0:00Good morning. Today we're doing our last anthropic IPO podcast before the IPO and TBD on whether or not it will actually continue covering it after the IPO because uh that becomes the domain of finance podcast. But uh I think it's still obviously very relevant to a lot of listeners particularly those of you that work at Anthropic but also those of you that work at OpenAI and uh everyone else connected in the ecosystem. Um, the IPO road show has been going on for a couple weeks and we know a bit more than
0:29we used to before. Uh, and we covered it a lot of it in the Jam and Ball podcast.
0:34That went really well. Thank you for all the positive feedback and uh, you know, I think I try to pull in people that I respect and look up to in terms of the crossover of technology and finance. Uh, because I mean I think that is if if anything my career has set me up to do that. Um so we have a price for the anthropic filing I guess uh between $7 to $78 which is an implied valuation of 1.3 to 1.4 trillion. uh it looks like it is roughly seven times overs subscribed
1:04and probably will price in the range of 80ish and then there's also forward contracts or crypto contracts on speculating where the final pop is going to end up if it looks anything like the SpaceX IPO there will be a small pop um it will be about 19 to 20% on the first day uh maybe would go up to $100 um and so that that is in that range of let's call it 1.5 4 to 1 uh 6 trillion, you
1:34know, and um that represents roughly 38 times trailing 12-month revenue, 20 times current revenue. And obviously this uh and and obviously that what what Ford revenue is is just dependent on what your projections are. It's still anticipated to uh so October run rate was 81 billion. That was also a new disclosure.
1:56Still projected to end the year at 100 billion. And what really people are thinking about is like well where does it go from here right like what is the market pricing um this is obviously the highest multiple that anyone's ever uh priced at especially at this size and uh you know ranges for the future like uh future years in like 2030 u or 2028 um really range from like 400 billion to a trillion uh in revenue and
2:25so uh that is very wide range. Um but you kind of have to do this kind of math in order to uh underwrite long-term investments in Anthropic. Um and so you can see why people looking ahead. You know, I think um uh 11 12,000 customers spend at least 100K per uh uh year and per customer, but once they land in a company, they also expand nine times um in the in the future years. So like uh
2:52there is uh NDR like a very good net dollar retention uh usually state-of-the-art is let's say 150% and this is 900%. Um so really the discussion about the road show has focused on three things. One which is cloud co-work um it is basically the sort of next phase as far as the market narrative is um how how well are you extending outside of cloud code. I think there is probably going to be a different form factor I would say than
3:22cloud co-work because cloud coowork is you know we did the first cloud co-work pod with Felix um if you scroll back in our archive and it's a it's a form factor that's there but I think it's not very cloud native and definitely the sort of autonomous like teammate that they are pitching for their S1 that we covered in the first podcast that is um definitely going to have to be cloud native that's cloud duet that is the instinct that is the gro
3:50bots that is the cognition bots all those things are like everyone is trending towards that direction and I don't see why cloud would not do that and probably that is going to be something different than cloud co-work uh the second one is chips uh entropic wants to run across Nvidia AMD TPUs tranium um and they also released information about how they were easily able to migrate much much more easily uh from Nvidia to AMD And I that is obviously part of the
4:19whole RSI narrative that people are concerned about. But I think that is obviously just makes a lot of business sense for them that like hey if you're I'm bottlenecked on Nvidia I'm sorry I I have to do this right and by the way it's verifiable and by the way it's uh one of the core tasks that we that is uh that that is easy for LMS to do. Third is margins. Um so we have a little bit more in terms of disclosures on margins.
4:45Um we know that margin went down from 19% to 15% and using an adjusted margin.
4:51Uh it went from a gap operating profit to a slight gap loss and that is just intentional not a big deal. Um Anthropic is projecting that it can get above 35 in the long run. Uh so that would be a very big deal because then it would be very profitable and that's something that uh again like we as we discussed in uh this podcast in the previous one with Jam that is just unheard of for a large tech company of this size IPOing. uh usually you're unprofitable like deeply unprofitable like like 100 negative 100%
5:20profits for I don't know 3 4 years before you even start to turn it around.
5:27Um and that's good. I mean this is growth with profit which uh never happens. Uh I think the final thing that happened this week is opening eyes model launch. Um as we already established model launches are incredibly incredibly important to the viability of the Frontier Labs. Um the shelf life of Frontier model is roughly 6 months and if you don't release the next model at the frontier then obviously the usage is
5:56uh not going to come to you and it it is very sensitive to benchmarks and uh vibe checks and so open shipped GBT 6.2 and Fable and topic shift Fable 5.2 and obviously Fable 5.2 is still on top and I think the vibe checks are mostly holding out that Fable is on top. So there's that and then there's the efficiency frontier which open does claim. I think people are a little bit more mixed in practice in terms of there's benchmarks and then there's just
6:24like impractical use and um I would say that there's basically both model launches have not really changed camps either way. If you're a cloud person, you could continue to be clawed. If you're happy with an open AI, you're continue to be open AI. Uh those key risks for the IPO that may have dramatically changed the story either for the positive or negative haven't really changed. There was no Fable 5.5 that people were anticipating and um there, you know, and Gemini shipped
6:53Gemini 4 that did not affect the narrative even though Gemini carbon did very very well. And that's about it, you know. So like those key risks are behind us for the anthropic IPO. Um Enthropic is still the top model lab in the world and uh it is IPOing tomorrow and it is going to be a blockbuster IPO and that's really good for the ecosystem because a failed enthropic IPO would have been uh would have read through really badly together with the Nvidia drama that is also happening at the same time. Um so
7:22that is the overall update in the last week. You are mostly prepped. This is our podcast that we recorded to go through some finer details in the S1 and I hope you enjoy and good luck tomorrow. Okay, we're in the remote studio. Uh this is another in our series covering the enthropic IPO uh with Tane Japur.
7:46Uh thank you for having me. Longtime listener of the podcast and excited to be on.
7:52Uh T, how come you're uh writing S1 breakdowns?
7:56Yeah. No, I think in in general I guess uh maybe for the last uh six or seven years I've been uh you know writing about various things like more on long form on on Substack and one of the topics I've been writing about just like breakdowns of interesting companies as they go public and I think the S1 is usually the time when as as an outsider you get a look into like you know all aspects of the business and the the reason I enjoy reading them so much is that you can kind of get a sense of like
8:25the story of the business so far and I feel like in every business there are some interesting lessons that you know for people to take away about business in general about you know specific kinds of businesses and so I used to follow them from that perspective and then obviously over the last year or so some of the you know the ones we've had go like with SpaceX and now with the Anthropous one these are like you know truly generational defining companies and so uh I think it's been even more uh even more fun to to look into those and
8:54I think they do give you a sense of the current state of um those companies specifically but then also the market more much more broadly and so with the anthropic S1 kind of gives you like a purview into everything from you know this this the idea of like you know the model development inference everyone's talking about like inference as this you know the next big huge market uh it gives you a sense of you know the current state of all the compute uh needs and like neoclouds and then also what's to come in the future with things
9:23like science and uh some of those things. And so I think it's obviously like a category defining company and one that uh and one that's still relatively early in in kind of crazy that in you know in five or six years it's uh created this much value so quickly uh and it's something we've never really seen before uh uh across any company really.
9:45Yeah, for sure. Uh and you know I I'm I'm happy to dive right in there. you have a lot of charts that are um have been prepared and I can go wherever you want to take this headline numbers um you know they're they're ramping revenue a lot it's really interesting basically up like 13 14x from uh same quarter last year I think the run rate numbers just leaked for October which we knew was was uh something that they had in store and they they kind of left blank in the S1
10:14but now uh they're reporting 80 billion run rate right um yeah So I think it's 80 divide by 4 is 20. Yeah.
10:22Yeah. And I think the the my guess is the way they calculate that 80 is this kind of more like on a monthly basis. And so I think the exact definition they might use is actually 28 days times 13 plus like monthly subscription revenue.
10:35And so it's it it's interesting because I think it's being even on Twitter I think the reactions to that will mix where there are some people saying oh this is like a very good very positive number and uh signals reaceleration.
10:46uh whereas some are saying that it's actually they'll it'll mean that they likely will fall short of the 100 exit run rate that they've been talking about. I I think it it's hard to tell but I think it probably is in that kind of 8 to 10% month overmonth growth which I think is I don't know if it's re acceleration but it's at least roughly in line with what they did in the last few months and it might be slightly up is my is my guess but hard to hard to fully know at this point because they haven't broken down you know Q3 and stuff by month.
11:14Yeah. Yeah. Totally totally. uh 80% consumption, 20% subscriptions, which is kind of funny. Uh it's like it's a pretty public secret how much of a discount the subscriptions get. Um so actually people are incentivized to make this 80% consumption base lower, but uh you have to sort of figure out the subscription system.
11:34Um and uh you know number of customers over over 100K is up eight times to 12K. Um yeah, you know, like I I'll let you drive like what what let's assume that people have looked at the headline stuff and they know about the profitability.
11:49What are the major topics that uh you think are the the main messages that people should take away?
11:56Yeah, I think like you know at various points across the last few months there have been all these leaks about you know roughly like their run rates and stuff.
12:03So I don't think like you know to me like it's kind of like those things were not as much of a surprise to anyone that's kind of been following this in terms of uh you know what to expect based on what has been publicly reported even before the S1. I think perhaps the biggest surprise to people was just around the cash burn of the business and and in I mean that in a positive way. So they they had raised like you know $140 billion of cash and literally like they
12:31still have on the balance sheet like 132 billion or so. And so they burnt like under $10 billion to get to where they have. And I I think it's like one no one I don't think other than you know obviously the people that actually knew the numbers from from being on the inside. I don't think anyone expected that they that the cash war chest that they had on their balance sheet was actually that large just based on uh you know based on expectations around the burn and the uh margins and things of of
12:59the business. And so it's actually been like a very very very efficient business. Um and um uh I I kind of had just did this quick analysis around you know given the scale they've gone to right which is you know roughly 70 billion at the time of you know if you do like the quarterly times four what's the capital consumed in terms of like the cash burn per dollar of kind of revenue run rate at the time of the IPO and there are very few businesses of any meaningful scale really that we've seen
13:28go public over the last five seven years that are anywhere close like I think Airbnb is probably the only one uh that that kind of comes close on a revenue run rate basis in terms of the cash efficiency in terms of the burn. And so I think this was something that was quite surprising. I think even from like an absolute dollar basis like I think Anthropic has literally burned like less dollars than Uber did to to get go like at the time when they went public which uh you know given what they've created with this is is you know pretty
13:55remarkable. Uh um and uh it's like you know the cash burn. I was thinking about this like like Meta is still burning like 10 $10 billion on their reality labs initiatives every like two quarters or so. Uh I think it's now down to like 16 billion a year. So maybe they burn like 8 billion but they're still burning every year. Way more absurd.
14:14Yeah. Than like you know this company that got off the ground and like built like you know basically built like long horizon agents and like you know was the first of so many of these things.
14:25Oh, wait. I expected them to to take it to zero. Like, what the hell?
14:29No. Yeah, they still losing like to photo.
14:36It's like they're burning like a million dollars per spectacle for glasses or something. Like, what the Like, um that's that's crazy. So, so I expected so my my actually pitch for MSL was look the uh they were burning something like 20 billion uh per year uh on on on on on VR and you just cut that 20 and reallocate to MSL uh and you you have a Neol,
15:05right? Like like and which is they they now have it. It's just that his pockets are so deep that he can just do both.
15:12Yeah. He was like, "Oh, I'll do that, but maybe I won't do the cut fully and like I'll slowly do the cut." Yeah.
15:17I mean, like clearly like he's enjoying it. He's loving life. Uh whatever. But like uh you know, these glasses are nothing compared to Muse, right? Um Yeah.
15:27So, uh it it's so funny. Oh man, I can't believe they're still burning five billion a year. Uh five billion per quarter.
15:35Yeah. Uh come on. All right. So, that's the cash burn. You know, I think um that's something that um I feel like is underrated. Uh so the the one number that you're not seeing on here is where's OpenAI, right? Uh because we don't have the Open S1. Um it is way more inefficient. Uh it it is probably closer to like the Snowflake Cerebras Core Weave range. Uh and there have been unofficial charts. So this was this is like not a completely new fact. if people were
16:03paying close to the the leaked S1 where people were already drawing um charts like uh Enthropic was let's say doing uh was was was known to be doing like this uh as like a like a projection and then and then OpenAI was like kind of like way deeper and like yeah something like that. Um and so so this is a known chart that like we just didn't know how big it was. Um yeah the I think the magnitude wasn't
16:32Yeah. Yeah. So, so, um, and I think I I attribute this to to a couple things, right? Um, Enthropic started later.
16:38So, they they they basically skipped GPT123 because that was done on Open Eyes. Um, and you know, that that R&D Yeah. It was done on their dime. It's the it's the same founding team.
16:49Um, so like they, you know, they they already paid the tuition there and then now they come out and and they can just kind of repeat that work from a a clean room perspective. Um but you know what else like what why is why is anthropic so efficient? I think that is like a really really um underexamined thing. Uh partially I think it's just straight up like there are now unit economics profitable uh which is cool. Uh but they also didn't burn that much to get there which is super unusual like like basically
17:19Yeah. Yeah. Yeah. And and if you go to the the next the the chart right below around kind of gross margin I think and this also probably contrasts something with uh with open AAI which is that uh because anthropic focused and ended up focusing on coding and on enterprise maybe more generally and more just like let's have the best models and uh both on let's distribute primarily through channel partners but then also let's leverage them for compute wherever possible and they would didn't have this because they weren't trying to do
17:49consumer they didn't really have this like massive consumer base that they had to support uh with like you know mostly free inference. They the the economics are like very very uh positive for them on on kind of like you know on supporting like their paid users. And so I think the most common number like you know is like this kind of idea of 85% gross margins that they report and the way to think about that is that essentially like as a percent of revenue their like inference on paid users is
18:17about 15%. Uh and there's a chart further below that kind of breaks breaks that down. Um uh and so that you know they reported yeah they reported these like 85% gross margins on you know essentially paid inference. Uh but then you know one thing that a lot of these companies obviously have been doing is that there's these like free free free free tier users and it it's been common to put this into sales and marketing spend because you know you could argue that it's like oh it's our like it's how we attract users to a platform and then
18:45we monetize them with if they become paid and so they put that free tier inference in sales and marketing spend and so you could be like oh maybe that should go in cogs and but I think the point is that for Enthropic because you know they primarily relied on enterp price and on coding and claude is not as big as like you know chat GPT is from a margin perspective and maybe like a cost uh cash bone perspective that actually is a positive for you because you don't have this huge user base you have to
19:13support on on the free plan and so they only spend like you know only about like 4% of margin points or about like five six% of their compute spend ends up going on like this free user inference and so I think that's kind of like the other aspect of it I I imagine when the numbers come out for OpenAI that's going to be significantly higher just given the size of that size of that user base.
19:35Um yeah for do we do we is four 4% is small right for any free tier. I I don't know how you you look at your other uh S1's and investing. Um usually free tier is like people just kind of treat this as marketing right like this marketing expense. Yeah, exactly. And and so obviously this is specifically on the compute usage and so it's hard like there aren't that many companies that have like you know that are public that have like free user with compute plans, but I think essentially like they're not using up a ton of their
20:03compute on serving these free users. And so most of it is either being used for training and research or is being used to support these paid users which I I think like you know if you if you think about like you know remember OpenAI getting rid of Sora uh one kind of reason for that I think was just that it one it was obviously not really monetizing because it didn't intend to but it was such a use it was such a large use of compute that it was like you probably have more productive uses of that especially given that they weren't really trying to monetize it
20:32into some kind of like you know like AI video gen platform. form truly. Um, and so I think you see that with with with Enthropic.
20:40I I think there was a case for, you know, Sam has always wanted to build a social network for OpenAI.
20:46And what do you do if you're not Twitter, you're not Facebook or Instagram. You do AI Tik Tok, right? Um, and that was a reasonable bet and it didn't work out. So, I don't know.
21:01I I think it was worth a try. I actually like I'm kind of like an open AI defender here in terms of yeah you can do coding in B2B SAS or enterprise but like you know AI is supposed to be a bit more broad and creative and weird than just you know we help you code better. Um and so I I you know I defend like the deep minds and the the opening eyes for doing things that are not you know just work.
21:27Yeah. And I think it's one of those where I think that's the Anthropic has worked partially because of its focus, right? And and if you see where they've gone to that that that's helped them a lot. But as you think about uh one big question that people have and you kind of start to see it in some of the numbers is that hey like Q2 was this like generational quarter for y'all where yall added like you know like 25 billion of like you know like on an annualized basis of like run rate revenue and already you see like a a decline in Q3 both like a massive
21:56slowdown where you went from like you know over 100% quarter over quarter growth to like 40 45% or so quarter over quarter growth but then also you ended up um now that the people are like oh was at just the quarter of like max token maxing and peak kind of like entropic share because no one else had caught up on coding and on like a good like you know like reasoning model like a long horizon reasoning model and you just basically captured all of that that market available and now everyone's kind
22:26of caught up now open wave models are strong token maxing is ending everyone's talking about like you know routing across models and so you're going to suffer from that without maybe that next leg to go on and I do think that is parti actually one of the risks I don't think it's as big as that in the sense that it's we're so we're still so underpenetrated in terms of just diffusion of you know this core technology that's already available to us but I do think there is this big question of what is is there going to be like the next thing off to coding that
22:55is like actually the size of this of coding where um someone you can imagine like mo some large base of people are spending like $8,000 a month or $10,000 a month on kind of tokens off that kind of for that kind of use case. Um it to me I think what the the if you think about just like you know product market fit of various segments like obviously I think coding is number one. I think customer support legal have probably
23:23been some of the others. I think the the one that seems to be coming and there have been various debates on X about like you know what percent of token use next year is going to be on around security. I think that could be another one where in some ways these labs have kind of created the problem but then also offer the solution where now attackers have access to all these models and can kind of go ham on on these. And I was starting to see some of this play out, right, with some of the PHP stuff and then just like more and more of these hacks happening and you know some of these like um the models
23:53that are like uh um ablated and like that allow you to kind of really uh use these and so I can't I I I see that as like a promising area in terms of like you know what's that what's another like massive source where like a lot of tokens can be used. Um uh but then I think what they talk about a lot uh for long term is around like science and uh uh uh life sciences and uh you know let's uh maybe use these to start doing
24:20things drug discovery and and all of those things. Um and and they touch on like you know some of the uh the wet lab space they have and and and those things as well.
24:31Right. Right. Um we we'll leave the weblab stuff to the end because obviously that's like newer and and less um uh we honestly just have less details about it. Um what other what like Okay, so you know we've talked about a lot of the headline financial things. Um what you know what really surprised you?
24:49What else? Yeah, I mean I personally was pretty surprised by in some ways the the share of the compute that was on training and research still like like you know it's basically they're using about 70% of their compute on training and research and only like 30% is going towards inference and that I think we're surprised both one that just means you know the inference is very efficient so like you know it's going right it it was 50 at at uh at the start of last year exactly and so I think they're really
25:18squeezing inference efficiency out which to me I mean if you think about it right a business of the scale of anthropic if if their quote unquote infer inference costs are only like call it like 10 billion this year maybe like10 to 12 billion a year that's kind of crazy if you think about it like and it it does make you question a little bit about like some of these sizes of these builds out build out more now again we're still very early in aggregate but the fact that they've gone to this scale and they're only spending like you know 105
25:47billion a year on on inference to support that size. I think that that was surprising to me. Uh I think the other piece of it was that um they kind of they didn't technically do this, but they vaguely gave you these numbers that let you break out like the data the spending on data. And even that to me was surprising in that it was until this past quarter, which they only kind of published through the amended filing like a week or so ago. It my best estimate on it was it was like about like 900 million or a billion of run
26:15rate spend on data, which I thought was actually quite small. given that you consider this how many of these companies are in there in Silicon Valley selling like various forms of data like RL environments of all these different niches and the scales that like you know that so many of them are at uh and so that was surprising to me. I think the one thing you did see is that there was at least if you on a quarterly basis there was a big jump to I think they spent about 600 million or so in the June to September quarter on on data and
26:44so it seems like they have very recently ramped up this spending quite a lot and so today you could argue they're you know now spending close to like two two and a half billion a year uh run rate on data. Um, and it it's interesting because now it feels like, you know, there's been this renewed interest in all these like buying operational data from companies and like these data brokers that are helping get like real company data onto people. So to me it's a sign that like this the this kind of RL environment paradigm we've been in and like just like post training and
27:13using RL to keep scaling across every one of these domains uh is working and they're seeing good results from it and they're continuing to ramp up their spend on on on that across all these forms and so I I thought that was interesting well and it roughly the spend on data has more or less been you know like five to six% of their spend on compute over time although it it kind of did have this jump as I mentioned in the last quarter from like and so it went from like 4ish% compute to 6%. But I think that that's another thing to
27:43Yeah, that's something I've been I've been tracking a little bit. Um do you think that scaling it by percent compute is the right way to do it.
27:51Uh I I I don't think so. I mean I think like one is just like what's the absolute number and like let's talk about that. I think but if you think about if you think about it as like it's it's ultimately a core input into training maybe percent of training compute could be a good like percent of compute that's going on training and research the other thing that's not captured in this is that this is the spend on external vendors that I my guess is there's also a sizable just like or not my guess but there is a sizable synthetic data effort internally
28:18as well to synthetically generate data that is probably just getting captured in training and research compute uh uh uh uh um as well. Uh and so so that that is like you know the other aspect they're also buying books and uh destroying them.
28:33Yeah. And so I think this was the they had the separate line item in the for one of the quarters in the S1 for this the settlement that they paid for 1.5 billion. What was this?
28:44Yeah. I I didn't touch on it but it's I think it's called like the bots settlement or something. And so they in one of the quarters there's a separate line item on the in the income statement for for specifically that. Um, and the thing I found funny with the whole uh with that whole thing was that the issue was not that they uh or the things that they ended up getting uh fined for, I believe, was not the ones that they actually bought, but that they tormented a bunch as well. And so it was that not that you actually used it. You bought some and you used it, but was more that
29:12you there was some that you didn't pay for which you used. And so if they just like bought all of them, I think you may not have had to pay the settlement.
29:19So uh I have a tiny bit of context about this. Uh this is actually probably from books three. Uh I don't know if you know the context cuz this is kind of deep lore and I can cover it briefly for people who are interested. Um so basically GPT ones and one and two was trained on this data set of books one and two and it it's literally called books one and two. You can go look it up. These are sort of uh copyright clear books. Now uh a random person uh I forget his name now. I think
29:49like his name is Sean as well and just from like nowhere just like open source AI person goes like hey books one and two are really good idea uh let's keep going and so he like creates a very very illegal data set of like actual books including like Harry Potter and whatever and calls it books three right uh you you can call things whatever you want but clearly you know books one and two were like official copyright clear books three was not books three but how however books three was like it's like
30:18pretty widely cited or used uh in in the model trading circuit. So um I don't know if enthropic was exactly caught by books three but it's going to be something like a books three where uh it was included by somebody in some corpus that was just like okay like you know researchers got used to training on this for like an academic project where it didn't really matter but now it matters for topic and so this decision cost 1.5 billion.
30:41Yeah. And I'm I'm sure after that settlement they they're going to be a lot more careful and they have much better prominence on everything else that goes into their training uh training sets.
30:53I mean so so that was the whole uh discussion that I think should have gone to the Supreme Court. It did not. Um I think yeah I don't I don't think it went to Supreme Court. Um but uh it could be transformative use like we literally call it transformers, right? like like we we are I am allowed to go and read Harry Potter and like maybe I have a perfect memory and I can go somewhere else and write down word for word Harry Potter from memory. Is my
31:24Like like that's that's like effectively what we're doing. Uh you know just because there's there's machines involved and you can kind of view LMS as databases like maybe that that's like a bit more questionable there. But um you know I I do think that there's some amount of dual standard or double standard when it comes to treating humans versus treating AI differently you know.
31:46Um anyway fun questions for now.
31:51Um so so yeah I think the other thing that I always press people on is like what percent is cloud code? Do we care? You know we used to have these disclosures of like oh cloud code is like 1 billion in revenue.
32:05CL code is two billion in revenue. The last update we got by the way was two billion in February or March of this year and then they stopped reporting right. So when people stop reporting usually you assume the number is bad. Um but the the truth is that cloud code as a CLI stopped being a thing like it basically just got adopted into cloud tag and cloud co-work and you know now like three out of the four products that they list on the S1 is like cloud codebased things. So like arguably
32:33cloud code is responsible for I don't know like 50 billion in revenue.
32:39So I I I don't know if that matters. I you know it's this is one of those things that like people debate.
32:46Yeah. Yeah. I think like the other thing I I would have liked to see but they did not disclose which I don't fault them for not doing so is that what percent of actual revenue is coming from you know like these first party applications that you've built on top like the true like you know like like like the actual like you know the end of like versus like you know someone yeah or just or or just anyone using your if they could use it directly from you but like like a hobby using you and their AP like via an API is pretty different than a
33:14consumer using like cloud co-work or cloud code or something where you kind of have like one more level of uh um abstraction that you've kind of built which allows you to like you know arguably have a little bit more of uh retention over time and the reason I say that is like a few things so one is so we do know for for example that about half their revenue is coming from channel partners I think that's actually now going down slightly it was on the way up I think it's now down to 44% it's coming from these channel partners and
33:44that they rough roughly are paying if you think about like the big cost right it's pretty much compute which we've touched on people which again is is as a percent of revenue just keeps going down because they are getting more and more efficient I think the revenue per employee is now up to like six million per per year from like from a million or something um and then the third is this kind of the sales and marketing but most of it's not like you know brand building sales and marketing it's like these the fees that paying these channel partners
34:11which is about 30% of the revenue 30 to 35% of the revenue that those channel partners are getting for them. Um, but I think the other reason this important is that they disclose that about 80% of their uh most of the model usage uh or most of their revenue comes from models that were released in the last 80% of the usage comes from models released in the last six months. And I think what that the dynamic that kind of ends up creating especially when you add to that fact that most of their revenue comes
34:41from usage based plans or from usage based pricing as opposed to like you know subscriptions is that you can't fall behind at all or you you you kind of are putting like a large percent of your revenue at risk. So like if like and I think it would be hard for this to happen but if imagine like for whatever reason like they just fall behind for 3 months and you know there's like you know Gemini's new model is very good, open's new model is very good and um and Anthropics is just like slightly behind
35:10for a reason and it's like you're still using Opus 55 is the best that they have and the others are further ahead. you could find yourself in a scenario where because people are just kind of using you at the API layer in some system where they can easily just swap you out that you have lost a lot of shareh suddenly and lost a lot of revenue and so I think that's like a tricky place to be in and so from a company strategy perspective I think you know the more they can get people to use your apps
35:38like if you're using you know claude you're not going to switch of like or switch all everything over to chat you might have a bunch of scheduled tasks and you won't do that switch if like you know it's behind for a few months or something but if it's the more it's just APIdriven usage the more risk there is of just people uh you know just like the revenue can go away more quickly as well and so I do think that that is one of the things where you're in this game where you just always have to stay uh and not necessarily at the frontier but
36:08also just at the like at every point of the price performance globe that kind of matters from a usage and workload perspective. Yeah. Amazing. Um I I think that's that's a really good recap of of the state of things. Um okay, so it's been it's been a couple weeks since the S1.
36:26Um a lot has changed. Uh we are now tracking um what is it like 50 something% uh probability of IPO? How critical of a moment are we in whether this or not this happens or not? Right.
36:40Um if if they don't IPO, is it the end of the world? If they if they do and it still flops, is that the end of the world or is there no end of the world and actually just keep on the fundamentals, screw the forget the stock market?
36:53I think that my my guess is that it would probably it definitely would be better for them to even go out sooner rather than later even if that means like you know 10 20 30% like a lower valuation. uh in that like like you know indefinitely delaying it might actually lead to more fear and uncertainty than just like you know going out at like 1.25 instead of 1.5 or whatever that ends up being. Um I I think the the other reason this ends the
37:22probably the area this also ends up mattering in is just that so much of liquidity in the in the ecosystem rests on you know these the IPOs of some of these these companies and SpaceX has is the one that has gone out and I think a lot of the continued funding of these very large rounds like you know it's like whether it's like these Neolabs and some of the other like you know like the comput buildout and ecosystem stuff also just relies on hey like these things actually going public and providing
37:52liquidity to uh the venture community and then to the the RLPs which then allows them to invest back in the next funds of the VCs that then allows them to invest in the next set of companies and so I do think from that perspective as well it is it is important now I don't think it's going to be the end of the world if it happens in January instead of November but I do think like you know like it probably from an aggregate perspective and I I I wonder what the what the anthropic things, but it might actually be better to go out
38:21now like you know like before the end of November as an example than maybe wait for the new year even if that does mean a a slight like kind of discount on the valuation. Um uh yeah cuz I I do think that the fundamentals of this business strongs I don't think I still think it will be the first of the uh between like it will be it'll be easier for them potentially to go out than it would uh and open AAI and I think if you're if you're anthropic you probably still want to go out in this this year you've had
38:50where you're still growing like you know at least like on a year-over-year basis like this uh enormous amount because I do think what what they what the risk for them is that you could have like some month where you have like some negative month overmonth growth and then that could actually freak people out a lot. Um uh and so you want to go out while you're still growing this you know 5 10 8% a month and uh uh uh um and already like people aren't viewing this
39:18as oh you went 10 to 100. I think that is true and they likely will go 10 to 100 but there's also the well in the last three four five months you actually you know your your grow your the growth is declining and so but it still is growing at amazing pace for the the scale it's at so you still want to kind of get out in in that period um um I think yeah you know like uh look there's the short term like you know is it this month is it next month like look if it's within the next like six months like where you know we're mostly it's like
39:46rounding error um when employees are I think they, you know, I I talk to developers and engineers and researchers all the time and they're asking me like, look, okay, it's 1.5ish.
39:58Um, and so now now they're looking for the next like 2x, the next 5x, right?
40:02Like is it three 3 trillion? Is it 7.5 trillion? Is it 10? And um that's really like the big unknown, right? They had this big jump in the past year plus. Um what is the ongoing prospects, right?
40:15And so, um, that's the other thing I I really want to emphasize, which is like they're trying to also tell that story about what's the long-term growth. It's basically agents, uh, they call it virtual collaborators in the the S1. Um, you know, it's and that's a that's a bet that you have to take, right? Like obviously the S1 is not going to tell you anything beyond like recent financial performance and maybe some risk discussion, but that's about it.
40:39Yeah. Yeah. Yeah, I think this this idea I think this question that people have been debating around what is the right amount over time for a company to spend on kind of tokens and and and and because I think it's like you know you people will have a view on that of like how big is this kind of spend on tokens going to be and then they'll have a view on like what is entropic share of that going to be and then like how is that going to monetize based on where you are in the frontier or like you know the profitability of that. I think other
41:09than stat that was interesting from there one was that only today only about half the Fortune 500 spend over a 100k a year on entropic. So I was just wondering like what what are the other two like how are there like still 250 companies in the fortune 500 that aren't spending 100k a year on entropic like that's kind of crazy to me. Um and then the other one was that like off the ones that are spending this over 100k. off that first set. Uh, Enthropic is is under 0.1% of their like aggregate
41:37expenses. And so it's like there, you know, you would think that everyone has different views on this, but what is token spend as a percent of opex of expenses for these companies over time going to be? And it's obviously going to vary depending on the kind of company you're in, but my guess is that at least it's going to be like, you know, in the few percentage points and even that could be like, you know, under counting and I'm talking about like, you know, three, five years from now. And so there's probably like a 10 to 30x if you just think about it on just what you're doing now. Even without that much
42:06further, you don't need like you know the intelligence to get smart. It's more maybe just the things around it and the actual like you know this this idea of the diffusion gap or the uh uh capability overhang that people talk a lot about. I think as that gets addressed I think this the token usage should just continue to grow. But then I think then the other questions become like okay like what is the if most of the you said you don't need these like the frontier models for will more of that go to open weight will more of that
42:34will will the margins on those be low because they're much more price competitive versus like the frontier models where you know people are willing to pay higher prices and stuff and but those might only apply to a smaller set of the use cases than than like you know everything and that's being used in the enterprise.
42:51Yeah. Yeah. That's great. Uh, cool. Uh, I'm going to play a little bit of like a read around the broader ecosystem and then we'll end with the wet lab stuff and the science stuff. Um, does are you bullish or more bullish or more bearish Neolabs and Neo clouds? You can you can sort of interpret that either way.
43:12I I think that a at the current pricing of the NeoClouds post some of this uh dips and stuff probably more bullish on th those ones from like a from if you were just to take like the average one like how that'll turn out. I think if you were to take like a basket of like the Neoclouds versus the Neolabs, I think what's going to happen is that you might you may end up having one of one or two of these Neolabs becoming extremely extremely significant
43:40companies. Uh and so I think that basket will may end up like outperforming the Neocloud basket. But if you just take like your average one like you definitely are better off with the Neoclouds, I think. And I think the other thing is like many of these neolabs will end up being the neoclouds because they've already bought the compute and if they can't find a productive use for it, they end up Exactly. Yeah. Yeah. Because most of them I think ended up, you know, ended up basically like securing like long-term compute in one way or the other. And so now they could end up if
44:10they don't have a productive use for it, then they'll end up being the the supplier of it.
44:15Yeah. I mean that that's the you know that's the failure mode of a neocloud um or neo lab is to become a neocloud that's for sure.
44:24Um but I do think like just in terms of the profitability of inference we have now have a couple data points of how profitable anthropic inference is how profitable deepseek inference is because they released the deepse inference stack and they they said it's something like 80 something%.
44:39Yeah. So like I think like that should make all neo clouds just gener generally more uh interesting as an investment because not only do we know that anthropic is um profitable but also the state-of-the-art stack with Deepseek is is open source and so you can just buy a rack of GB300s and go nuts and obviously it's not that simple but um you know like the payback period for
45:06a GPU is not 10 years it's Not four years, it's not maybe not even 3 years, it might be a year.
45:13Uh yeah, I think SpaceX SpaceX on their owning squall said that they're seeing the sub sub oneear paybacks on their on their buildouts, right? Which is which is kind of crazy.
45:22So So then like why are people skeptical? Like why why are people like why why is if that is actually true? So that this it's probably not true. And the reason is the debt markets are the smartest people in the world, right?
45:35They do by far the most due diligence and they are hiking interest rates on all these guys, right? The whole reason we're raising equity is because you can't raise debt.
45:44I don't know. I to me I think one aspect of this is that obviously we are in this moment in time where definitely there is this you know imbalance of supply and demand and so I think people just have a different view or or the different view in terms of when that does end up clearing. I also do think that if you look at like you know anthropics inference numbers I don't think that the at steady state or even right now for example like the open-source inference market where it's like you're just
46:12serving inference on these on on the models that are open source like you know you you see different numbers thrown around for the the atcale players there right like the fireworks the base stands of the world uh but I I don't think you know I I don't think at steady state these things end up close to like what the entropic numbers uh for inference just because like that model is available to everyone and someone will will price will will undercut you on price to kind of just get share and I kind of see this playing out. I think I think open router is just
46:42like fascinating to look at like you can go look at um you know like go look at like a Kimmy K3 or you know GLM53 like it doesn't really matter what model you look at and you'll see like the set of players on like the official you know the lab itself that made that model like a Kimmy and like the price that they charge at and there are like five and 10 plus like players now that are basically like at like 80% on like basically 20% of that price or even 10% of that price but they are like undercutting the, you
47:10know, the the the little what the price that like the Deep Seek themselves might give it at or that like Miniax or whatever might give it at that, they're like undercutting them by like 10x and these are like smaller players, but I do think you're going to continue to see this trend of like, oh, if you try to charge too much, especially in the open source model and depend on the licensing of that model requirements, like there'll just be like some new player like, oh, I have like I'll just capture share now and uh and I'll run it at like 20% gross margins or even 10% u and and
47:39as opposed to and and just like capture that share. And so I do think it'll be interesting to see how some of that plays out and settles out. My guess is that I I don't expect any open source inference providence provider to have at least like maybe o over at least like 40% margin longterm on like the vanilla open source tokens which I think is partly the reason why you know you see many of them uh pushing into things like post training and uh uh helping customize models because then that
48:07becomes a slightly different uh more lock in slightly different model you're serving and you can maybe uh you know you can benefit from like as you optimize it further and the maybe charge higher have higher margins on those.
48:20Awesome. Um so I just want to wrap it up with the the wet lab and the sciences uh work. Um Daario has been pretty open about his personal relationship with uh AI for science. Um we haven't seen that much yet. Um what are people first of all what was disclosed and what uh are people talking about there? Yeah, I mean so maybe for a step back I think the the other thing is like you know like the the obviously signed like these like 900 billion of comput of compute commitments
48:50which has been you know one of the kind of like eyepopping figures in the S1 that has some people worried and I think to me the question for seeing the applic they have like a rough schedule and so let's say that you know next year they might end up having compute expensive 70 to 80 billion the following year onwards it ramps to like more like the 100 120 billion range And so the question becomes like what happens given that you know these are like mostly contractual other than the SpaceX one. Uh what happens if you don't have enough demand
49:20to serve it which is you know like is obviously not a great scenario but it could happen and the reason it could happen is just that um uh it doesn't have to do with anything around even just competition and things. It could just be that diffusion takes longer and so you don't have it's it just you know beyond coding that next thing isn't using as much tokens yet you don't have that next thing and so I think the life sciences stuff is important because it provides like a some use of that compute
49:47that is more in anthropic's own control where you can focus your models and then hey like let's we will be the primary user of our own models and and help with you know you could partner with companies obviously, but like you'd be then looking for these novel discoveries and and it doesn't have to be just life sciences or it could be any kinds of things in science, new materials, uh new things like that. And so I think it's really important there because it it it it then means that at least you have an avenue where you can direct
50:17compute to which has another way of being productive that isn't dependent on timelines of of enterprises and uh and others to adopt your technology. And so if this diffusion gap ends up bit being long being wider or takes longer you have this kind of this other business segment where you can direct some of the uh compute to and so I think it's important from from that perspective um uh uh beyond just like you know just further the other obviously big use of computers RSI as well and and just
50:46making the models better and better but I think that's may not still solve the customer demand piece whereas like having like these life sciences bats and I think they also touched on robotics a a little bit but not as much as life sciences. Like having those then also puts you in a different um having some of these other things that are expanding to physical world but then also things that may be a bit more in your control where you are operating these to try to find like make novel discoveries and that you yourself are you know potentially with the farmer partners
51:15taking through the the the pipeline.
51:18Yeah. Awesome. Um, all right. Any parting thoughts, um, messages for people, founders, you know, talking to Wing, whatever.
51:28Uh, no parts. I mean, I think it's going to be very interesting to see how this plays out. And, uh, I'm curious to see what the next couple months look like.
51:36And I think until they go public, the the number that everyone's going to be watching very closely is just like the month over month the the actual revenue growth for them. So yeah, it's going to be which is why it's better for them to do it sooner rather than later. Otherwise, it's like they have to like somehow like encourage a token max for the next Yeah, I mean in that case they need to release Fable 5.5. Um yeah, and uh you know, whatever is beyond that, but yeah, I mean that that would obviously be a driver of more attention
52:04and and usage and that would that is a lever that they have to pull.
52:09Um but regardless, thank you for taking us through. Uh, you know, I I'm I'm really just trying to give people a good grounding on what they should know going into this. I think a lot of engineers are wondering if they should join, if they should cash out, you know, same thing for just like the general broad rest of the market because Anthropic is probably, you know, one of the uh the three IPOs that are, you know, happening that are the leading frontier labs in in this sort of one-year period. Uh and I I
52:38do think it really matters for the OpenI IPO which um if I you know anthropics fails in whatever shape shape or form and anthrop OpenI was just not going to do it. Um and that that's that's sucks for multiple reasons. Um yeah so uh pretty pivotal moment. We'll we'll see. Thank you for covering it.
52:58Yeah. Yeah. Thank you for having me. It was great to great to discuss and uh yeah love your love the stuff you put out and so uh we'll we'll continue to listen and excited to follow along.
53:08Yeah and same same for your work uh you know everyone go read his uh substack as well as uh deploy codes. We're we're actually starting a for deployed engineering track uh podcast uh and we're launching it with ode um which uh which I'm we're pretty excited about. Uh, regardless, the AI is here and we have to deploy it, right? Whether or not things are public or not, we have to deploy it.
53:32All right. Thanks, Jenny.
53:35Thanks. See you next week.