You got to have billion-dollar positions in everything in big-boy venture today.
I think billion-dollar positions are nothing. There's no point in investing otherwise with large funds. 50 million, 100 million seeds — you can't make money.
So what are we talking about today? Are the AI cracks starting to show? CoreWeave is hit pretty hard. Next up we have Anthropic, the talk of IPO town, could be priced at 1.5 trillion, not the 2 trillion originally suggested. This and so much more in the episode today. If the end-user demand is there, then everything is going to work out.
CoreWeave is B-tier compared to Nvidia and friends. Oracle is B-tier.
We're all B-tier assets. How about that?
I actually think code review is the single most important category today in software.
Coding is the mother lode of LLM value. We are just taking revenue multiples too far today. Bending Spoons is like the Berkshire version of software. Ready to go?
Boys, I just tweeted something that has got a rather large response from the founder community. I think venture is broken today in the way that I'm meeting more and more companies that fit this profile. I just met a company today, and I have a company that's 20 million in ARR, growing incredibly fast, 600, 700%, profitable, brilliant business with a great team — crickets. VCs do not want it. Is venture broken today?
Well, look, I'm sure Rory has a lot of thoughts. I'll just add one ahead of it, which is we were talking about this company, which will be nameless, right? I invested in a vaguely similar company, like doing similar things for similar size companies. They raised about a 300, $3 million seed round, and they're planning — that's it. Even with this type of growth, even with it all being agentic, even with it all being all the things we're supposed to be excited about, right? AI and voice and resolution agents — it does actually something that's pretty somewhat unique, but they're literally planning to never raise again because, you know, you get the 15%, 18% a month venture update, but no one wants to meet. It's like, well, I'll put in a little more money. But it may mean some ways in the old days where your early investors may need to carry you if you're anything that's not hyper-growth and the right boxes to check, right? The three-box check — your investors may have to carry you. It's like the old days. We need Rory to carry us through the IPO.
But Jason, you're right. We really need those reserves. Jason, you're right. You said earlier the kind of specific box has really narrowed in terms of what is interesting to investors.
Yeah. And I'm seeing that more and more with this concentration of capital going to fewer and fewer names and everything else being wholly unfundable, which more so than I've seen ever in venture before, which is surprising for a sheep-like market like venture. But what you're saying is, if you think of the role of venture financing innovation, what you're saying is there are lots of companies that look on the surface to have very good numbers that can't raise money. That's what you mean, correct?
Correct. And I think that's probably true. Let me give you the contrast, though. At the same time, I don't think it's just going to 10 or 20 companies in total, because at the same time there are a whole bunch of projects that probably in another era would not get funded at all that are getting venture money, right? I mean, you know, data centers in space, quantum computing — so we're very risk-on for stuff that has ultra-high upside. And I think the stuff that's struggling to get funded is, as you say, it's software for specific markets where there's some inarticulate fear that AI will make it all go away. Is that kind of a fair next-level summary?
Yeah, I think so. I think you're absolutely right. I think it's this like power law on steroids requirement where it's like unless you can paint me the trillion-dollar market cap, I'm not interested.
To answer your question a little more, to be more tactical, Harry, I think you're right that you roll with a high-powered crowd. I think if you want Sequoia and Benchmark and one of the better partners at Andre to fund you, the box is very specific, right? It's Cursor better, right? OpenRouter is okay, but let's see what we can do better next time. I will say for companies like you described, still growing at numbers that approach the iconic outlier metrics, right? Here's where I think actually the world is today. You got to meet a hundred VCs. You got to meet every VC. You got to meet a hundred. It is like the old days. And you will find someone that takes that bet. It will not be the seven Benchmark partners and all the rest, but someone will take that bet at a reasonable but still elevated valuation because it's the best deal they've seen this quarter. Someone will take it. You just got — it's the old days where you got to beat 100. And there are growth investors whose names are not Index and Andreessen and Bessemer, and they will take that bet because it's a good bet. To Rory's point, it will lead to a Lovable Labs or better in some cases. You just better meet everyone on planet Earth now if you just have top-decile growth numbers.
But top decile for a while was like an automatic term sheet, right? For years it was an automatic term sheet, right? You'd walk into Rory's office. Rory knew every deal since the dawn of time. You would know exactly if you're the top decile. If he believed in you and trusted you, you'd walk out with a term sheet from Scale, right? The world just doesn't work that way today. You might still walk out with a term sheet from Scale. But you're saying is there's a small number of extraordinary potential startups that the very best VCs are focused on almost to the detriment of anything else. That's your point, Harry.
Correct. No, I'm saying the whole venture landscape has realized that a certain few categories are the most interesting and exciting and everything else is uninteresting in a way that has not been before.
Rory, back me up here. If you're raising a $5 billion-plus fund, you have to have billion-dollar positions in everything. You even need a $5 billion position here or there, right? Which is why we need a $2 billion Lovable Labs — just doesn't, is not good enough because the odds, unless you have 51% ownership, which most founders are not cool with, it's pretty hard to build a billion-dollar position. I mean, if I were working at one of those mega funds and we were playing our game, that would be my final question to the team: how are we going to get at least a billion-dollar position out of this? And I'll take 1% of Anthropic. I think that still gets me to a billion-dollar position even with dilution, right? The Eric Schmidt story. But you got to have billion-dollar positions in everything in big-boy venture today, I think — and I don't mean 'boy' with a gender attribution, big, big, big folk. You got to have billion-dollar positions or nothing. There's no point in investing otherwise with large funds.
I agree. Well, I still go back to the statistics we cited last week, which is, you know, what, 10 to 20 decacorns per decade in the entire US tech industry, and we clearly believe that there's going to be 90 of them this year, right? And one of those two facts is going to be at variance with the other, and I'm willing to bet on the 50 years of data. Lifted up for growth. You know, this is not just predicting on the past because, you know, you allow some kind of growth over the years for bigger exits. But I don't think you can confine yourself to saying I'll only do deals where I can get a billion dollars to work, or a billion, or it's a — what's the outcome? A $25 billion outcome. I mean, you want some of those, but it's going to be — not the whole industry is not going to be able to say we're all going to do those and have more than five funds because it's not enough.
Maybe five is enough. Jason, I think Rory is not going along with our thread here.
Listen, I haven't sat in an Andreessen or Index partner meeting, but I'm pretty sure the math is something like how can we make at least a billion dollars off this deal. It's not even enough. Like that's — I don't know. You know, everyone has like the four different scenarios when they write up the investment memo. A billion-dollar position, not an exit — billion position is just the B outcome, right? This is 10% probability analysis: 25 billion or higher. 25, you know, this percent we only make a billion on the deal. We're raising these funds every 18 months, these 5 to 10 billion funds. We have to have so many billion-dollar positions, gains, to even be playing at this game. It's just math. It's just the maths.
To take the other side of the argument and actually to talk about seed, which is your world, Jason, you probably — I'm assuming you would agree that from a multiple perspective it's going to be far easier for the best seed investor to generate a 5x net fund than it's going to be for someone who's writing billion-dollar checks into things.
No, it's harder. It's much harder because unless you really do the Roger Ehrenberg, find something at Antarctica, right, the entry prices are so high and the ownership typically is lower unless the seed fund is very large. That actually — and all these valuations are cutting my throat. Rory, how am I going to make it up on the back end with my SPVs and my carry? I can't make any — 5x is so much easier. I'll do it at $1 billion in town and we'll raise it $5 billion next week and I'll take the rest of the year off.
I got it. I got it. The entry — 50 million, 100 million seeds. You can't make money.
And not only just to add to that, not only can you not make money, you can't make money even worse because they're a multi-stage product for those big players in terms of they're the ones who lead them and they crush your allocation. So, not only does the price go way higher, but your allocation goes way lower. But just pause. And so you have 300 to 500K at best. And I've got this in a company that is going now to 18 billion. It still doesn't move the needle, Rory.
Because what? You only got 300K in.
Because we got 300K in and it's at 100 million price.
Yeah. No, that doesn't move the needle. No, that I would agree, right? That's —
And so not only do you have a way higher price, but you have a way lower check going in.
Look, over time in a rational capital world, the return multiple from early-stage investing should be higher than growth, right? Just logically over an extended period of time because otherwise why would you do it?
I agree. But would you not say for the first time ever that we're actually going to see that inverted?
I totally agree. I think the last five years are going to be clearly inverted. It's clearly going to be — I mean, look, someone said it. It was a very good line. He said the aggregate market cap of SpaceX, Anthropic and OpenAI went from around $400 billion three years ago to something like five trillion, four trillion today. Right? That means trillion. Add change billion into a trillion. 400 billion to 5 trillion. Right? So you've got four trillion plus of value created in the latest-stage business. Nothing else matters, right? And if your theory is there's going to be 10 more than next year, then you're right, that's all you do. And if your theory is that was amazing, but it's probably not going to be a run-rate business, then you make different bets.
We will go back to schedule, which is: are we starting to see a crack in the AI trade? CoreWeave have had a tough week. I admit I'm a holder. It has not been great. Rory, can you please provide some context as to what is going on and what do we need to know as background?
Yeah, look, I mean, I think Core had a tough week last week. I think down 15, 20% kind of — bond prices went up implying, you know, lower default. But zooming out, the way I think of it is, when you think of the whole AI trade, these kind of neoclouds, they're always going to be more risky than anything else in the system because they have, you know, they're underwriting high growth with large capex and lots of debt, right? So you try to thread the needle. If you grow too slowly, you don't meet your customers' happy. If you grow too quickly and demand slows even slightly, then you're screwed. Because I'm doing it from memory here. I think, you know, I checked it this morning. I'm sorry, but CoreWeave, I think it's like $7 billion run rate or whatever, 10 or 20 billion of capex, you know, 35, 40 billion of debt, right? I mean, that's the way you have to do it because you got to build these data centers and fill them full of compute. So it's just very tight, right? It's not like a software company where you can hunker down and cover your nut, right? There's just no way out here. So they're always going to be risky. So anytime someone wakes up and says, 'Oh, AI is growing a little less quickly,' all these things are going to go down. And anytime anyone says AI is growing amazing, they're all going to go up. And last week was an AI go-down week, probably a little bit because of the Anthropic numbers, which in retrospect a bit silly, but there's no doubt last week was that things are slowing down week. And when that happens, Core jumps down.
Rory, was this a gross overreaction? And if the end-user demand is there, then everything is going to work out, and including CoreWeave, because look, I think their biggest customers are people like Microsoft, OpenAI. Like, provided the demand is there, they want the data centers, so it will all true up under those circumstances. Is that not like a generational buying opportunity though? It's like down 25%. Of course demand is not cracking. We all see that every single day more precisely. But the problem —
Yeah. Well, if it was that easy, we'd all be doing it, Harry. But I mean, the real question is the growth — not is the demand going, but how does the demand compare relative to what these guys are spending, right? You know, it's like can they fill everything they have now? Of course they can fairly quickly. But if you're continually investing for next, like I said, if your capex budget is 20 billion, it's not enough that demand go up. It's that demand go up by a lot every year.
The beating heart of the whole thing, and we're seeing people share it, is you add up OpenAI and Anthropic's numbers together. So that kind of gets rid of the shift in market share, and the growth rate of that overall is what's going on in the AI economy, right? And that's the growth rate you have to look at. You know, the question is how does CoreWeave's capex plans look at relative to that.
The simple learning is the markets are going to be very sensitive, right? As Rory said, any bump with Anthropic, any bump with OpenAI, any bump anywhere, of course CoreWeave and Oracle should be hyper-sensitive because they have massive debt loads and massive risk. But listen, but they're also the B-tier players. CoreWeave is B-tier compared to Nvidia and friends. Oracle is B-tier, right? So I don't know. You can make money in the B-tier, but why would you want to hold the B-tier? CoreWeave exists because Nvidia and friends need it to exist for now. It has no inherent reason to exist. It is not producing a singular set of silicon and GPU and software today that no one else can do. Oracle just can't sell enough enterprise software. People don't want to install enough of this crap on their servers. So they're stuck renting out their excess capacity and buying GPUs and doing Stargate. These are B-tier assets, and B-tier assets should trade at a discount. And I think more importantly, they should be more sensitive to shocks in the system. The B-tier assets, the A-tier assets should be in theory less sensitive.
Yeah. Fine. The only comment I'd make is I think, you know, B-tier, A-tier, CoreWeave versus Nvidia, I understand what you're saying, but it's almost like you're comparing apples and oranges. I mean, Nvidia's core IP. Yeah, it's the dominant provider of the whole stack. Look, compared to Nvidia, we're all B-tier assets. How about that?
Yeah, but why buy the — Harry just asked, listen, why buy the B-tier? It's like IPOs. Why buy — like Braze is getting acquired for whatever, 5.8 billion. It's down 40% from its IPO. Like why buy any IPOs that aren't A-tier? Why buy any — you just, there's just nothing in the public. You can make money as a — if you get your nice seed entry at 8 million post, you can make money on the B-tier, right? But you know, it's the start of this conversation. It's a tier bust. There's just the discount's too high. What the hell am I going to do with my CoreWeaves?
It's five or six. I'm not going long on — I want to hold, I want — I'm not passing Core to my great-grandchildren. I'm not putting it in their 529s that I'm giving to my great-grandchildren. You know, there's no way.
I mean, this is the Jason stock-picking is dead. This is the public equivalent of the private comment: only the biggest companies matter. And again, the great thing about this is I can prove you wrong more easily in the public markets because, you know, when you go back over a year, I mean, you know, Nvidia is up 18%, and Nebius is up 50%. You know, socks — I mean, that's — Micron's up 350%. I mean, the truth is you buy the second-tier assets because you think they're going to pop, right? So, you know, I hear you, but no, you know, the bigger you are, the more likely you are to —
Let's see if you put 20% of your personal portfolio into those, into the Micron. There's a lot of risk there, right?
So, Rory, you know, we have Jason's IC. We're going to have Rory recommends. Okay.
No, we're not going to do it. Genuinely, we're not going to do it because I'm not going to sit here — one thing I've learned about doing this now is pulling stuff out of your ass because you've done a little bit of work is just not right.
That's not — yeah, so no.
Okay, okay, I'll leave it there. I'll leave it there, it's okay. And we can talk about what maybe caused a slight crack in that, which is Anthropic's numbers. Anthropic announced that they turn profitable on an adjusted basis. Growth slows. IPO price potentially rumored to be going down from 2 trillion to 1.5 trillion. Q3 revenue was 16.8 billion, up 45% on Q2, 85% gross margin, which we'll talk separately about. What should we take from these numbers?
I think it was good. I mean, I think that like it depends. Look, I guess —
You should tell your voice that, Rory.
Sorry, long day. I mean, look, growth is clearly slowing fairly quickly, but from an unsustainably high number. I mean, it had 10x'd for the last three years. It's highly unlikely to 10x. You know, what was it, run rate coming into the year, 10 billion? If two more 10x's and what was said, 100 billion, you'd be at a trillion. So growth was going to — this is all about how fast is this going to grow, how quickly is the growth going to slow, right? And honestly, I thought — I'm sure people are sitting there going, if you add linearly, if you did five in Q1, 11 in Q2, 16 in Q3, you kind of lean in and you're at 20, 21 in Q4 if you keep adding 5 billion bucks a quarter. I'm not saying they will, but right, and you do that. So you start next year, which is when you go public, the future year, if you're at a $20 billion run rate adding five a quarter, then you do 100 and, you know, 105, 110, and now you're saying it's 12, 13 times forward revenues for the most exciting company in the world. It doesn't feel crazy. So, you know, it doesn't take a lot of squinting to make that work.
You know, I'd say I'd add just one thing for what it was, just tactically. I thought — I think, Rory, you'd agree. I mean, Chamath's tweet was just raise whatever you need at 1.5 trillion and move on, right? Win the day. Like, if you can raise at two trillion, but it's going to stress your capital needs. Like, it's just a valuation. So in absolute terms, everything's great, right? This is unprecedented on all levels. Great team, great application. I think though what I suspect will be a reminder is it's going to be very choppy after the IPO because going to Dev's point when he was on the show, developers are not loyal, and the amount of market share — since this show started, right, the amount of change in Anthropic versus OpenAI versus open weights versus Qwen, right, which didn't even exist what, 45 days ago, right? This is a winner-take-all, but the amount of variability in the incremental growth and in the market share — there's only so much predictability in this revenue. There's only so much R in the ARR here. And I think depending how effusive the markets are, it may create stress. It wouldn't surprise me if at the end of next year, Anthropic's market share is down, you know, 10% from where it is today. And it still could be great on an absolute basis, but down percent wouldn't be a shocker at all. Not just in tokens, but in revenue. Easily could happen. Look at — despite whether OpenAI at 70 or 50, it's a lot.
When you say the, you mean the share is down.
Yeah. Market share could be down, right? And so when you're building your models, right, not just DCF models, but your top-line models, you could be materially off from your good case, right? Your best case. And it should be expected. It should be expected, which to state the banal is a little scarier than the average recurring revenue business, right? And that's what people are wrestling with.
How do I think about — I mean, if you have continued market share shifts between the key players, then the next shoe to drop is typically, well, I want to keep my market share, so I'll look at price. And I think, you know, everyone's trying to work out from a game theory perspective, you know what, you're right, what does that look like as a P&L going forward, right? Which is why going back to the boring thing I keep saying is that provided the end-user demand at the top is there, you can, you know, everybody can win, and we all love businesses where everybody can win, right? If that starts to slow down and then price competition becomes, you know, literally the thing you got to do to take it from the other guy to keep your growth rate going, then it could get pretty tougher. So that's —
You think Sam was super strategic raising his supposed round at 1.4 just in terms of some form of anchoring and message to market?
I think when you need money you take money, and I don't know if there's a big sophisticated, oh, 1.4 is a messaging versus 1.5. I think it's as simple as, you know, we spent a lot of money in compute, we didn't get our IPO done, been good, we spent all of Masa's money, better get another 40 at a price. I mean, you know, what are the constraints? It's got to be up on the last round otherwise the messaging is, and it's got to be down below reality to leave room for the upside. You know, there's only a finite number of digits between — well, there's an infinite number of digits, but there's only a finite number of one-decimal-plus digits between one and two.
There are a couple of elements beneath this story, one of which — or beneath Anthropic — is a weak one: was Replit made OpenAI its default? Jason, I am intrigued here. Anything from you on that, especially given your Replit power-user status?
I think it was a stupid media story. I mean, we do use Replit about 14 to 15 hours a day basically. Replit has its own router in it, right? Its own model selector. It can be invisible or visible to you depending on what category you're in. And OpenAI models are just cheaper for the same rough quality output in Replit. So they just made it the default, and TechCrunch miswrote an article saying they were no longer working with OpenAI, which wasn't true. They just flipped the default from Opus and Fable to Astra. That's just the default, and it's what we're all routing, right? We're all open-routing one way or the other. It's just if you're Replit or Lovable or Base44 or V0, you have these extra incentives because your average user just wants to have a high-quality output, and they're very incented to get this quality-price thing just right for you, right? Unlike, you know, the average idiot configuring OpenRouter themselves, hoping they get it right. These guys are — it's more interesting to watch what Replit, Lovable, Base44, V0 do because they have an incentive to nail the routing, right? They have a very strong incentive. They can only partially pass it on to the end user, right? So we can actually squint and see the future based on what these guys are doing. And what they're saying is — listen, what John's saying is the quality is sufficiently peer-level here and the price is 30 to 40% better for us. What today? Why would we use Fable and Opus? And you know what? Next week he may be at a different event and get on stage and just go into this market share thing with Anthropic. It may change, right? It may change. They may not have fully thought through how to use Sonnet 5.5 and Haiku 5.5, and they have Kimi and others. But if Kimi actually was so great, you know what? All those guys, Replit would be using it entirely, which is maybe the more interesting learning that was under-discussed because it will route through them, too. And it doesn't. Amjad didn't get up on stage and say we're routing everyone through Kimi, did he? He said we're going from Claude to OpenAI because it's just as good and materially cheaper. What's the take? That's all he cares about.
What's the takeaway from that? That Kimi is not as good as people think? That —
