Anthropic's IPO prospectus is big on promises, short on answers to tough questions

Brandon: Hey everyone, and welcome to the latest episode of the Registers Kettle Podcast. I'm your host, Reg Reporter Brandon Vigliarolo, and this week we're focusing on what I think our contributor Carly Page aptly described as the peak of the AI hype cycle we've been in for the past few years, and that's the public release of Anthropics' initial public offering filing. With me to talk this over is our systems editor Tobias Mann and senior reporter Tom Claiburn, both of whom I'm sure have plenty to say about this matter. Welcome guys.

Tobias Mann: Good to be here.

Tom: Okay, good to be here.

Brandon: Yeah. So, I'm sure we've got a lot to cover. you know, six hundred and fourteen pages is a lot to go through. Tobias, I understand you were combing through this thing with with with a fine tooth to comb. talk us through the hard numbers before we get into some of the more sort of wobbly parts of this of this IPO.

Tobias Mann: Sure. on face value, it it you know, you going into any S1 filing, it is a kind of mess of you know, risk mitigation, kind of covering your butt. You you get a they all pretty much read like you know, a doomsday prophecy. and

Brandon: Right.

Tobias Mann: the S one definitely did that. But on the numbers side of things, The the the take for the market was kind of interesting because there were some people looking at the numbers and going, Okay, this actually seems reasonable. You know, their their operating losses were only about seven hundred million dollars, which, you know.

Brandon: Yeah, hey. I think on on sixteen point three billion net, right? That was what they reported.

Tobias Mann: Right.

Brandon: Now they also mentioned, I think, and I was gonna ask you about this, eighty point seven billion total losses, but but I understand that was an accounting adjustment of some kind. Did they

Tobias Mann: Right.

Brandon: mention in the IPO what that meant specifically?

Tobias Mann: So so not specifically. It's a little it's a little fuzzy and it's always kind of like this.

Brandon: Course it is, right?

Tobias Mann: you know, Anthropic's business is extremely cash heavy, and so they had to go to partners and say, Hey, we need capital in order to pay for a compute to train these models. we can either take it in debt or you could also, you know, get repaid in valuation, you know, our equity. And so I believe a large part of that loss is they're trying to write off a lot of the equity that they're gonna have to pay off when

Brandon: Okay.

Tobias Mann: when they they do go public. And so I do believe that's the large com you know, the lion's share of that that loss. Because it does look huge, but also when

Brandon: Right. yeah.

Tobias Mann: you think about how companies go public and how they raise capitalist startups, it's not really that surprising. It's just a huge number because anthropic is arguably the largest IPO in US history at this point.

Brandon: Right, right. But but bottom line is their actual day to day business losses was six hundred and ninety three million. But that was one point two billion. Right.

Tobias Mann: Right, and for the first time ever. I'm sorry, go ahead.

Brandon: say it was one two point one point two billion, I think, for what, the first half of the year or or same time last year or something like that. With so so it's gone down. Yeah.

Tom: Yeah, yeah. Yeah, and it's what it's what they're

Tobias Mann: Yes.

Tom: going they're they're trying to go out at two trillion now? Is that the that that the the the number?

Tobias Mann: Well, the number depends, on whether they're they're going to be able to sustain this. I was gonna mention for the first time ever, their revenues outpaced their compute expenses two to one. this has never been true in the the company's history, and it's one of the reasons that people are feeling more confident about this. The the issue that really comes from all of this is that And and what has gotten missed somewhat in in the more rosy interpretations of the S one is that they also are signed up to have to repay upwards of eight hundred billion dollars and counting of

Brandon: Right.

Tobias Mann: compute commitments out through the next two decades or so.

Brandon: Yeah, you outlined in your story just a whole host of people they owed billions of dollars to for compute. you know, and you like you said, you know, it would be great. That sixteen point three billion in profit would be great if those compute costs were gonna hold steady, right? But they're not, right? They're gonna keep climbing and there's a lot of a lot of bills that are gonna be due. So, you know, profitability is relying on making more. Than that nearly trillion dollars in compute obligation, and that's a big if.

Tom: Yeah. yeah, and and the big there's a big question mark in terms of, you know, what the technology does for inference costs, because I think a lot of this is betting on, you know, you'll be able to get the the same hardware that is much more efficient down the road and

Brandon: Mm-hmm.

Tom: maybe not drop your prices proportionately and, you know, profit on the difference. but i it it remains to be seen whether there will be enough competition to just have the price go down with the the increasing cost and and you see that a lot. I think the inference costs have have dropped pretty significantly over the year. I mean just like on on on average. I mean it's really hard to calculate inference costs because there's so many different factors and it's really situational depending on, you know, the workload and how much caching is going on and things like that. But overall it's a lot cheaper to do inference now than it was, you know, a few months ago. Yeah, yeah.

Tobias Mann: Per token, yeah.

Brandon: Yeah, yeah. But again, I I feel like if you're betting if you're betting this two trillion dollar IPO on the fact that, we're gonna keep getting better and cheaper at inference, right? I mean, I know, you know, Moore's law is kind of a thing of the past, but at this point, like, you know, are we what if we run up against another compute bottleneck, right? And and we stop before they need like if if inference stops getting cheaper before they need it to. I mean Obviously any IPO prospectus is is one big document about risk, right? hey, you might make a ton of money investing in this company, or we might go complete belly up and and you lose everything, right? That's always a possibility. But I feel like y go ahead.

Tom: Right. And yeah, and they and they still haven't demonstrated the the sort of that the use case for AI is wide enough, at vo you know, at volume that, you know, they claim it is. I mean certain sectors it's pretty obvious now there's gonna be AI for a long time. I mean, coding, everyone kind of agrees, yeah, that works and we know how to

Brandon: Mm-hmm.

Tom: to manage that. And then there will be, you know, adjacent fields to that. But you know, once you start getting into more general office work and, you know regular business tasks that are not super compute heavy, you know, I I don't know. I I it's a good question. we'll I suppose it we'll see if if people are gonna use it that broadly, but I I just don't see it being I I see being overly optimistic.

Tobias Mann: Yeah, the the the big question I have right now is is there enough users or demand and to support a even a two hundred billion dollar a year revenue stream? Because

Brandon: Right.

Tobias Mann: even going off just commitments that they can't get out of made in the first half of of this year. that's four hundred and seventeen billion dollars running out through about Twenty thirty three, twenty thirty six. you know if

Brandon: think on the upper end the upper end of your estimate there was a total, including newer commitments of about eight hundred and seven billion. At the upper end,

Tobias Mann: Right.

Brandon: seven hundred and eighty two billion at the lower end. That's not that much of a difference when you're looking at hundreds of billions of dollars.

Tobias Mann: But if they can't make a profit with a two to one revenue to compute ratio, like you know, if they have to make twice as much in revenue in order to pay for their compute and still not break even, y you just you extrapolate this out and they're going to have to be making two hundred billion plus, assuming no additional compute contracts are made during that period over the next year, which is definitely not going to happen. they're they're they've already said they're going to keep making compute contracts. So this number just keeps getting more absurd the the further out you go, unless inference magically gets dirt cheap, which we haven't seen.

Brandon: Right, right. And it could cap. I think also you made a really good point in your story, Tobias, when you mentioned that, some of their biggest customers are also some of their biggest competitors. They're leasing space from hyperscalers, they're making deals with people to host their compute. But these are also companies that are operating their own AI models. And, you know, you might end up in a situation where They're like, an MVNO, which, if anyone's not familiar with a mobile virtual network operator, it's, your your your cell phone companies that are not s T Mobile, Verizon or ATT, right? Anyone who's not those big three in the United States is operating a virtual network that piggybacks on someone else's infrastructure. And when that infrastructure is busy, you get less priority. You know,

Tom: Yeah.

Brandon: so I mean we could see them run into a similar situation there, I imagine.

Tom: Yeah, th there's an interesting trend that there's another piece I'm working on later today that that touches on this. That's apparently s some of the the AI open AI or the the AI gateways like open router are now starting to move away from sort of just focusing on pricing price metering and looking towards things like, you know, times you know, fastest time to tokens and all the sort of quality of service metrics in terms of the way that they price and deliver the models, because the inference pricing is such as sort of has become very commodity in terms of what they're dealing with. they're not necessarily s serving all the frontier models at very high prices. They're dealing with a lot of open weight models. And if you just focus on the lower end of the market, it's there's not a lot of margin there. So it makes sense to go, well, you know, maybe if we deliver this faster or we, you know, have better better reliability or service we can move up the price and get a little profit. And I think that everyone in the industry is starting to realize that the music is stopping and you know the musical chairs and they have to take a seat at some point and you know make the pro the business model work. And so there's a I there's gonna be a lot of pressure on that. And that's also showing up in things like the licensing of models. A lot of those open weight models have seen their licenses get more restrictive over time as you know companies realize well maybe we don't want someone competing with us you know with this thing that we've made. And so I think there's a lot

Brandon: Yeah, Mistral, yeah, that reminds me of Mistral just released the open weight version of what is it? I think Large Four. and they're yeah, Le

Tobias Mann: Let's try.

Tom: Yeah.

Brandon: Chunk. And the the the licensing model basically said if you intend to use this for any, you know, down market purposes, like you've got to sign a an agreement, like a fairly restrictive agreement with us, you know, so that you're not just using this open weight model to do things, you know, to without us benefiting from it, right?

Tobias Mann: A lot of the Chinese models, the big ones, Kimi K three, I believe was one of the first ones to move back in this direction, basically said, Hey, you can do if you're gonna just gonna deploy this internally, we don't care. if this

Brandon: Mm-hmm.

Tobias Mann: is just for your enterprise, but if you're going to be selling an API service like open router around it, you gotta cut us in. and so like there there is definitely that. And Anthropic in in their S1, again, every S1 reads like a doomsday prophecy. And this is one of the things that they they raise is open models are a threat in terms on multiple levels, but one of them being that it a lot of the their customers are startups who are sensitive to economic downturn. And so those customers will likely be the first to jump ship to cheaper alternatives. that are maybe not as good, but they can afford if the economy ha you know, runs into any issues, which, you know, sure seems like we're heavily.

Brandon: Yeah, wonder wonder if that'll happen or not, right?

Tom: Well and and that and that, you know, it was I think was it even Elon Musk who was tweeting about the end of the neo clouds recently that you know with the financing getting tighter, a lot of the neo clouds that are living on sort of borrowed money are gonna find it harder to sustain their, you know, to sort of their their loans and not gonna survive this. And so, you know, there will be a shake out at some point. I mean, does it hit the the big companies? I don't know. I mean you the all the cloud providers look pretty happy because they get business regardless. I mean even if Google's Google's AI models are not, you know, num you know first in people's minds, they're still serving all of this stuff. Same thing.

Brandon: Yeah, we've said that on on multiple episodes, and I know in multiple articles we've written, right? Is basically, if if the AI in industry does implode, right, Google's gonna be fine, Amazon's gonna be fine, Microsoft's gonna be fine, these guys have other businesses, right? Anthropic and open AI, mm, not so much. That's all they do. and and in many cases they're renting compute space from these other providers. So they might take a bit of a hit there, right? But the by and large, they're gonna be okay. They've got compute, they'll turn it around, they'll find other people to use it. which, you know, is interesting, again, right, from your article, Tobias, you're you're you kind of summarize that the entire pitch that this and that this IPO makes is that bigger is better, right? but as we were just talking about, right, w there's a lot of reasons to say that that's probably not the case, right? It makes me wonder if they're still kind of living in the token maxing era, right? whereas, you know, we're like I just wrote a story this week about AWS basically telling its engineers like back off, stop using so many tokens, be more judicious, use AI well, don't just use it, right? And I know Microsoft said this even months or in August, they were saying the same thing. Microsoft was saying that token maxing doesn't factor into the way they want to do business. Right. So you got two very big consumers of AI who are basically saying, y'all need to slow it down and be more purposeful. So bigger, you know.

Tobias Mann: Yeah, and Anthropex business is predicated on the idea that more compute means more revenues. However, there's the disconnect is can revenues scale fast enough? Is there enough user base? Is there enough token demand to support the scale they need to pay the chec you know, to cash the checks they've already written? and and the it does really look like a house of cards. And they warn about this in the S1, they have to, that if they can't, if their revenues can't keep up with their commute compute commitments, they could default. And that default could turn into a cascade where

Brandon: Mm-hmm.

Tobias Mann: they if they can't pay one bill, they can't get enough compute to generate revenues that are necessary to pay the next bill, and it all falls apart.

Brandon: Yeah. And I feel like if it happens to anthropic, right, the other AI frontier labs are not far behind, because that's that's gonna be a symptom of a larger problem. Right. And that's that people are basically saying what a lot of non AI companies or companies that do AI along with other things like AWS, right, are basically saying we don't need these massive frontier models for the vast majority of business purposes. They're not necessary.

Tom: Right. Yeah, and and a lot of the enterprise AI products that you see now are about trying to, you know, figure out ways to reduce that cost too, you because they don't want to run, you know, the fable, you know. twenty four seven, so you've got a a way of you have some sort of gateway or or harness system that looks at your jobs and routes them to an appropriately cheap model so that most

Brandon: Mm-hmm.

Tom: of the time you're not using the you know the the top tier intelligence you're maybe using you maybe just a a decision model or you know a regular classifier or or you know some you know something like Gemma that to just to do language translation. I don't know. It's but you don't always have to use that expensive model and that price pressure is gonna hit them too.

Tobias Mann: Yeah. actually to Tom's point, Cadence Systems, which does a lot of electronic design automation and circuit board layout and that kind of thing, just rolled out a new product line called Maestro that's currently built around NVIDIA's Nemotron, OpenWaights models. But the idea is that they had been using proprietary APIs to power their agents that assist engineers with kind of like PCB layout and and chip design. but those agents generate a ton of data that could be used to train and fine-tune open weight models. And so what their latest pitch is hey, we know you guys our customers like this, but it's getting kind of expensive for them to continue using it with these proprietary models from OpenAI or Anthropic or Google or whoever. So we're gonna give you a way to take an inexpensive model. Use the proprietary model to generate a bunch of training data that we can then use to fine-tune this open model. And then you can decide how much of the work you want being done on the expensive proprietary models and the cheap open models. And you can deploy those models on prem, in our cloud, on your preferred cloud. We don't care. Just please use our agent platform and please pay the subscription fee associated.

Brandon: Yeah. Yeah. Right. And that that again, right, I I can't imagine most businesses are gonna say, well, we're gonna fork over all this money to these proprietary model providers when we don't need those resources. it it just it really makes me wonder, how much of this of this IPO is is basically forecasting the the bursting of the bubble. I know I feel sometimes like we talk about the AI bubble bursting, you know, like we're where doomsday prophecies are, you like who are it's gonna happen soon. It's it's happening now. It's happening now. But I feel like there's been more signs recently. Like I think there's been a big collapse in not necessarily collapse, but Chinese AI stocks have been hammered recently from what I understand. you know, there's there's I think there's been some recent earnings reports from American companies that have not looked good, you know. so I see really Yeah.

Tom: Yeah. Go Google was Google was hammered with they lost like five hundred billion in in, you know, stock value based on the fact that their search was underperforming. that's even that bad, but you know, that's you know, that's

Brandon: Yeah, that's that one I wasn't able to think of, that minor little blip, right? Yeah. Yeah.

Tom: AI overviews not monetizing as well. So I I think that there's a lot of yeah.

Tobias Mann: Yeah a problem they created for themselves.

Brandon: Mm-hmm. yeah, absolutely. Google creating

Tom: Yeah. But

Brandon: its own problems? What?

Tobias Mann: Yeah. but yeah, I I I I I would agree that anthropic's survival d does seem like it depends fairly heavily on pushing the frontier forward. And in order to do that it needs a disproportionate amount of compute that it may not be able to afford or on a continuing basis. Especially if Microsoft, Google, all of these Cloud CSPs that provide compute for them and are among their biggest customers, which directly or indirectly subsidizes that compute, decide, well, okay, we don't really need Cloud for this. We've we we've reached a point where our own open weights models or our own small proprietary models are good enough for drafting emails or generating code bases for our customers. So we're gonna start buying less tokens from Anthropic or Ape open the eye, when they eventually IPO. They're likely to go through very similar challenges because they're also getting most of their compute from companies like Oracle and Microsoft at this point.

Brandon: Mm-hmm. Mm-hmm. Which actually was one of the questions I wanted to ask kind of as we wrapped up here was, I mean, do you think OpenAI is going to be any better position once their IPO comes out? Because I mean, know that neither of them own a a lot of their compute. And anthropic or not, sorry, OpenAI's attempts to build its own compute seem to have maybe kind of fizzled out or are are fizzling out a bit. you know, but could they be in a better position or do you think it's gonna be just more of the same?

Tobias Mann: I don't know with OpenAI whether they're going to be able to overcome the same risks that Anthropic has. It does feel like Anthropic is sidestepping a lot of hard questions by pushing forward with their IPO now.

Brandon: Mm-hmm.

Tobias Mann: I with Anthropic i going public this year, if OpenAI goes public next year, I don't think they can avoid the scrutiny that is going to, you know, because the world is going to have woken up to the risk factors associated with these giant model houses. And I think they're going to face a much harder time with this, even before they get there.

Tom: Yeah. And OpenAI has, got the distribution of Chat GPT, which you know, is demonstrating itself to be a useful tool that people like to use. But, you know, that's not where they're gonna, you know, their monetization, you know, gold mine is gonna be. I mean, unless they just basically become the Google of search, you know, of AI search, but Google already seems to be trying to get to that space too. so I I don't know how how how that they're gonna make it into the enterprise business and you know, get the kind of margins that they need to to really pay back all the money that they, you know, have taken.

Brandon: Yeah. Yeah.

Tobias Mann: OpenAI also just has a different business model than Anthropic. and to some a pretty hilarious degree, if you recall the advertisements during the Super Bowl, about you know injecting advertisements into to to GPT and something about Cougars, I forget.

Brandon: Mm-hmm. Yeah. I j I don't remember now. I just remember that it was a big kerfuffle. I'm pretty sure we wrote about it.

Tobias Mann: Right. But the the the point being that open AI's revenue streams may be sufficiently different to insulate them from this. they have been trying to position themselves as an AI advertising company as another revenue stream, which anthropic has flatly rejected. And so it's possible that OpenAI's business model is sufficiently different by next year. 12 months is a long time in this industry. that if if they're IPOing, if Altman takes them public this time next year, the conversation might be very different. Having said that, I think Wall Street is also going to be their eyes are going to be open to a lot of the risk factors in a way that Amade is sort of sidestepping right now by being kind of

Brandon: Yeah.

Tobias Mann: the first really big model house to go public.

Brandon: Which is, honestly I don't know, it feels sort of surprising to see open AI kind of press the brakes while Anthropic presses the gas here. I mean, I'm assuming, like you're saying, right, they saw an opportunity to get out the gate first and avoid answering the hard questions. Cause, you know, open AI basically said, Hey, we're gonna pause our IPO while we assess these risks, which obviously they've kind of been in the doghouse when it comes to AI behaving badly, you know, this year, giving Anthropic that opportunity to kind of Scoot past and not have to deal with these things.

Tobias Mann: I think the other tell with anthropic is actually that two to one compute ratio, because that's the first time that's ever happened.

Brandon: Mm-hmm. Right.

Tobias Mann: And it will be very telling if in their first public quarter or their second public quarter that just disappears. So it the timing might be related to the fact that they can now claim that their their their cash flows are on a positive trajectory towards being cash flow positive and and and moving towards profitability. OpenAI delaying their IPO, I suspect, is largely bent on they need to get to a similar point. Otherwise, people are going

Brandon: Mm, mm-hmm.

Tobias Mann: to look at this and go, this has no possibility of ever being profitable. At least with anthropics, S1, given the current state, halfway through 2026, it sure looks like they could be profitable. Whether that you know, whether that's true in a in their first public quarter is another matter until

Brandon: Yeah, yeah. I think to kinda to kinda close us out here, I think this statement from your article, Tobias, the only way to justify this kind of spend and risk is if you believe the upside is enormous, like the biggest economic transformation in the history of humankind, enormous. And that kind of I think raises my final question for you both is like is it gonna be worth it? And what's the chance that Anthropic is gonna flame out spectacularly in the next year or two?

Tobias Mann: Tom, I'll let you take for first track of that one.

Brandon: Ha ha ha.

Tom: Yeah. I don't I don't know how how spectacular it will be. I mean I think that I think it's gotten to the point where it's sort of too big to fail, but I could certainly

Brandon: Mm-hmm.

Tom: see, you know, some desperation movies where, you know, one of the these big AI companies strikes a deal with Apple for distribution because they really need, you know, the kind of money that Apple was paying Google, you know,

Brandon: Mm-hmm.

Tom: or sorry, that Google is paying Apple for for distribution. And Yeah, I can see something like that happening because th they really need revenue. I I don't s really see it just blowing up because it's it's everybody. And so they're gonna figure out a way to to, you know, har do the least harm if things go downhill. And that may mean parting out part of their businesses. Who knows?

Tobias Mann: Right. I I would I would tend to agree with Tom. I I think about how kind of what happened with the dot com boom as well as something of an indicator. Now the the the situation is quite a bit different because the kind of the AI industry has wormed its way so deep that it's not kind of the surface level situation that we saw with the dot com boom. Yeah, there was a ton of fiberlade during the the dot com boom. And that didn't get utilized until much more recently. but you know, it did get utilized. And so I think yes, AI is probably bigger than the internet and, you know, going to be go down in the record books or the history books as kind of the next industrial revolution. In much the same way the internet felt like that. After the dot-com boom, it

Brandon: Mm-hmm.

Tobias Mann: was big. It just wasn't as big as people prescribed the value to. and it took another 10 years for it to actually become the thing that everybody wanted it to become. It eventually, you know, rose to that valuation. It just took much longer than the market could sustain. and I think we're probably

Brandon: So in other words,

Tobias Mann: in for that.

Brandon: yeah. So in other words, maybe don't don't buy during the IPO. Don't buy high and hope to sell higher because you might have 10 years to wait before you get back to that IPO valuation or above. Who knows? Maybe it'll go faster, you know. no matter what happens, there's a good chance we're gonna be right here to talk about it because my gosh, this sure can fill a lot of podcasts and a lot of articles and a lot of digital ink. is gonna be spilled on this topic for years to come. And we hope you'll join us to hear all about it on the kettle. Thanks for listening.