0:00Okay? Because the reality is is open and profit, they have far more powerful models that they've developed that they just haven't released yet. If you compare the dollar spend in compute a year ago versus today, their compute that went towards serving models has gone up by 2.3x. Their revenue over the same period of time has gone up 14x. So embedded in that is basically a 7x [music] efficiency gain in serving inference which I mean this is a
0:28business that I think is almost accidentally profitable. Right? Maybe my my favorite stat of the whole S1 was $141 billion of capital raised to date. That's inclusive of equity converts and safes and a $134 billion cash balance today. right now. You want to go raise multi-billion dollar debt for GPUs? No bid. No bid. They're like debt markets are closed. They're, you know, big debt deals that aren't happening.
1:06Okay. We're here in the studio with Jam Ball from Ultimator. Welcome.
1:09Thanks for having me. Excited to be here.
1:11Uh, Anthropic is IPOing. I was thinking of who do I know would be the perfect guest to cover the S1. You're the guy.
1:18You're the guy. Uh, we met before I think you started your Substack. I maybe was just starting mine. How old is it?
1:25Uh, six and a half years now.
1:27Okay. Actually, no, you're older. Uh, so Cloud of Judgment is like one of the top like what what are the brag statistics about cloud judgments?
1:34Oo. Uh, brag statistics. Um, probably getting close to 100,000 subscribers, you know, every week. I I think in six and a half years I've I've missed one Friday. uh try to be pretty consistent and the the the general like my takeaway is like here's like public markets comps for the startup investing that you do but ultimately yeah the the idea was hey how can I provide some commentary on metrics around software companies and you know the universe of companies that that
2:02typically has public metrics or public companies um as a as a way to kind of not guide but uh provide a blueprint to early stage companies and founders of like hey here are some of the metrics required to be public one day Um, and so there's lots of tables and charts and graphs on public metrics and valuation multiples. And maybe three, four years ago, I I kind of started writing a little ramble up front, which is more of a I kind of treat it as a personal journal, like what's on my mind, what what have I been thinking about, what's what what's top of mind. Um, and you
2:32know, that'll range from anything from, you know, some macro commentary to market commentary to, you know, trends in AI or what, you know, whatever's trending at the moment. Yeah, I think every generation there's kind of a VC blogger who's like sort of the guy to cover what's going on. It used to be Fred Wilson if back in the day.
2:53Uh he used to write ABC I think and then uh more recent like Tomas Tongas. I mean he's still active but like uh I feel like you are the king of like charts and public stocks right now.
3:03No, I appreciate I mean Tomas was a huge inspiration. So I was at Redpoint for for quite some time with Tomas and you know and now now he's started theory.
3:11Exactly. And you guys Red Point. Okay, you guys are all leaving.
3:14No, Redpoint's great. They're they're great. I mean, I owe everything I learned to to those guys and I I have a ton of respect for them.
3:20Yeah. Yeah. Amazing. Uh okay, so we're going to get right into Anthropic and uh you have a a sort of some spiel. You've had some public dustups with people that we won't name.
3:31Um people are just generally very confused because they a lot of them have never a lot of our audience are engineers and researchers.
3:37Uh they've never seen an S1.
3:38They they've never been through an IPO process. is they don't know what a road show is. They don't know how this is done. But more importantly than that, just the that's the mechanics. More importantly than that, what does it mean about our general state of AI, right?
3:50Like this is the most visibility we've ever gotten into anthropic.
3:54And like well now we know a lot more things than we used to.
3:57Yeah. Totally. Totally. And just just to say it up front um because I've been asked to say this. So these are I work for Altimter. We have a big position in Anthropic. Everything I say is you know these are all my personal views versus versus uh versus my firm views. Uh but yeah, it's a big deal. I think you know the big labs, OpenAI and Enthropic seems like every quarter their numbers leak.
4:16Uh but it's a subset of the numbers. It's not it's not everything. And now we kind of get a full open kimono view of the financial profile of you know one of the most important and iconic companies maybe of our lifetime. And just kind of a a quick process uh process refresher.
4:31They'll file an S1 uh probably wait about 15 days. There's usually a minimum until they'll start a road show. the road shows and they actually start and they pitch investors, IPO investors that, you know, travel across the country, across the world to pitch the IPO. That'll usually last a week and a half and then it will start trading publicly. So, we're probably looking at end November.
4:51Yeah, like mid end of November. Um, you know, kind of 15th to 20th, 10th to 20th of November of of when it actually starts trading publicly.
5:00Yeah. Do people IPO past Thanksgiving?
5:03That's like a weird dead zone. It's, you know, between Thanksgiving and Christmas is is kind of a dead zone. Um, I would say, you know, post Labor Day to Thanksgiving is generally a kind of like a hotbed uh time for companies to go public. People are around, people are engaged. Um, notation and it's anthrop is I think people would people would make time for work for Yeah. Exactly. Exactly.
5:27Uh, yeah. So, okay, that that's the normal mechanics. I I think um and and that's it, right? So, so off the top of your head, uh, you know, people should understand that an S1 is just a filing that you have to do. Um, opening has done theirs as well confidentially and then in the legal public.
5:43Um, people have filed S1's and not IPOed. Cerebras famously did that.
5:48And then they did it again. Um, and uh, it's just part of the routine process and you have investor day and you have whisper rumors and all these things, right? So, uh, as someone who is unless you're literally a banker or fund manager, that's all you need to know basically, right? there's not there's no this is part of the process. Um and there's a lot of indication to try to make it a successful IPO. Successful is a mixed or loaded word this these days because um typically you you hear things like IPO is so successful the stock popped 50% on
6:16the first day. That's actually potentially a failure if you're um Bill Gurley.
6:21Yeah. Exactly. Exactly. [laughter] Who was like well you could have got 50% more.
6:25Exactly. And you know, and just just for the audience, what what that mechanism is is in an IPO, it's like a fund raise on steroids. You're going to go out, you're going to raise a bunch of money at a certain price.
6:34How much are they raising?
6:35I don't think I don't think that's been okay.
6:37I don't think they've decided. I think the rumors are like 75 to 100 billion.
6:41Um, it'll be a big raise. I mean, typically, what you used to see price I I know what you used to see in normal IPOs was companies would float about 10 to 15% of their company, which means like what you would sell into the IPO would be 10 to 15%. Yes. And there's an employee lock up and ex Exactly. Uh now it's it's maybe as as the companies get bigger and bigger and bigger, they might float less of the company because hey, you can take less dilution. You know, it's no different than a company raising a series A, B, or C. It's you're going to raise money at a
7:10certain dilution. Um and yeah, the Bill gurly thing is well, if the stock pops 50% on day one, why wouldn't you have just raised money at a higher price, right? And taken on taken on less dilution. And that's a whole separate topic. Yeah.
7:24lots of nuance to public markets and IPO out there's tradition there's conflicting incentives and then there's there's the chance that you actually you don't pop yes um so all that let's put it at a high level why now why IPO now right they uh they could do a Stripe they could do a data bricks right they're not they're IPOing why yeah I I mean I think there's a couple reasons well one it's all speculation it's hard to know but I think there's a number of reasons why companies have historically chosen to go public um there's kind of the obvious let's
7:52provide liquidity to our employees and our investors in like an unrestricted way. Um because of course, you know, now big companies, data brick, stripe, open eye, anthropic kind of like tenders are are pretty regular. Uh once a year you you'll get like well if you want to cash out here's here.
8:08Yeah. But you know you'll have you'll be limited in how much you can sell. Um you know now once the company's gone public once they've gone through the lockup you know which is anywhere from you know it used to be just 90 180 days 6 months and now it's kind of like every company has a bespoke lockup that's like dribbled out some are shorter some are longer but it's kind of like unrestricted access and you can generate liquidity for your employees um help morale whatever or um you need a liquid currency for acquisitions and hey a company might be
8:38more willing to take an acquisition from a public company because the currency you're getting back is is liquid versus in a private company. Maybe it isn't. You don't really know the value. Yeah, it's monopoly money. Um I would say these quasi public companies, you know, OpenAI, Anthropic, Data Bricks, I think their currency has almost started to be viewed as cash, not monopoly money, because they're very high quality.
9:00They're very late stage. Um that effect probably means less. I would say a a bigger reason to go public um because and it used to be companies went public because now you could access much larger quantum of capital. You know, you if you wanted to raise $100 million, you know, it used to be the only way to raise $100 million is to to go public. You know, now you you seize and raise a series A or a seed a seed of of more than that.
9:25Um and so it's not necessarily the quantum of capital. I think whether it's US institutions getting bigger um maybe folks who used to act as LPs, large pension funds and endowments acting as direct investors in a company themselves or kind of the emergence of big sovereign international funds investing direct [clears throat] in companies.
9:44There's a huge capital base available to private companies and so like I don't know if the argument necessarily holds hey you if you want to raise a lot of money um go public. I think Openai raised a round of like 120 130 billion you know some biggest of all time. Yeah. Exactly. Um I think maybe a more important angle is is more of like the debt piece. Um and so once you go talk about it once you go public um actually let me take a step back. If you
10:11want to raise debt, um, or if you want someone else to raise debt, not on your behalf, but like as a Neocloud, a Neoclad might want to raise debt to build a data center for Enthropic to to use. Debt providers are going to look at Enthropic and they're going to underwrite Anthropic's creditworthiness.
10:28And how it works is they'll go they'll say, "Hey, we're going to syndicate however many billions of dollars of debt for you anthropic or via a neocloud for you, but we're going to need we're going to need like full open kimono on your financials. We're going to need to know exactly how this business is doing uh so that we can underwrite you. And then everyone we syndicate the debt to on the back end, they also need to see it." And I'd say historically companies have just been hesitant to go full open kimono with their financials to the debt and credit markets because they don't want
10:57them to be fully public. Um, but if you want to be rated by the by the rating agencies who basically they'll give you a credit rating, um, investment grade is the highest, your credit rating will kind of impact the rate with which you can raise debt at. And the higher your credit rating, the lower the rate that you'll receive. But you, you know, have to be open kimono with your financials.
11:19So going public um is a great way for companies to basically you're going to get raided. I think SpaceX very famously got raided very quickly after going public. Now you can tap into Yeah.
11:30arguably too I mean there there was a lot of people who said it was hey it was too fast they need to season more as a public company. Um and and added to the S&P 500 something like that new rules. Exactly. And so I I think hey a lot of these companies they will be capital intensive whether it's for training or building out compute clusters and you want to tap into the credit markets and I think if you want to tap into the credit and debt markets you can do that much more efficiently as a public company and and so again it's probably a confluence of all those
12:00things added up to hey why why go public. I think there's also just if you ask if you ask these executive teams and management teams, hey, they want the everyday American to be able to participate in the equity upside of these companies.
12:12I was going to bring that up. Yeah. Up to including giving shares to everyone.
12:15Totally. Totally. And hey, the reality is is private markets are more of like a cabal. It's a small subset of the population um either qualified purchasers or accredited investors or or you know, big funds and pensions and endowments get to participate in the equity appreciation of these businesses.
12:31But you know your doctors and dentists and your everyday American doesn't and there is just something that doesn't doesn't feel right about that right like straight up you'll get social unrest right and straight up you'll get you get voted regulation because like well those guys have their AI and Amazon vote against it I have my right that's right that's right and so like that is actually really an interesting governing mechanism of democracy versus capitalism where like capitalism encourages centralization and democracy encourages um decentralization
13:00somewhat Um anyway, so there there's all that I think it's very very valid. Uh I'll point out for people who don't know about capital structure that uh equity is is junior to debt and so it's nice to have two trillion in junior Yeah. Exactly. Exactly.
13:13in your enterprise value. Right.
13:14Exactly. Uh and that's like well that's that's uh that's public. Um and I think part of that also I'll I'll throw in the why now which is a little bit more advanced than than people know is um interest rates are spiking for for um uh for data center debts. uh globally, right? Like is there a source to look that kind of stuff up? Is it just Bloomberg or is Yeah, I I I mean I just charts that people tweet around, but most of these big debt deals will price
13:42um and there'll be, you know, like an interest rate, an o, you know, there'll be a bunch of things associated like with the price and like often times those will be public once they've priced. Um but yeah, rates are going up.
13:55Um risk-free rates are going up. the Fed paused you decided not to hike yesterday or earlier this week. Um I don't know if that's necessarily a reason to go public. Um but it it's part of the why now if if it does help with debt that if my debt went from like I don't know uh 3% to 5% that is billions of dollars.
14:17Um and if I can keep a lid on that that's very important to me.
14:21Yeah. I mean a lot of the data center debt deals now are are getting done over 10%. Um and so it's it's getting it's getting expensive.
14:29Yeah. And so so and then the other that's the other part of this discussion which is um uh the ROI on a data center, right? Because that there there are a lot of AI skeptics who are like this is a the biggest buildout in human history and we are throwing money after this illusory dream of AI. Then on the other side the AI pro people are saying well your payback period is one year.
14:52Yeah. Yeah. Why is there such a huge difference?
14:56This was kind of the best thing about the anthropic S1, which is a lot of the metrics that were leaking were your your ARR, you're this or you're that, but you didn't really get to see a lot of below the line. Um, you didn't get to see the cash, how much are they burning? What's are they profitable? Are they not? And like that I think was one of the biggest surprises of the S1, right? Here's a business that showed 85% gross margins, over 20% free cash flow margins, um 8% gap operating margins, and so even on a gap basis, um highly profitable. And
15:25maybe my my favorite stat of the whole S1 was $141 billion of capital raised to date.
15:33That's inclusive of equity converts and safes and a $134 billion cash balance today. And so cash and cash equivalents, but yes, exactly. And so you think, okay, well, you've you've raised this much. You have this much in the bank. That means you've burned $7 billion to get to, you know, the Q3 numbers that aren't in the S1, but that were kind of like broadly reported a few weeks ago. $16.8 8
16:00billion of quarterly revenue off of $7 billion of lifetime burn, which and the burn we should we should get into cuz like that was a a huge debate on like well have well it looks like they did raise debt or are they really profitable? What is this net loss on the P&L? There's I'd say lots of lots of things out in the world that aren't fully true.
16:20That is important. I do want to cover it. that's that there's a little bit more in the weeds than than than at the start of this conversation. But like that overall efficiency is by far uh if you've been following the relative fates of let's call it SpaceX uh anthropic openi the the ones that are going public um they are the most efficient lab in the world by by like order of magnitude.
16:43Yeah totally. Um, and you know, so I I when when I wrote up the S1, I was like, "Okay, well, this is the the the the mystery, like why what's going on?"
16:52Like, how are they so much more efficient?
16:55And uh I they in their S1, they attributed to the loops um the improvement loop, the investment loop, something like that.
17:03Um so people can go read that up. I I already wrote it up. Um but like, you know, maybe they're just smarter.
17:08Yeah. [laughter] Well, I I think here like one of the best numbers to look at is like what is their growth relative to their what is the the change in growth relative to the change in serving compute spend and they do they break out their compute spend into more like training and um internal use and by the way you have charts to yeah let's play we do have I do have a chart on that um so here is here is a chart that shows the let's do full screen on this one yeah yeah
17:37just just so yeah it's Perfect. It's a little small.
17:39Yeah, exactly. All right. Here's a chart and this is broken down in their S1 that shows the percent of their compute that goes towards basically serving models. So, inference revenue generating or internal R&D kind of training or maybe safety and alignment or you know non serving of uh of tokens. And what's interesting is you can see now I mean they're at 27% only 27% of their compute today is going towards serving models.
18:06Um here's why this is so interesting. If you compare the dollar spend in compute a year ago versus today, their compute that went towards serving models has gone up by 2.3x. Their revenue over the same period of time has gone up 14x.
18:22So embedded in that is basically a 7x efficiency gain in serving inference which I mean this is a business that I think is almost accidentally profitable, right? where I mean a 7x increase in efficiency of serving tokens. If you were if you were monetizing, you know, a megawatt at 10 $15 million, now you're monetizing it at $70 to $100 million.
18:47And here's this is a little bit of like a a random rant. Um, but software companies have historically told the world and told investors, "Hey, we're an annuity of cash flows. Once we acquire a customer, our net retention is high and they they'll stick with customer lifetime value."
19:01Exa Exactly. and we're going to be burning a lot of money early on. Um, but we're just in growth mode. We're in customer acquisition mode. Once we exit growth mode, we'll be more in harvest mode. And that's when we'll start to become we'll flip profitable um because we're not spending a lot of money on on um on customer acquisition costs. But, you know, the the general model of a SAS company was unprofitable over time then profitable. What's happened with Anthropic? And let me pull up another
19:28chart here. Um, I believe I have their ah here are their free cash flow margins.
19:33Oh my god, it's so negative.
19:35Which [laughter] I had to cut the chart, right? Like like negative 4,000% negative500%, you know, super super negative. If if I actually drew the axis to scale, it it would it wouldn't look real. Um, and so yes, the the axis is broken. Um, but this is the free cash flow margin of anthropic, right? 21% free cash flow margin in the most recent quarter. And so what you're seeing is a business that all of a sudden because they were able to drive such efficiency in their compute, that's kind why I went on that
20:03like compute ramble. Um they've been able to monetize at much higher rates and actually deliver positive free cash flow margins pretty early on. And so this is a business that says, "Hey, we've proven that we can operate highly profitably. What we've probably done is underinvested in compute in compute buildout and we need to be more aggressive on it." And so my guess is that you're actually going to see this level off and maybe go negative again as they say, "Hey, we're going to reinvest.
20:27Um, and we're going to invest in a lot of compute. We're going to build out a lot of compute. Um, because now they've shown themselves and the worldap.
20:35We can we can flip the switch. Um, and and be profitable." Um, and for listeners, uh, a couple things on the accounting side, just because I we got it like educate people, right?
20:46Um, FCF is typically like the most honest version of what profit looks like. Um but uh in this case you can mostly just regard it as what an operating margin would profit margin would look like because basically FCF just focuses on cash and so there's less tricks to play with acred receivables and li and you know offsetting of liabilities and what have you. Um and so like that that is the roughly rough thing and but also this the re the reason people have adjusted revenue and
21:13adjusted profit is because of nuances in the running the business like this right like GAP which is the official generally accounted accepted accounting principles that you have to report um doesn't recard how you want to run the business and split out investment uh it does have investment decisions versus like sort of profitability of of core business decision decisions but um it all gets smushed into this decisions right it's like so basically what what I'm trying to sort of guide people towards if you
21:43do see a change in this margin profile it is a conscious choice and actually you should probably split it out uh further uh between what is actually the core business and you know is that actually inflecting or is this is this more investment therefore SCF went down yeah and look I mean at the end of the day cash is king um and like your measure of profitability should be is your cash balance going up or is your cash balance going down. Now, if you wanted to poke holes at this, um the the reality of gap accounting is there's two
22:12forms of well, very broadly, there's two forms of of expense. There's the cash and the salary that you pay people and then there's also the equity and then the non-cash that you pay people. So, non-cash uh stockbased compensation, equity compensation that doesn't show up in free cash flow. It's actually added back uh because it's a non-cash expense.
22:30So, there are a lot of people who look at the free cash flow metric and say, "Well, it actually overstates your profitability because the reality is is if you didn't grant equity to your hires, they wouldn't join you." And so, like, it is a real expense. Treat it as a real expense. I don't have a chart for this. Um, but if you were to adjust free cash flow for stockbased comp, you're still at an 8% free cash flow excluding stockbased comp. And so, you're you're even profitable if you include the big by normal rules.
22:57And so, like if you've never seen a software IPO, this never happens. Yeah, it's very rare. [laughter] I mean, I think there there are a lot of software companies that are free cash flow positive today, but if you factor in not at IPOP um but even today, the mature ones, even if you factor in stockbased comp, most of them are free cash flow neutral or negative if you treat stockbased comp as a cash expense.
23:19So, all those companies are dying and everyone's just in buying in topic, right? As a I think there's Well, you're not quite that. Um but but certainly, why would you like anything else? Yeah. I mean, look, here's a crazy stat. Um, in the S1 is like a Brett versus depth one. Um, they called out 50% of the Fortune 500 is using Anthropic.
23:39However, if you look at what that cohort, the Fortune 500 is spending on Anthropic as a percent of their expenses, it's.1%.
23:48And so, you ask yourself, where's that headed? I think John Collison on he had Sam Alman on his podcast the other day. I think he mentioned 1% of Stripe's expenses are on tokens today, but he expects that to go to 15 to 20% over the next few years. And so obviously Stripe is a very forward-leaning tech company.
24:07Your average business in the Fortune 500 isn't going to show that level. Um, but it could show a same order of or same magnitude of growth. So, if Stripe is saying, "We're going to 15 to 20x the spend on tokens relative to other expenses in the org," maybe you see something similar and the Fortune 500 go from 0.1% to 1%. Um, but I do think it's, you know, ultimately where do you think that settles out? How deep can they penetrate their TAM? It's kind of
24:35like an important question that, you know, folks should think about. Uh, but to me at least, what it demonstrates is they really haven't, despite, I mean, despite the insane ramp in revenue, you know, growing over a,000% year-over-year in the Q2 quarter, there's probably still a lot of room left um for them to kind of continue to penetrate their TAM.
24:57Yeah. Uh, and, you know, I think the the word is virtual collaborators or, you know, I say agents, but you got to you got to make it a friendly term for people. Um and uh you know I really start to tell people uh you should just start thinking about it in terms of like what your employee spend versus your token spend is and most people are at let's call it 0.1 or 0 point or like 1 to 5%.
25:18Y uh the median AI companies at 50%.
25:22And the top AI companies uh you know let's say 90th percentic person is like 10x uh token token to salaries. Yeah.
25:30And so, uh, that is probably the right comp of like, well, what are you spending on humans?
25:35And like probably 30%. Right.
25:37Like, and well, okay. So, could be more even. I mean, your humans could be like four, you know, total headcount could even be depends what kind of business you're running, right? Yeah. Exactly.
25:44And so, like if you're physical business, fine. That's like until we have good robots, then that's not that's not getting happening. But what are you spending on knowledge worker humans?
25:51what and knowledge worker humans that includes coders but also just um admin and ops and what have you and and legal and whatever and all that and some part like that percentage the to ratio of tokens spend to human spend was zero to whatever and now it's going to rise and how far can it go totally [laughter] totally can go a lot I mean look I I think there's there's a lot of like hyperbolic statements being made both positive and negative about
26:21OpenAI, Anthropic, the big labs. Um, I think it really boils down to three main questions. Can Anthropic stay on the frontier? Is there a business and a durable business to be built at the frontier? And if so, how much is it going to cost? And and then in reality, two and three kind of like relate to each other. Um, can it make enough money to cover all the cost to be at the frontier? And that's that's the one where I think I actually have a couple I think I have one on this. Let me Let me
26:50see which one. What did I want to bring up? This one. Yeah, let me pull up. Um, no, let me pull up this one. This is a good one. Um, so this is kind of the second question like how much will it cost to be at the frontier or or or even maybe let's go to like what what can you build a business at the frontier? The other interesting stat that Enthropic had in their S1 was for the 3 months
27:17ending September, 80% of their revenue was driven by models that were 6 months old or younger. And so what does that tell you? The vast majority of their revenue is coming from their most recent, most powerful models. And that's something that I kind of expect to continue. I think you know when it comes to frontier tokens down to open weight tokens I think there'll be a massive market for both. Uh I think the trend will continue to be on a as a percent of
27:47tokens open weight kind of continues to eat share but as a percent of wallet share in dollars Frontier continues to take the line share. So I expect Frontier tokens to be the most economically valuable to capture the most market share. I expect openw weight and cheap tokens to capture the most market share in terms of total tokens.
28:06And I think kind of this stat on 80% which is much bigger than I would have expected just in models six-month old or less kind of demonstrates the value of being on the frontier. Okay, so then let's go next go to the second question which is well what is it going to or the third like what is it going to cost? Um and here's a chart I think and there there's I'll have to caveat this chart.
28:27Um, but a lot of people look at the Anthropic S1 and say, "Oh my gosh, there's a $856 billion of off the books of like not even off, but like committed deals that they've signed. How are they going to pay for it?" You know, they're at whatever whatever their ARR is, something in like I know 70 75 80, you know, whatever. I think the last number was 65. Um, the Q3 number was 16.8, but whatever 70 plus or minus um how are you going to pay for 855 billion? you know,
28:55like have you read the rest of the S19?
28:57Well, what I what I so what I tried to do what I tried to do in this chart is I broke down the the because they broke down their commitments. Um, they didn't do it by year, but this is normal, by the way, that you would have year by year.
29:09So, I feel like this is unnecessary.
29:11Year yearbyear. This is a little bit of the jam in um best interpretation. They'll give a dollar.
29:17Oh, no. I've seen variations of this.
29:19Yeah. Yeah. Yeah. It's and so I think what you have to ask yourself is in the commitments that they've made and I've tried to annualize the commitments if they said this is a 5-year deal. just made it, you know, by the way, every banker in the world has done this already. Like they did this on on the night of the the S1, right?
29:34So, you have to ask yourselves, can you pay for it? Okay. Well, at a fixed basis and then the the gray bars is maybe kind of like less hard commits, but commits they want to make.
29:44So, the the X SpaceX one is in there, right?
29:47That's the one they can walk away from 1990.
29:49Exactly. Um, will they be able to fund 100 billion a year? And so, it isn't 850 billion a year. It's can they fund a hundred billion a year? Now there's a couple ways they could do that. Well, one, do you expect their growth to continue? Um, which is a question of like will there continue to be a business at the frontier? Um, or you go back to the well only 25% of their compute today is allocated towards um
30:17revenue generating or kind of inference serving use cases. if they were to just double that and add zero compute, they would they could in theory double their revenue. And so like you look at the headline number of 855 billion, you're like, "Oh my gosh, that's a huge number.
30:32How are you going to do it?" Some of that is very backweighted. Um, but I I I I think when I think about the trajectory of the business, um, the TAM that's been penetrated today, I don't think it's far-fetched at all to say, hey, like the business trajectory will more than cover the revenue will more than cover the um, the expense. And I think I have another chart in here. This is another great one. This is the compute margin. So this looks at if you were to take their revenue and subtract
31:01all compute re uh uh internal models training same quarter matched right exactly exactly so even as what you're seeing is even as they've shifted more of their compute to nonrevenue generating use cases they've still scaled their compute margin to well over 50%. Mhm.
31:21Which what you're So what you're seeing is they're getting hyperefficient at the compute they bring online. The compute they decide to use for revenue generating use cases is becoming it's become 7x more efficient over the last year. I think you have to ask yourself will they continue to ramp compute and capacity? I think the answer is yes. I think if you ask anyone anthropic again this is all speculation they'd probably say we wish we started doing it sooner.
31:46We wish we were more.
31:47They're trying to commit to more compute. That's something if you don't talk to any of these guys, you come away with like, oh my god, they're, you know, they um these guys are absolutely crazy and being fiscally irresponsible. Uh but I think if you have any confidence or if you see any um if you see up close the demand for intelligence that people are snapping up and immediately consuming whatever you give them, then like yes, actually you're trying to get more compute and that that's the game is not to reduce the commitments. It is actually to get more commitments. They would have a
32:16trillion if they could.
32:17Yeah. I don't think I've spoken to anyone in the AI industry, certainly not any with the big labs who's who has said, "Yeah, we're, you know, I feel I feel we're good. We got the compute we have." It's it's almost always and and I think a underrated factor on compute demands in the future will be compute that's used for safety, alignment, and verification. Like I think more and more compute is going to go towards those functions. Whether that's on research or just on
32:44verification and alignment that happens at inference time that will be an increased compute demand for all these.
32:51But we're probably not going to break those out, you know, but like it'll it'll be like maybe 5% now and 20% inf I would just tell people like if you think this is bad now, wait till you see the open one.
33:03Uh and then like it's not bad. It's like that's the whole point is like they're actually really trying to get compute from every source they can get and like open buying TPUs like [laughter] like that was a cold day in hell TPUs right they have their own chip uh initiative Jalapeno which has pretty rave reviews staffed up a compute team to build their own chips obviously the hyperscalers are building their own chips it's a I mean every mad dash it's a mad dash
33:30only you know only winners are TSMC and uh Nvidia and ASM ML uh whatever in the supply chain stack.
33:37Totally. And and I think look it's it's getting harder and harder especially in the United States. You have a lot of push back. Um I mean you have all kinds of push back. Data centers are too noisy. Don't want them in my neighborhood. They add a lot of air pollution. Don't want it. It's going to the water. Oh my god. The water.
33:50The water. It's which is it uses too much water which it doesn't. Um it's the electricity. It uses a lot of electricity. It's going to increase my electricity bill if I live in close proximity. or it's just the simple AI is going to kill us and I don't want to enable that which again you have all kinds of push back. It's become hyper hyper political but I think the unfortunate thing is it has become a lot harder. You do see a lot more push back.
34:16Um midterms are coming up right who who kind of like takes control in like the House and the Senate like could have a big impact on regulation or lack thereof around data center and intelligence center buildouts. And the good thing is that it's getting harder and harder to build these data centers. Uh behind the meter is harder and harder. And so you do have a natural governor on the buildout and you're we're I think we're a lot less likely to overbuild too quickly. That leads to a supply glut, right? Kind of the the internet bubble analogy of the early 2000s where we
34:46built way too much fiber. There wasn't as much demand for the internet. There was a lot of dark fiber, right? Fiber that had no bits and bites traveling over it. We have no dark GPUs today. Um, and so it's it's probably a good thing that there's some rate limiter. It will stop us from entering a huge bubble, but it's also bad because we need, you know, the industry needs more compute. Um, especially as these models get more and more powerful. You know, the security concerns go through the roof. You also need a lot more like the defenders need more compute. The defenders need more
35:14capabilities. Oh, yeah. Which, you know, I don't I know, personal belief, I I think we're going to be compute constrained for for far longer than um than the world thinks. Yeah. Um I mean I'd be curious if what the what the markets price the world as thinking and then doing the work on like what do we actually think and have the delta is probably you know could be 3 years could be 30 years.
35:37And and that's a very material thing to figure out.
35:40Yeah. Of course. Of course. I I think one thing to guide people towards is well there's all this political stuff and honestly I would say um Meta does a lot of work and I think increasingly OpenAI does a lot of work in like well here's how we we make sure we're a positive um presence in the communities that we're in right so yes you you do need to take care of that um and but I think like uh coming back to again research relevance ROI and um what it means about
36:08the industry um this profit uh move here. I I I I do wonder sometimes and this is a thing that has come up in our other podcast particularly tying in the research side is how much left is there to improve, right? So, was this kind of like a onetime upgrade of our inference stack which you noted was like something like a 7x improvement in in our inference efficiency because like you can't like probably like there's no improvements
36:35for ever until the end of time in our inference stack like unless we fundamentally changed the chip.
36:41Um we pro like we probably started out pretty unoptimized and we're now a lot more optimized than we were 2 years ago. Okay. Well, there's a limit to how optimized you can get just like physical like jewels moving around and bits and bites moving around.
36:54So, uh that that is my question in terms of like the the how much efficiency there's left to squeeze and maybe there it might cap out, you know, in terms of like that that year-on-year growth.
37:05Yeah. Well, look, I I think um I I have a chart I'll get to in a second, but I think this is a debate that's existed for quite some time, right? You had, you know, the I don't remember when this was two years ago. uh you know like the Ilia moment of like hey the data and the internet is a fossil fuel. We've we've exhausted it. There's no more data left to train on and like we're hitting a plateau. Yeah, that was years ago. Then we hit inference time reasoning time compute and reasoning models. You hit a new paradigm. Um, but I think maybe
37:34embedded in your question is well, is the gap between the frontier and the leading openweight models is that shrinking? Like is is the gap that the frontier Yeah, I don't mind the gap. It's not something I talked about. It's but I think it's interesting because like what the real way to think about or to answer that question is not the the gap between the best leading openweight models today and the frontier models today because the reality is is open anthropic they have far more powerful models that
38:02they've developed that they just haven't released yet and like this is a chart that Mark Chen published uh from open eye actually switching gears from anthropic to open eye he's been on the pod cooking pod oh nice we we made some soup for researchers amazing amazing that's I love that bod um but you know when they saw Navier Stokes that big math uh problem you know Mark came out and said hey here's a model maybe dark blue light blue is harder to see but like the top bar is an internal model we used to solve that
38:30problem uh the blue bar is Astra hey look at how much more powerful the internal model is to what's publicly out there which on some ways that like could scare you a little bit right like hey these labs have even more powerful models so if you think about what's the gap between openw weight and frontier today. I think it's actually far wider than people expect because there are just far better models at the frontier that haven't been publicly released yet.
38:57Either whether that's safety, testing, alignment, um some combination of everything, you know, unclear, but I do think you're still seeing the frontier accelerate. Um even if it may appear that um what's released is plate. I'm not even sure if it it does look like what's released has been plateauing.
39:16No, no, no. No one's saying it is plateauing today. It's it's it's we're all discussing could it plateau in the future? Yeah.
39:22Right. Because that that is a pretty important thing to figure out for your investing.
39:27U which is uh part of the the deal here.
39:31Um yeah. So so that is a very well-rased point. I thought you were going a different area where you were going to argue about bitter lesson and you didn't. So kudos on that. Um and like I think that's a very very fair point that I I I think we're uh we've got still every generation model generation still shows quantum leaves. This is only in math but open has a pretty good track record of generalizing beyond just math or or coding. Right. And so I I I do think that is my point was a little bit
39:58more as the financial analyst hat on which is you had this one time 7x improvement in revenue efficiency or whatever.
40:06Um and like it's not like they raised prices. Yeah. Yeah.
40:10Right. So, like where is it coming from?
40:12Oh, here's another good one.
40:12I mean, they didn't raise prices here. Oh, I think I have.
40:15So, so you know what I'm saying, right?
40:17They have 100. They have one.
40:19Yeah. It's actually opposite. This is an awesome chart. This shows the cost to complete.
40:24They're charging less.
40:25It Yes, exactly. The cost they're charging less and making more money.
40:28The cost to complete a task with each model family. And so what you see is with whether it's Fable or Opus and then you have Fable 51 um and Opus 55, you see about a 25% 25 to 40% reduction in cost to complete the same task with each successive model family. And so it's kind of like yes, the costs are going the costs are going way down. Their compute efficiency is going way up. And I think a lot of that just demonstrates
40:57I mean again it's going to sound a little like VC cliche um but like Jebman's paradox like as things get I know I know I knew I knew podcast on AI without that coming up it is this is the appropriate time to bring it up.
41:10And as as things get you know you go back to like the early days of the internet like how many bytes were traveling across the internet in 2000.
41:17How many bytes are traveling across the internet today? Like many many many many orders of magnitude more. And guess what? The price to transfer bytes over the network has like dropped like like many I mean many it's P and Q, right? Price has gone way down, quantity has gone way up and I think you're going to see the same thing uh with people buying and serving models and tokens.
41:42Yeah. So so I generally pro supported that. I I do want to remind people that that is uh true objective fact. This is not like VC like this is like actually how the the businesses are running and it is surprising to us.
41:56That that I get while yeah we read about this in theory but like to actually see this in financials is like really ridiculous.
42:03You'll continue to see it. Think about like Jeb and decision models you know and like the price point of hey instead of you want a a binary classifier uh you don't need a language model to do that. Just go use a decision model. Super cheap efficient. I think you're going to continue to see efficiencies like that.
42:19You know, I go back to S3 and EC2, AWS's big initial products, storage and compute. You've seen, you know, 99% price reductions in both of those services from the day they were launched. But like what's happened to the EC2 and S3 business or just the broader storage and compute, you know, more CPU compute, uh, it's it's gone up exponentially. I mean, all the hyperscalers reported today, they're all over a hundred billion dollar run rate businesses today if they were individual companies. Google Cloud's grown over 100%. It was just which is crazy.
42:48Azure's growing almost 50. Amazon reported today, we're recording on Thursday, 39 almost 40%. Like these businesses are just accelerating and growing at it's crazy these there's you should drop everything you're doing and just go start one of these things.
43:04Uh yeah, it's October 29th for those those uh keeping track. And and yes, uh a fun personal anecdote, I was in my hedge fund when uh the day they broke out AWS from Amazon. M no kidding.
43:14And at the time we were not long Amazon because we were worried about uh AWS margins and also like oh is it just startups and what if like you know we go through uh a startup winter and like all that. Yeah. It was so nonsensical. So like it it the the funny thing about finance is that you can do all these works and make worse decisions.
43:33Yeah. [laughter] We definitely did that. to totally totally I mean that that's the thing like market timing is impossible like and also fixating on the wrong thing. Yeah. Which a lot of people are having to encounter for the first time. That's right.
43:47You have 2,000 data points that were released in this thing and you'll fixate on one thing and they be like that is that's the everything you need to know and like well no like you you lost sight of the bigger picture.
43:57Yeah. I mean that's that's for me. I mean again there's a lot of things that I believe that others would disagree with me on but I believe there's one question that matters and that's canthropic stay at the frontier. Yeah.
44:06And if I believe the answer to that question is everything they've done historically has demonstrated that I think they'll build a great but then like there's a right price for that.
44:17Of course. Of course. And this is this is not like will it be priced right or this or that. That's more of hey if you're able to take a longer term view on will the be will the business be successful over time. I think to me that's the most important question. Now whether you think the IPO price is cheap, expensive, just right. I mean, we don't even know what that's going to be yet. Um, prices can change of public companies for idiosyncratic reasons, reasons based on individual company performance, or they could change for macrocorrelated risk. Like the reality
44:45is is if anthropic went public um a month ago, you know, the the big route that you've seen in public markets recently, especially in semis and the AI trade, anthropic would also be down even if the business was doing well. So, stocks could go up and down based on individual company performance as well as just market beta. Exa. Exactly.
45:02Exactly. you know, uh, you should be prepared for those kind of discussions if you're getting into this. I think, um, actually, just as a a side question, there's a lot of these like, um, non-public market trackers of the stock, like like here's my proxy of OpenAI stock, here's my proxy of Enopic stock. Do you guys put any weight into that or like zero? You I don't Yeah, I don't.
45:22Yeah, I'm just I I'm wondering if I should pay attention to those things more closely because they they do give some indication of like, well, here's the stock price reaction when the model just re I mean, it might it might show some market sentiment, right? But I think Ramp had some data.
45:36I think I copied a RAMP chart um the other day. I mean, this was anthropic spend um on some of their most recent models. The dark blue is Opus 55 and like, hey, what what the ramp data shows is people switch.
45:50Yeah, exactly. And and with with Opus 55, you saw some reaceleration. And so, would it surprise me if you saw a reaceleration of revenue and run rate in October? And October won't be in Q3. That's the first month of Q4. No, it wouldn't surprise me because I do think a lot of the third party data is showing and suggesting that. Yeah.
46:08Um well, so okay, let me this is this chart is super interesting, right? Because you you just said like when you when you say stay at the frontier, there's different levels of that.
46:17Stay at the frontier, meaning we're one of the top two, one of the top six, one of the top 20, whatever.
46:22Um, usually people mean top two.
46:25Usually, but sometimes you can expand and say top six. Yeah, Google now has a has a claim to be a number one in some metrics totally.
46:32Uh Meta doing super super well. Uh SpaceX doing super super well. Not at not number one, right?
46:38But number three, number four, you can say that.
46:41So look at that like summer slump in anthropic revenue.
46:46Um was that because OpenAI was briefly ahead?
46:49Yeah, I mean I I think Astra was a great model.
46:51And I think a lot of people started using Astra. Um and it wasn't just Astra, right? It was what was Google's models? Um what was the one they just released? Carbon or argon?
47:01Carbon. Yeah, carbon argon. Yeah. Yes.
47:03Yes. Next yesterday. Uh they haven't released it yet. This is my gripe at Google. They never released anything.
47:07They just announced a weight list and then like god damn it. They just I I think it was also you saw a massive explosion of you know Kimmy K3, the GLM series, Deepseek models. You saw uh openweight models get really powerful um and and get good quickly. So I think you kind of saw all of that but it's it's crazy like the far right bar like what was nonopus 55 or sonnet 55 like that
47:34has gone that's h haveved or more maybe more you know in what is that 2 3 months and so like you do you imagine if they didn't have opus 55 or sonnet 55 right like who knows what the revenue growth and trajectory could have been so like that's kind of why I do think it's important to be have a frontier class Yeah.
47:55And so I I will say from watching these labs for the past 3 four years, they used to have more of a reserve like I will have internal models that I can just release whenever I feel like it.
48:07And um and then I think they towards the GPT56 era, everyone just released whatever they had when they had it.
48:16Now I feel like the reserves are building up again. I don't know why, but like I it definitely seems like like as you saw with the internal model, uh it it seems like they they have more wood behind the shred for stuff like this. Like I can I I sitting on Fable 55. I can release it the day before I IPO and like like that'll that'll make my IPO go well.
48:34You know, [laughter] I mean, look, I I I think there's I think there's good reasons, you know, that the models aren't released right away. You know, on one hand, there is just a lot of like safety and alignment and all that. But as a PM, as an as a as a executive like playing this game, yeah, I can play it pretty well if I have a background.
48:51Oh, really? I mean, and and it's also think about um you know, open eyes patch the planet initiative, which is basically, hey, let's use these super powerful models to go find vulnerabilities.
49:02Not even it was 56. It was 56 soul. They found the PHP vulnerability. It wasn't frontier.
49:07Yeah. Which is crazy.
49:08It was just a spam a bunch of like like our last gen model.
49:11Exactly. Exactly. and then let's let's find all these vulnerabilities like the PHP one. Let's go create the patches so that we can instantaneously release the patch while announcing the vulnerability. The challenge is attackers are using these models are using really powerful openweight models and they're finding these vulnerabilities themselves.
49:27Yeah, it's a race and it's it's a race and it used to be uh you know you had weeks months from like vulnerability discovered to like exploit. I mean now it's collapsed and it's basically zero you know it's zero days like vulnerability announced or discovered exploit available like it's it's almost negative. Yeah. Um and I think in you know OpenAI's announcement of the PHP thing I think hidden in there was I think they basically said like oh and we we there are more serious vulnerabilities that we know about that
49:57we're working on patching right now.
49:59Which I mean more more more critical than you know something that touches 70% of the internet. um is is is pretty serious.
50:08Like the 70% of the internet is just a proxy for old.
50:11And like of course it's older and it's going to be more vulnerable. Yeah. Yeah.
50:16That's um I think uh there's a couple of takes. I'm not that interested in like dwelling on on that stuff only because there's going to be smarter people than me on that.
50:25Of course, the there's one angle of this which is kind of funny which is like well we're spending a billion tokens so finding all the vulnerability and we're going to spend a billion more to fix the vulnerabilities. What if we just didn't do either and it would look like net the same position as we were at?
50:39Yeah. Yeah. Yeah. Yeah. Well, I think cats out of the bag on that. You can't put the toothpaste back in the tube.
50:44So that but that is one argument people make like that blame the AI instead of blame the they like kill the messenger.
50:48The the other part which is interesting is like I and this might actually uh be an issue for basically nation state actors uh who actually had all these exploits already. Yeah.
50:59And now like well you know civilians are finding them. God damn it. Like now I have to work harder.
51:04Yeah. Exactly. like all my zero days are gone. Like I had them for 10 years like whatever.
51:08But anyway, so so these are fun anecdotes or consequences of this.
51:13If you want to really do superst and makavelian if you're a nation state actor, what do you do to to preserve your edge because like you cannot have the a channels that you're relying on lock up.
51:25Yeah. Uh, so then then then you send the Saudi decoy to put the thing behind your TV so that people so that so that they get distracted by that so that your real spy Yeah.
51:39[laughter] That's just some war games.
51:42I I I watch too much TV to to to think that what we are told is the full story.
51:47Yeah. Yeah. I think that's fair, right?
51:49I think that's fair. Too much is at stake.
51:50Yes. Too much is at stake. Um, okay. So coming back to the actual again finance like I really want to be a primer for everyone who are new to the economics of models and model labs and what what have you. Um one thing that have that people have said is that the R&D per model let's say like what does what does Fable 5 s 5 opus 5 cost let's call it a billion dollars that lasted me 6 months right because you just said the dep the depreciation
52:20it lasted that six months. Yeah.
52:22And so isn't that worrying, right? Like about like well these things are the fastest depreciating assets in history. This is something that me and Sargo have talked about before.
52:31Yeah. Yeah. Yeah. Yeah.
52:32Um and I think the the net is this no but like it's a qualified no of well as long as my next model run is better than the than the than the current state of the art we're good.
52:44Yeah. Y it's a little bit of a house of cars if if the next one doesn't work out.
52:48Yeah. which um briefly affected OpenAI.
52:51So, a lot of people know by now that, you know, like GPT45, the the big the big run wasn't wasn't that successful.
52:59Um, and they, you know, they they they since fixed it, but like that a slump in a in a run or a failed run, uh, can really really screw with the the ROI on the thing. And look, I think there's a really good analogy to the trip industry. And I remember speaking to someone, you know, much older and much smarter than myself who did a lot of semiconductor investing and, you know, in the n 80s and 90s, um, about all these chip companies.
53:22They've seen it all and, you know, they've seen it all and they were highly skeptical. Um, and, you know, sometimes you have industry vets that are highly skeptical and, you know, they miss the next big opportunity because they never think the the incumbent can be disrupted. But the point of view they had was it's very different to tape out a chip that is competitive with the leading chips at the time from building a machine that can continuously tape out chips that compete with, you know,
53:50the the chip factory. And so he was like this this individual's point of view was maybe a chip startup can build one chip, but guess what? they're not going to have the machine to constantly pump out a new chip every year and stay at the frontier um from the chip standpoint. And that's why Nvidia is king. It's not because they have the best chip now. It's they have the best factory process for chips.
54:09Exact. Exactly. It's the machine that builds the machine that and so it's kind of one of these I mean this is the bare case on all these neolabs which is you know and and you talk to many smart folks and what they might tell you is hey if you want to really compete and be have like a soda model at the frontier you can't you can't lo you can't uh go more than one order of magnitude of compute less than the labs you just can't compete if you don't have you know one order of magnitude of compute of the big labs um and so as a Neil lab you know the question isn't can you build one model
54:38it's Can you attract compute and talent required to continuously pump out new generations of models every year to ultimately compete in your domain? And so I think for the labs the reality is is I think it's very hard for anyone unknown players to catch up to kind of like the generalized frontier big models. I think you'll continue to see domain specific models um or models or
55:08like new types of models, decision models like Jev. Um but I think the I think the big labs they are on the treadmill of continuously pumping out new models because guess what? If if anthropic stops, OpenAI and Gemini will continue and then you'll just kind of see re you'll see revenue fall off which is why I do think it's so important to stay at the frontier. Now what do I expect and like what have I seen from
55:34anthropic now in the data it's as their model families get better and better they get more efficient but they also capture more revenue that could be because they're doing more economically valuable work or they're doing the work faster. Um but what you're seeing from now kind of black and white is hey as Enthropic has 2xed their compute spend uh to build and train these models they've 14xed their revenue and like I'm
56:03not really sure what would cause that trend to stop. Um, I think models are only going to take on longer running tasks, um, more economically valuable tasks. And, you know, if you can't tell, I'm a hyper optimistic person. And so, maybe I'm just drinking the Kool-Aid too much. Uh, but I I do think we're nowhere close to hitting a limit of the dollar kind of like economic value that leading models can capture. I think what's maybe
56:33more interesting is you know OpenAI you know Luna class models kind of also trying to capture and compete at the low end of the market with much smaller uh much cheaper models you know and kind of span everything from um frontier to you know distilled versions of their own models that are smaller in and you know cents per million input output tokens not dollars.
56:58Yeah. um uh cosign all of that. I think uh that is a good way to think about it. Um I I I think uh you know funny I I'll just comment this because it's in my head. Uh 61 Terra has been pretty pretty successful. 61 Luna like that that whole series is very interesting. The approaches of the different labs is also super interesting.
57:16Um Anthropic to this day has no decision model and they're like well we're good like use type safe whatever. Um, and like well they don't have a standalone voice API.
57:26They they have a voice model. They just don't don't market it. They don't they've never done images. You know, it's a lot of enterprise focus and coding focus and OpenAI has been a little bit more of like we'll do everything and we'll try Sora, we shut down Sora.
57:40Remember when Sora was like the number one app in Yeah, exactly.
57:44Um, and then Gemini has been the most like we will explore everything. Um, and so like we appreciate them because if they didn't try it, no one else would, right? like they have the funds for it.
57:53Um, but like I think often about anthropic in terms of like well okay enterprise and coding and they they're first to it. They nailed it. They still continue to nail it right and everyone's chasing after them.
58:04Um is that the story going forward? Is is is it just more of the same or like how what should we expect?
58:10I think it's I think a lot of this boils down to like a new type of innovator's dilemma, you know, which I almost call like the compute innovators dilemma.
58:17Like let's say you're open and you have Sora, you have many different model families, you have image models, you have everything. Well, the reality is is how do you choose what to serve? Maybe Sora demand explodes. Are you going to allocate a lot of your codeex compute to slot videos? Yeah.
58:37Um because you probably will have to like if you actually want to make I mean that's exactly why they shut down experience good you need to. And at the end of the day, what a lot of people have decided is, well, guess what? A lot of the the highest ROI, most economically valuable work has been in coding. So, it's hard to prioritize anything else other than that. It's why there's probably opportunities for other companies and startups to build models and build products in those domains because guess what? What is open ad anthropic chasing? They're not chasing the incremental 10 billion in market
59:05cap. They're chasing the next trillion.
59:07And so, the reality is is for a startup or a Neolab, great. You could go build a build a company, build a product, build a model where you exit for $10 billion and it's a home run, grand slam. Um, that's awesome. And simultaneously, it's too small for OpenAI and Anthropic to matter. And so, like, what is Anthropic done just a bangup job at? They said coding is our thing. We're going to be great at coding. We're going to build the best coding harness in cloud code, and we're going to continue to nail it.
59:32I think everyone is kind of starting to ask themselves well when do we maybe you asked a version of this earlier when do we tap out the coding tam um I think one point of view is I have asked a ve very many times this question but I think one point of view is hey coding is actually the best generalizing model and framework and like that can be applied you know you think of it as coding but when I use cloud code like I'm not coding but I'm using cloud code when you use codec are you are you coding no
1:00:00these are they're generalizing and coding is a good framework for generalizing a big push by both openthropic into life sciences and drug discovery. I think they've both been quite public um about their interests um to kind of like capture that domain as as a new field and I do think those are just different.
1:00:23Um it would seem like Enthropic is investing more than OPI on specifically the life stuff.
1:00:28Yeah, I' I'd agree. I'd agree. Um but now I can't remember the question you asked. Um but I but I do think you're starting to see these companies evolve the scope of what they want to go after and it has been very targeted on coding given it's been the best generalizing.
1:00:44You know you want to generalize knowledge work. These models have generalized the best as they've been stronger in coding. Um but it also can give it some a soft underbelly to other areas. Maybe it's visual thinking and reasoning. You're seeing great models there like Allorean. Maybe it's decision models like Jev and Types safe, right?
1:01:00Like there's there's a lot of different uh soft underbellies they may have, which is opportunity, which is great. It's opportunities. Like you'd love to see better, cheaper models in other domains. You know, you don't need to drive the Ferrari engine to the grocery store. Hey, maybe you a Toyota will do just fine. [laughter] Uh and I think that will continue.
1:01:18Um and good shout out. We just interviewed Andrew on his benchmark and and glimpse I think glimpse is the the new are you guys invested or we are promotion. Wow. Okay. All right.
1:01:28Uh it's it's it's weird to like this is a whole different discussion but ultimately I don't get as like well you will lead the giant rounds in like you know the anthropics of the world but you also care about Laurian. It's like way smaller and like you know probably worth like a minute of your time.
1:01:43Well no like like we're worth a ton of my time. I mean I would say like we barbell. I mean, we're life cycle investors. We invest early. We invest late. We invest in the public markets.
1:01:53And I think there's a lot of value. Like the way we think about it is some of the best founders today. They need latestage capabilities early on. And latest stage capabilities isn't just capital. It's relationships with execs. It's relationships with public companies. I mean, it's companies are built on on hyper mode now, right? It's not let me go raise a round and let me, you know, add 10 million of AR in 18 months and I'll go raise my next round and then two years later. It's like you're you're growing to 100 million super fast.
1:02:19You're you're raising nine figures of of money super quickly and it's and it's hard. I mean in many ways you like you have to like break the laws of physics to how like cultures can be maintained. Lawyers lawyers are not set up for this. They're like what do you mean when we come on a weekend to because like Monday might be different.
1:02:33Yeah. Exact exactly. Exactly. [laughter] I mean, it's, you know, you used to think about as a board member, an investor, hey, let me think ahead. Let me make sure we hire the VP of sales and the CFO, you know, 6 months before uh before we need to, so we're not behind.
1:02:47Now, it's like, hey, 6 months before you need to, may have been before the company was even incorporated, you know, and so how do you how do you do that?
1:02:53Like, when do you hire the right execs so you're not doing it too late? How do you hire ahead but not too ahead? I mean, it's I feel like the pressure of building companies has never been greater. Now, the size of the prize has never been greater. Um, but it is it's as you know it might appear as easy as ever to build a company but I actually think it's the opposite. It's almost as hard as ever to build a good company and maybe more importantly to build a durable company which is I think what what everyone aspires to do.
1:03:17Yeah. Uh I'll also shout out the you know on the model factory thing. Um we also interview poolside which acquired by Nvidia they were pitching this model factory really hard and and I do think like uh the business that builds a business is kind of like the the general uh takeaway that I want people to know.
1:03:32Um, one of the things I wanted to I was diving towards I didn't I didn't quite start it yet but I'll start now which is typically when you have an S1 you have a clear sense of the business uh let's call it market segments product segments the org chart the whatever um and we got cloud code cloud code work miscellaneous or platform whatever u and I'm like okay what about legal what about design like all the stuff that you've cloudai the consumer business for a while that was a thing
1:04:02Um, do we care? Do it's just all like one revenue bucket, you know?
1:04:07I I think we do care like what the products, right? Like this claw code and what else?
1:04:14I I and this is where I do think and there was there was a section of the S1 um where they broke out some of these like vertical applications they want to go after. I mean, it's kind of funny you bring this up. Like I had a hypothesis, you know, years ago that like, hey, maybe the labs will stop releasing models and instead they'll just build applications and products powered by their models. And if they're powered by the best models that are never released publicly, hey, maybe they have the right to build all the best applications
1:04:42that you kill Harvey by only using Claude Fable 6 legal, right? Or it's, hey, you know what, cursor, guess what? We have well, you know, years ago, we got the best model and we're not releasing it. So guess what? But we're going to have the best coding agent because our model that's the most powerful isn't released publicly. And I do think it's why some of the leading application companies today are moving towards in-housing their models. They want to control their own destiny. You need to I think it's going to be really hard as an
1:05:10application company over the long haul to survive without a research team internally. Um it's one of the like an investment thesis that I have that if you know investing in applications like I want to believe there's two things really strong product org but also a really strong research org and I also want to believe that they're aligned and as the research gets better the product gets better which sounds counterintuitive but there are a lot of research orgs that are just doing research for the sake of research versus closing the loop and actually like
1:05:38having a a positive cycle of of product improvements with research improvements. I'm I'm not a you know I'm not in a topic where I have the benefit of being friends with fair fair amount of people in the topic. Uh and I would say research is king topic.
1:05:51Yeah totally totally by far. And so part of my whole thesis around the model labs versus agent labs is uh model labs will you know respect the model and ag and try to build AGI and products will be a final last mile.
1:06:06So the delivery mechanism whereas the agent labs will will respect the customer the most and build whatever they need and then and they'll they'll backwards integrate to models as and when but they'll never be probably never be at a frontier for for those things and both can coexist both have a marketally um maybe to your question I do expect to see a much bigger push by both labs anthropic and open AI into vertical applications and kind of like diversifying out of just delivering models. I I think I I think that the actions of Enthropic show so far have
1:06:35shown no serious intent at it because cloud design is just a tab inside of cloud.
1:06:41It's not like a whole Adobe you know replacement cloud legal is a markdown file you know markdown repos uh which like that is the AGI pill is if if research is king and like you know one model does everything that is what you do.
1:06:53Yeah. This is classic enterprise and org management. You have a cash king product that think your sales reps, what do they want to sell? The thing they know how to sell that they've been selling that is crushing it. What's harder to sell? Like new things, new SKUs. And so it's inertia. It's momentum. And this is just classic like big company boring stuff.
1:07:12But it's just like, you know, I I think for someone like I've been through maybe 15 years of IPOs. And for an IPO stage company, much less a $2 trillion you've never seen like, well, we have three SKs.
1:07:27Yeah. Yeah. Exactly. Exactly. But but I think to anthropics credit, they've been really good at focus and like they haven't been, you know, enamored with the next thing or no robots.
1:07:38No images, you know? It's And again, you look at Open Eye, it's tough. Like who are they competing with? They're competing with Nvidia. They're competing with Google.
1:07:45They're competing with Meta. They're competing with Anthropic, you know, like they're they have different lines of business that are all competing with giants, like the biggest tech company in the world. And it's like it's it's a tall task to take on one at a time, let alone all of them at a time. But hey, that's that's the game you take on if you know when you're a trillion dollar company, you know, the the stakes are just higher.
1:08:05Yeah. Um, okay. I want to round out the entropic discussion and then we can go to general commentary on on other stuff. We've typed the main debates that you've you've been engaged in.
1:08:18There's um there's two others. So um this is going to be very like financially in the weeds. Um there if you look you want to talk about convertibles.
1:08:31Yes. [laughter] Okay. So, um, on the balance sheet, you know, earlier in the episode, we talked about, hey, 140 million of cash raised, uh, capital raised, 134 billion of cash.
1:08:42That's what this the balance sheet shows you. What the balance sheet also shows is a I'm I'm going to I might mess up the exact numbers because I don't have them in front of me, but I think about 108 billion of convertible notes payable as a liability. And so, the question is, well, is that debt? Did they raise debt?
1:08:58Um, and what you also see is a huge portion below the operating income line of like other income and expenses that makes it look like they have a huge net loss. They're both related. And let me explain. Um, so any company and any startup founder listening to this would understand and appreciate this. Like there's different ways to raise capital. You can raise preferred equity, you can raise converts, you can raise safes.
1:09:20There's a whole different host of options to raise money. Enthropic just like every other startup did everything.
1:09:26They raise some preferred equity, they raise some converts, um they raise some safes. And along the way, there's different reasons that certain investors of the company might prefer one or the other. When it comes to strategic investors, there's even more reasons why they might prefer one or the other. And I think it's it's very well known that some of the strategic investors, Amazon and Google in particular, were very large investors and very big owners of Anthropic early on, right?
1:09:51Yeah. Go ahead. Sorry. and and so early on Enthropic and Google 23 24 25 a lot of the dollars that they provided the company were in the form of converts and safes and notes. Um those the value the the value that those converted at at the time was pegged more towards the valuation those companies were raising at at the time. Why might they prefer a convert? Well, they're public companies uh Google and Amazon. If they have an ownership in a third party that exceeds
1:10:19a certain limit, they now have to start reporting more on that look through ownership. Right. If you own 20% of another company, you almost have to.
1:10:28It's consolidated financial statements.
1:10:30Exactly. And that company's losing a lot of money, you now have to report that loss on your books. And so it's they didn't, you know, strategic don't want to hit certain ownership thresholds. So, hey, give us a convertible. Give us a note. We'll give you cash. You haven't given us shares yet. And so we don't have that equity ownership. We don't we haven't gone above those thresholds.
1:10:48Maybe Anthropic didn't want to give them voting shares. I don't know, right?
1:10:52Let's let's have your shares convert into non- voting shares because we don't want the governance of a of a supplier or strategic like I'm I'm totally speculating and hypothesizable reasons to do reasonable reasons why hey they were converts but the conversion price was at a much lower valuation relative to what it is today. Okay. And so when you look at the balance sheet what and balance balance sheet have to balance you have assets equal equity plus liabilities they have it has to
1:11:22balance and so when you raise a convert you get cash your assets go up well if your balance sheet's going to have to balance your liabilities also have to go up and it makes sense a convert is a promise for shares in exchange for cash and so Enthropic took on that liability and at the time the liability ility matched the cash raised. Now, what's happened to Anthropic's valuation since 2023 and 20? It's it's gone way up. And
1:11:51so, now what happens is the dollar value of your liability, because you promised shares at a price, the dollar value of that liability has skyrocketed because the valuation of the business has gone up. What does that mean? Okay, now the liability looks big. It looks like a $ 108 billion debt that Anthropic raised.
1:12:13The reality is is they didn't raise $ 108 billion of debt. There was about 11 billion in those combined convertible notes. About $6 billion has already been no $5 billion has been converted to equity. There's 6 billion outstanding.
1:12:25The market value of that 6 billion is 108 billion because the company's value has appreciated meaningfully. And so there's this debate of like, well, they have 108 billion of debt on the balance sheet. They haven't raised 141. they've raised 141 plus 108. Um that debate is just misguided. Um there is no debt on the balance sheet and the 11 billion of converts is inclusive in the 141 billion of total capital raised. Um and so debate one of the say good job
1:12:56Yeah. But but it it actually goes deeper. Yes.
1:12:59Um that's just the balance sheet. Okay.
1:13:01Now let's look at the income statement.
1:13:04They have an operating profit. You could say great gap operating profit, but what comes below operating profit? Net income. Someone could say, well, in the first six months of 2026, they had negative80 billion go of net income. So, you could say, and this is a debate. This is what people are saying. Again, it's very misguided. They're losing $80 billion. Like, it's true.
1:13:28What a terrible business. Technically, by gap accounting, it is. But, okay, let's go back and I'm going to go back to the balance sheet. the liability has gone way up. That liability is now 108 billion. Okay, let's think about the the balance sheet. Assets equals liabilities plus equity. As liabilities have gone up, two things have to happen. Either your assets have to go down or your equity has to go down. Well, your assets haven't gone down. That means your equity the the value of your equity has
1:13:58to go down. This is where it might get a little complex, but how the how the three statements link is yeah, people can take accounting 101.
1:14:07On the P&L flows to retained earnings on the equity value of your balance sheet. And so what has to happen as that liability has gone up, retained earnings has to go down by the same amount, which basically means net income has to go down by the same amount because net income on the P&L flows to retained earnings on the balance sheet. It's all mathematically linked. You can't break it.
1:14:29It's all mathematically linked. You can't break it.
1:14:32But what's happened is in order for those statements to link and for those statements to tie, you need a huge decrease in value on your income statement. And what what happens is guess what? Anthropic's value has just skyrocketed recently. Um they've raised big rounds over the last 9 months. Um, and so as those converts have appreciated in value, the value that those notes have appreciated shows up as a loss on the P&L. And of the $81
1:15:01billion of loss in net income on the first 6 months of this year, 72 billion is from a series of converts for Anthropic uh for AWS and Google and 9 billion is for a separate set of safes and notes. And so 100% of the loss in the first six months is just the value appreciation of their converts. If Anthropic never issued converts to Amazon and Google, just did straight equity, that loss would it wouldn't be
1:15:31And so it's kind of one of these like technically from GAP accounting, they do show an $80 billion net loss.
1:15:38Do I view that as an accurate representation of the business? No.
1:15:42Because the reality is is 100% of that loss is just the appreciation of their equity, which everyone would say is is a good thing. And as those notes convert to equity, those go away because you no longer have this dynamic of your liability going up as your stock price goes up, which means your net income has to go down over here. So, you know, all the financial statements tick and tie.
1:16:07And I Yeah, sorry if that got too like financial.
1:16:09That was a really good recap of the discussion. That's should be a definitive answer for anyone uh who has come across that topic. Uh there's another discussion about the circular financing of Google and Amazon. We don't have to go into that. I mean I think that is small in comparison to the numbers we're talking about today.
1:16:26Um and uh in general like you know book value accounting is not that relevant compared to overall enterprise value. Um oh maybe one other thing related to this I think there are a lot of people that say okay well the only reason the liability has gone up is because the value has gone up.
1:16:41Well, is that does that mean that Anthropic issued shares today? And did they give did they give Amazon and Google $108 billion worth of value? But did they give him those shares at the price from four years ago? Did they just give them shares at a huge discount?
1:16:55It's like no, that didn't happen. What happened is they raised that money years ago in normal financings like a normal startup would and over time it's just appreciated in value.
1:17:04So, so you know, I think a lot of people if uh they're just not financially educated to that level because they've never had that exposure. So it is what it is. You're just educating. What was the second uh debate that you were referring to? You said you had two debates.
1:17:16Oh, those related. Yeah. So it was the balance sheet. Oh, wow.
1:17:19Have they raised more debt? And then it was the income statement. Are they actually losing money?
1:17:23As an investor, why do you get involved in these kinds of Oh, this is more of like I mean I'm you're just having fun.
1:17:30I think in thro look alttimeter is a firm. We do venture investing, growth investing, and public investing. And I spend a lot of my time on the venture investing the early stage. However, I just I consider myself like a student of finance and Anthropic is such an iconic company.
1:17:45You shouldn't have to educate people on the three statements of accounting. Like that's Yeah. Yeah. But but like hey, I mean I think a lot of people look at the balance sheet uh the income statement, the P&L, but the reality is is there's a lot to learn from the balance sheet and the cash flow statement too. Um and so this was more of a hey as just someone who's intellectually curious. I mean anthropic is is a really important company to understand. I think if the anthropic S1 came out and the financial
1:18:11profile that was pictured was doomy and it didn't look like a good business, I think we have a lot to be worried about.
1:18:18Um because if they're not delivering profits and high ROI, what does that mean for everyone downstream of them?
1:18:24Like that's it's I think it's really important to understand the anthropic P&L. But what I understood from the anthropic P&L, we let me just rip through these really quickly. Um here's what you're seeing about their gross margins. Lots of leverage in the gross margins. They're improving rapidly.
1:18:38Which uh one one nuance on this is this is the half that is not through cloud cloud partners.
1:18:43Exact. Yeah. Uh this this No, this is just a gap gross margin, right?
1:18:49I I I don't know. I I read something about like how obviously cloud partner margin would not go into gross margin at all and that's 50% of revenue account I believe is a lot of the cost that they pay out the rev share is in sales and marketing which is below the line which is which I think what you're saying which by the way $3 billion of sales and marketing yeah I think it's like 40 something% of the revenue comes from cloud partners it's it's not a triply amount 48 48 okay yeah I was close in so you see gross margin profile improving and and
1:19:18sure maybe it'd be lower if uh you kind of factor in the third party, you know, that the cloud providers, but like the trend is going up. You see operating margin going up, positive gap operating margin, you see f free cash margin going up. And so to me, what I see is a highly profitable business that is probably going to flip to reinvest in compute and buildout. I think the way compute gets built um is going to look different in the future. It's going to be expensive.
1:19:43There's going to be more upfront payments, right? like I do think you should expect to see the profitability profile of these labs um kind of go the other way in the short term as they shift to investing in growth and investing in compute. Uh but the biggest takeaway for me was it's a highly profitable business that's still operating in growth mode that's seeing some reaceleration probably you know in the third party data from the new models which gave me a lot more confidence for the rest of the industry. So long as the revenues show up, the compute
1:20:13obligations will be hit. Um the supply chains will, you know, continue to be saturated and constrained and companies will continue to perform. And OpenAI and Anthropic, the reality is is they're a huge percentage of the overall AI revenues. And so everyone looks to them as a proxy for the rest of the industry.
1:20:28So if they keep growing, which again I think a lot of the bears would say they're going to stop growing because openweight models are going to eat their lunch like this, this and that. Um but I think those are the important kind of questions.
1:20:40Yeah, very important uh very good questions. I I think like Yeah. Anyway, uh do you do you dabble in like forward projections of numbers like what what like we we what what should we expect from FY27? What is expect Yeah. I I think that's that's too hard because the the the delta is very high, right? given the year they just had, it's it's a business that is so hard to project because the future is so like unknown and unknowable. Um, you know,
1:21:09they'll go public. The sellside research analysts at all the banks will will build their models. We'll kind of have a forward a forward number. I don't know where that's going to shake out. Um, but I I I do think you'll, you know, you you'll see um a lot of optimism um about what the future of this business could look like. But again, it's would it surprise me if it was much higher than expected or some you know anything could happen. It's anyone trying to predict the future es especially in the short
1:21:37term it's it's as hard as ever. I think businesses at the frontier are the most predictable because they're the least likely to be disrupted but um I don't know we live we live in a highly uncertain time.
1:21:50Yeah. the the kind of A+ financial analyst as a PM in my hedge fund I would have wanted is to do physical limits uh calculations on like okay you have all these inference optimizations where like how much juice is there left on the current thing um let's let's say the nextG has has this level of inference how much can we charge on that and uh given the competitive profile and exhibited u pricing history of of these things uh and then finally how much GPU
1:22:19supply uh has been locked in and secured so that that translates to this number of tokens and that's what it can sell for because that's like the physical limits of what where these things can go because I think a lot of the financial analysts and I used to do this you you would just kind of had this spreadsheet of like here's the past performance and you couldn't click the right corner and just drag it out yeah exactly how it works can't do that I mean there's not good an there's not good analysis right totally I mean it's I think there's limitations everywhere everyone knows like hey Nvidia can only produce so many
1:22:47chips you know challengers cerebras TPUs can only produce so many chips, but I think there's other constraints. There's only so much land power shell um that's approved and permitted.
1:22:56Correct. So, you do that build up, right? And then you're like, "Okay, who has what share?"
1:23:00I mean, we're we talked a little bit about like debt and credit markets.
1:23:02Like, right now, you want to go raise multi-billion dollar debt for GPUs. No bid. No bid. They're like debt markets are close. They're, you know, big debt deals that aren't happening, right?
1:23:15Almost the equivalent of like an IPO getting pulled. Big debt deals. Why is there 100 billion in equity but no 100 billion in debt?
1:23:20Well, I think I'm going to turn around and do those put those money right in big credit funds. They're all looking at the look through exposure and any neocloud or a hyperscaler that's raising debt is the biggest question the credit providers are going to say is well who's the offtake? Because the reality is that's actually who I'm underwriting because that's the customer. That's the one paying that's the end payer of the bill. And if the end payer of the bill does poorly, the bills don't get paid.
1:23:47And right now, the ender of the bills are Enthropic and OpenAI. So many of the data centers and the the the being built by the Neoclouds are for OpenAI and Enthropic. And a lot of the credit providers are saying, I'm feeling kind of full, you know, I just did a ton I just did a ton of debt um for either one of these. And didn't matter if it was Coreweave or Nebius or an Upstart like a Volta or someone else. It's like, hey, I'm feeling full. And I think debt markets just are going to go through a little bit of indigestion. They need to
1:24:16figure out, well, do I feel good? Do I feel great? Hey, some of these projects that were promised, are they being built on time? Is the revenue flowing back to me? And I do think the credit and debt markets will eb and flow. Um, because there are just a thousand reasons why, you know, a project could get delayed, right? A Neocloud could go sign a contract with OpenAI. They could sign with uh Nvidia. They could go have Dell package them up. They could raise the debt. And then you know what? There
1:24:45could be local community uprising and picketing. I'm lit. I'm not even being kidding. And then it's delayed.
1:24:53Well, the debt was raised. Someone's got to pay the interest.
1:24:57What happens? Yeah. You know, and you have creative delayed draw facilities where you don't, you know, you the debt is committed, but you don't draw it down. But like all of these have limits, you know, like and you get super in the weeds on like SLAs's and and who was responsible for it. Was it a delay in the supply chain? Was it a delay in the construction? You know, the data center developer? Was it a delay by the operator? Was it, you know, something else like community unrest? And, you know, that's that's everything that,
1:25:26you know, credit markets are are thinking through. Um, and then there's a lot of other super technical reasons, you know, that we don't need to get into that, you know, make it difficult with credit default swaps and hedges going through the roof. And so now it's for debt providers, it's tough to hedge the exposure they have. And so as as the hedges get more expensive, your willingness to take on more risk goes down because you can't hedge out of it.
1:25:47Like it's this is all the function this is all symptoms of the debt market running out of money.
1:25:51Yeah. Well, I don't I don't know if debt markets are super deep. I think it's more of like an appetite. And of course, there's a price for everything. Hey, great. You know what? go take out 5 billion of debt at 20%.
1:26:00But you don't see this. You see this at the equity layer and like so again, how come there's hundred billion dollars of equity Yeah.
1:26:06that's going to go into Entropic and and Entropic cannot raise $100 billion or $50 billion.
1:26:11I I think um I would I don't know if I would say they can't do it. Um I don't know if well one they need to be rated.
1:26:18Um do they want to? You know, my my guess is Enthropic and OpenAI are happy that Neoclouds exist to take the balance sheet risk. Maybe they don't want to take that balance sheet risk because again, it could be things out of their control that blow them up. You have a huge debt stack in your cap in your cap table and things out of your control going like you don't you're not the one holding the bag. And again, this is like the beauty and the curse of of why Neoclouds can be
1:26:46great businesses or they could be brutal businesses if the contracts aren't airtight, if things aren't structured well, your balance sheet business, you know, and and just like fintech balance sheet businesses can eb and flow, you know, there's a there's a ton of Oops, my computer may have died.
1:27:03It's fine. We'll edit we'll edit.
1:27:05Okay. You know, balance sheet businesses, right? Like there were a ton of fintech companies in 2020 that did super well when interest rates were low that got blown up when interest rates were high and balance sheet businesses can be really tricky and very cyclical and you don't want to own a balance sheet business when a market cycle tips over. So hey on on some level I'm sure you know especially for internal chip efforts right like the openi team will might may want to develop their own data centers for their jalapeno chips right
1:27:33and anthropic same thing for their chips um but I'm sure they're also very happy for other people to take on that risk um because it's someone else's core competency and you know a lot of debt could add a lot of blowup risk um and there was no debt on on the anthropic balance sheet now they have facilities in place that they could brought down.
1:27:54And so would it surprise me if you start to see debt on their balance sheet in the future? No. Um I mean, but again, it's like you want to build a gigawatt, right? 50 60 70 billion. That's that's not that's not a tripling amount just for and that's just just one gigawatt. Um I think both open and anthropic have said they kind of want to get to five by the end of the year. Who knows if it's if it's that or something more or less than that.
1:28:18Are they about equal?
1:28:19I don't know. Okay. Um that's interesting to because I think Open I was ahead.
1:28:25Correct. Right. By by a lot but I think closed the gap a lot. Uh and open at least started like Stargate was a plan the first stage in a plan to get to 30 30 gawatt.
1:28:37Yeah. So and so I was like okay well that's I haven't heard that plan from but again like let's let's do some fun math really quick on the fly. Let's see if I can do it. Getting tired. Um, OpenAI, let's say, I think when they came out with that, it was maybe a year ago and they were what, like 1 and a half gawatt, something like that. So, whatever, round that down to zero for just the sake of the argument. Let's say Open Anthropic each want to go from 0 to 30 over 5 years. So, you're going to add 60 gawatts of AI capacity, just the two labs. What's a good assumption for their
1:29:06total capacity? I don't know, 40 50%. Um, let's say it's 40%. And I don't know, now we're getting too tough in math in my head, but you know, let's say that gets you to, I don't know, 110 120 gigawatts of total capacity brought online over the next five years. Let's say 100 just so the math is easier. And let's say it costs $50 billion to build out a gigawatt. That's $5 trillion.
1:29:28$5 trillion over the next 5 years, right? And so call it a trillion a year. You're you're talking about a nontrivial percent of GDP, you know, real GDP. That's like five% plus or minus of GDP.
1:29:47Like that is a massive bonkers, massive buildout, you know, rivals the railroad buildout. You know, I I think I've seen some charts somewhere that kind of compare those expected numbers, but like f you know, when you're measuring things in like whole number percents of GDP, uh you're that's that's huge. And so like the credit markets, [clears throat] debt markets need to be super super super deep and super super open for that
1:30:14to happen. You need global buildout. Um, you know, it's kind of funny, you know, you know, semi analysis number for next year is like 50 and, you know, I think Brad at the all-in summit said, "Oh, maybe it's lower." And I think there were some trolling going on back and forth. And I think the end result of that was like, "Hey, maybe the 50 gigawatt number, I don't want to speak for anyone, was more of like what's constructed." You know, Brad was more of what's constructed and live generating tokens because there can be a gap um in what's constructed and ready to go versus what's actually live and
1:30:42generating tokens. But I think it's like a real debate in the world, you know, and then you could see a pretty wide range of outcomes, you know, do we really add 50 gawatts of capacity next year as an industry. Um, or is it 20 or 30, you know, like who knows? I just know everyone you talk to in the supply chain is saying, you know, we we have we have orders or indications for orders that are well over 100% of our capacity.
1:31:07And like I don't have line of sight to that slowing down for the next few years. Yeah.
1:31:11And Yeah. Yeah. And and the limit like land shell power all these but also silicon like straight up totally like compute memory every every part of supply chain totally that's the work that's why 700 is so powerful right now.
1:31:22I mean look at these recent Kimmy models Kimmy K3 you know how much memory you need gigabytes of memory just to serve that you're not running that on your laptop you know you need like a rack to serve one model.
1:31:32Um you know again that's that's those are just the big models right like you're also going to have tons of small models. Um but I don't know the big models are getting bigger which only require more memory um and more compute to run.
1:31:44I used to say they were capping out. I used to say like the max size is one to two and then like we would just optimize from there. I've been wrong.
1:31:50What's interesting is it does feel like the open weight models are capping more around like a 100 billion active parameters.
1:31:56That's right. Yeah. It's mostly like an infant speed thing. Totally.
1:31:59But like I think it's actually really interestingly tied to the generations of GPUs that they have, right? like you wait for the Ver Rubins to to roll out and like whatever China has and and well then then they'll get it you know like then so they're really memory constrained at the infra layer and that actually shapes the model architecture which is like kind of cool totally.
1:32:19Um oh so that is anthropic uh powerhouse tour. Congrats. I don't know how you felt coming into this but you you did great.
1:32:27Oh this is fun. Uh, and like obviously more to do and more to analyze, but I just definitely wanted on the podcast to do this because I I know you.
1:32:34Yeah, of course. It's been fun.
1:32:36And like when else you've been meaning to do this for a while when else we going to get it? Yeah, exactly. Exactly.
1:32:40Um, any other commentary? When you don't spend time on, you know, the labs, what do you what are you spending time on?
1:32:45Ooh man, it's um again in a world that's highly unknown and unknowable and uncertain, I kind of fall back to what are my core beliefs that I do have conviction in and that that is the world's going to consume more models. World's going to consume more power.
1:32:57World's going to consume more comput, networking, and storage. And we're going to need to build a lot more data centers to serve the intelligence we're building. Those are kind of like my six core truths. And so like a lot of what I'm looking for are what are businesses that are a direct derivative of one of those one of those core beliefs that I have.
1:33:13Yeah. And any any other expressions of those core beliefs that you've uh done recently?
1:33:17Done recently. Um yeah, certainly. Um there's a chip company still in stealth. That's super exciting. Maybe we can talk about later. Um there's, you know, power company American terowatt. They want to build a new grid, an industrial grid to serve industrial offtake. Big huge bold vision. Um you know, we talked about Allorean and Glimpse and kind of the visual thinking.
1:33:39Anything on the comput side? Uh so effectively like slicing out compute uh in order to increase the efficiency trading futures trading futures on comput all I haven't I I haven't spent as much time there. Yeah I haven't spent as much time there.
1:33:53Yeah you know um we spend some time with an who's doing national comput and and AMP grid and and SF compute and all those those guys. uh in in New York we launched the comput trading track where we're we're like mostly it's just to increase utilization is the broader goal but that's how you can magically create a whole bunch more compute which is better use the ones we already have totally uh and so you know if the MFU whatever choice of utilization is at 30 to 40% right now taking up to 60% you've
1:34:22doubled the number of GPUs you have yeah totally so I mean I think there's going to be so much cool innovation it's why like this is the most exciting time to be in tech, to be investing, to be around all these startups cuz like you're just you're seeing innovation at the silicon layer, you're seeing innovation at the capital markets layer, you're seeing innovation at the product. I mean, like it's like every creative destruction everywhere.
1:34:43I notice you're not doing much on the application layer, you know. Look, I I would say the agent layer.
1:34:48No, c certainly are. Again, like I I I mean, I'm speaking more personally than necessarily on the the altimeter behalf.
1:34:55um want to believe there's a strong product research or combo. Um there's certainly a lot of great teams and we think you know early on in market cycles you know we kind of use these like triangle analogies where you have um compute or infra like like hardware infra um software infra software apps and in a mature market the biggest the biggest TAM is the apps and you know the
1:35:24smallest TAM is the the hardware um early on in cycles it's inverted and you know the biggest markets are at the at the at the well both top or bottom of the triangle more of the the physical hardware and then the infrastructure and then the apps and like eventually the triangles invert and so hey at some point in the future I think the apps and the agents will be the biggest part of the TAM we're just early on and there's lots of opportunity in the other parts of objectively that is correct yeah I I I
1:35:52do feel like and and by the way you can see various versions of this in all of computer history right down to like the uh western micro like the like when hard disks were were getting built out or whatever. Anyway, so I I I don't think those are the best examples. Western digital.
1:36:08Western digital, right?
1:36:10Um anyway, so um there uh there Yeah, I I I think that is true, but like to me domain specificity is very interesting and to me the app layer, the agent layer is supposed to capture that.
1:36:21They they're just like the biggest outcome here is cursor right now and that's 60 billion. Yeah, we're here talking about two trillion.
1:36:28Very very different.
1:36:29Yeah, totally. Totally. I think I also just like skew a little bit more infra than Yeah, you know, just personal bias. Yeah. Yeah, we we we're in for guys at heart.
1:36:36Uh okay, great. Um my closing question was going to be favorite um Brad story. I mean Brad's a character.
1:36:44Uh and you've worked with him for like I don't know three, four years now.
1:36:47Six years. Okay. So So cloud judgement ends. Okay. All right.
1:36:50Brad run around. Yeah. Yeah. I I feel like I'm getting my time either my timeline is wrong or I met you way earlier than I thought I met you.
1:36:56No, I think we met earlier cuz I I was I think I was still at Redpoint.
1:36:59Oh my god. Okay. All right. All right.
1:37:00Yeah. You Where were you? Netifi. No.
1:37:02Uh Airbite. Uh no. No. Before that.
1:37:04Before that was Temporal.
1:37:07That's when we met.
1:37:09Jesus. That's that's ancient history.
1:37:12Uh anyway, B's a character. I I just think he's he's interesting and like what's what's a good story you have?
1:37:18Oh. Oh, man. You put me on the spot.
1:37:20Yeah. Um, honestly, I would say favorite Brad's story just how much he's worked at invest America like and I don't know if people know this initiative. It is the idea of like how can we get people invested in capitalism early on in their life. I think like right now in the world right you mean you survey you know America you know most people have a negative view of capitalism because they haven't participated in it and like a lot of people view capitalism as a great thing
1:37:49and it's the people who have participated in the equity upside of America and have kind of compounded um into capitalism but most people don't have that luxury and so how can we educate people with firsthand experience about the benefits of capitalism by helping them participate early on the whole idea is an investment account funded by the US government at birth.
1:38:09That can be funded by anyone along the way. That could be philanthropists like like Michael Dell um who made a $6 billion uh donation. I think Elon Musk has pledged one share to every kid in America which is you I think $10 billion and has some version of like the public benefit fund or something.
1:38:26Exactly. And so I I think just seeing him work on that behind the scenes for years um and kind of seeing it come to fruition all very quickly. You know, it's the classic overnight success. And so there's not like one specific story. It's more of just seeing the grind that goes on. Yeah. Exactly.
1:38:43And is it done by the way? I actually don't know that.
1:38:45So it's passed. Yes, it's passed in the law.
1:38:46It's passed. So So what? Like if I have a kid now, they get you open an account. Yeah. Yeah. You open an account. Um any new kid born gets $1,000 of birth. Um parents can contribute to it. Employers can contribute to it. Um friends can It's like it's like a baby 401k.
1:39:00Exactly. [laughter] Exactly. And you know, you have people like Michael Dell who say, "Hey, here's my donation, and I want it to be gifted to I can't remember the exact specifics, but I think it was like, hey, bottom 75% uh income by zip code." And the money just That's better. Yeah.
1:39:15So, Exactly. The money just cuz, you know, as Brad would say, look, what's 529 plans, which are kind of education plans, like, hey, I can fund money that can be used to for my kids education one day. Like the reality is is the people using 529 plans are generally like the wealthiest part of the country. Like 529 plans and the tax advantage wasn't made for the wealthy part of the country, but that's who uses it the most. Uh what he wanted to do was, hey, let me create investment accounts for the people who
1:39:44actually need them. Um and I think it's, you know, it's an awesome initiative that's just getting started. I think you're going to see a lot more direct philanthropy. What's a big issue people have with philanthropy is well you're going to donate a bunch of money and then middlemen are going to take half of it.
1:39:58I have seen literally directly this tons of loss. It's used for corruption like Exactly. But what's better direct contributions direct to bank account of people who need it most. No middleman direct accounts that then people can see appreciate in value over time. Can't touch it till you're 18 years old. Um and so you get to see the appreciation happen over time. maybe learn and and participate in in the upside of America along the way. Um, and I think it's it's it's really like it it's caught on and
1:40:26just seeing how passionate he is about it. Um, has been kind of inspiring to see.
1:40:32Okay. My one push back is like I'm, you know, homeless on the streets.
1:40:37Food stamps, everything.
1:40:39I have $1,000 sitting there.
1:40:42You then like Well, give it to me.
1:40:44Yeah. So, yeah. you. [laughter] Well, it's made it's made more for kids now and so hopefully you're not well there there are plenty of homeless kids.
1:40:52Exactly. Right. Like so so like that that's the thing like it's like well you know did you know you have a bank account and you're participating growth for capitalism.
1:40:58That is the biggest thing.
1:40:59I don't care you know like so that that's the biggest thing now which is um people need to claim their accounts. The government opened up accounts for every kid in America and parents need to claim them. It it is sitting there now. The money is sitting there for newborns. Um, you know, if you're 15 years old, money doesn't show up in your account. I think donations will. And so, you could be a 15-year-old that does have a $250 donation for Michael Dell sitting in your account.
1:41:23Um, but they need to be claimed. And so, like now the I would say the focus has gone from let's get this passed into law. Let's get it up and running. Let's get the infrastructure set up so people can donate.
1:41:35Now, it's let's go convert those created account to claimed account. And that's just a that's like a grassroots word of mouth game because there are people whose accounts are sitting there with money in it and it just claim Yeah.
1:41:48Yeah. By the way, most people have money sitting in the government that has not been unclaimed. You can look it up.
1:41:53Yeah. Unclaimed property. It's like a great use case for like have an agent go look up your name.
1:41:59Uh you probably have a few thousand dollar sitting somewhere that someone returned to you and it's just you just need to know your address, your prior addresses and then go have instinct or muse. highest ROI on this podcast is that not not the anthropic.
1:42:09Exa Exactly. Exactly.
1:42:11Um, cool. Thank you so much for spending your time with us. This is great. Uh, we'll have you back. Uh, we'll see how the IPO goes.
1:42:18This was fun. Thanks for doing it.
