in the next episode. I'm joined by an original guest of the show within the first, like, 10 episodes, maybe less. Tom Dautin, he wrote an amazing piece of Vanity Fair about Larry Ellison and his big, wet son, David. And actually, Larry Ellison also has a daughter. I'm surprised I didn't know this. Anyway, it's one of the best interviews I've done. I think I just said interviews, but we're going to keep going. It's one of the best ones because I've just fucking let loose. And that's what I'm going to do with the show going forward. Because talking about the tech industry in this cold, calculated, ultra-financialized way without having the disgust necessary to talk about these

companies is just kind of wasting your time.

That's a little harsh on me. Either way, I really enjoy hearing from all of you.

Shoot me an email. Shoot me a glute on slub or a slub on glute. Those are two different social networks and two different buttons you can hit on them. And I will catch you next week. I love you all.

Kapitelklicken springt direkt hin

Shownotes

Transkriptvom Podcast bereitgestellt · Zeitmarken springen hin