The entire value of AI right now is being realized by writing software. Agents are just quickly spun-up applications, and as enterprises discover they have spun up too many inefficient agents, they will realize strong software engineering capability and a strong IDE are required to fix the mess, making Cursor IDE position extremely valuable.

(00:00:00) Jason, you are the unique person that is at the intersection of both the and the SPLC files. (00:00:07) Do you have a comment? (00:00:08) Do you have a comment? (00:00:10) No, I'm not in the SPLC files. (00:00:12) Yes, you are. (00:00:12) You're adjacent to the vice file. (00:00:13) SPLC adjacent and you're. (00:00:16) What does that mean in the Venn diagram? (00:00:17) I thank you, though, for putting me in the crosshairs of all the... (00:00:21) I got a really good way to select. (00:00:23) There's a reason why I'm carrying this guys is because the. (00:00:27) What the fuck is going on? (00:00:30) There's a reason why I carry a stiletto at a P-35. (00:00:33) What the fuck are you doing? (00:00:36) There's a reason. (00:00:36) If you want to jump the feds, feel free. (00:00:40) J-Pal is ready. (00:00:48) Let your winners ride. (00:00:50) Rain Man David Saxon says. (00:00:55) We open sourced it to the fans and they've just gone crazy with it. (00:01:03) All right, everybody, welcome back to the greatest podcast in the universe, episode 270 of the All In Podcast, your podcaster's favorite podcast. (00:01:12) With me again, your Sultan of Science, David Freberg, the dictator, Chamath Palihapitiya, and yeah. (00:01:21) The Rain Man is back. (00:01:22) Yeah, it's definitely David. (00:01:24) David Sacks. (00:01:25) He's definitely in DC with POTUS. (00:01:28) Yeah, POTUS lets him drive in the driveway. (00:01:31) Sacks, what's going on? (00:01:32) You pushed back, you big shotted the entire crew and pushed the show back an hour. (00:01:37) Simple text. (00:01:38) He's like, with POTUS. (00:01:39) Start it. (00:01:40) It's unbelievable. (00:01:41) Start later. (00:01:42) OK, we'll just wait. (00:01:44) OK, Daddy. (00:01:45) Look at him. (00:01:46) All right. (00:01:47) big shot. (00:01:47) What's going on? (00:01:48) No, look, I was in DC today and I was at the White House and I just asked if the president had time and he made time. (00:01:56) And we did have a little meeting. (00:01:58) And so we did push back the pod for that. (00:02:00) One thing I just want to say is just (00:02:03) What a pleasure he has to deal with. (00:02:05) when I read in the media, they're always describing him in a certain way that, he's yelling at people or he's moody or something like that. (00:02:13) And that's never, ever been my experience with him. (00:02:15) He's always pleasant to be with. (00:02:18) He's always genial. (00:02:19) He asks questions. (00:02:20) He's interested in the subject matter. (00:02:22) It's just a completely different portrayal. (00:02:25) I don't get where the media is coming from at all on this. (00:02:27) He's charming AF. (00:02:29) Let's just call it what it is. (00:02:31) I mean, maybe if you double-crossed him, maybe, I don't know, but I've just never seen any evidence of how they describe him at all. (00:02:38) And I think on our issues of AI, I think we're really lucky that he's the president who's in the White House when this AI revolution is happening. (00:02:45) I mean, doing old history, Sax, what would happen if Kamala Ding Dong was in right now and we'd have like no data centers? (00:02:52) would have no data centers, and they'd be using AI to censor us, and they'd be promoting DEI values through AI. (00:02:58) That was in the Biden executive order. (00:03:00) President Trump just wants the country to win and be successful, and he doesn't have these like doomer neuroses about it. (00:03:08) That's not to say we don't support any regulation at all, but we should have specific solutions for specific problems, as opposed to being cowering in fear over this and just trying to halt all progress. (00:03:20) And I think a really good example of that was his idea around data centers, where he said over a year ago, before data centers even became a hot political topic, that we should let our AI companies stand up their own power generation behind the meter. (00:03:32) And that's a much better approach than the Bernie Sanders approach of just shutting everything down. (00:03:36) So I don't know. (00:03:36) I think we're very fortunate that he's the president during this critical time and developing this technology. (00:03:41) And like I said, he's always been interested in it. (00:03:44) He talks to a lot of business leaders. (00:03:46) I'm always actually very impressed with what he already knows. (00:03:49) He listens. (00:03:50) It's like all the top guys in the industry and he synthesizes what he hears. (00:03:53) I think he's very good at that. (00:03:54) He was talking about the Anthropic guys and he was like, these are brilliant guys. (00:03:57) And he was like giving the flowers to them and how genius they were and that they were working on a deal. (00:04:02) Any insights there about the relationship between the White House and Anthropic? (00:04:08) I thought what he said was very balanced and accurate. (00:04:11) Like you said, he said that they were very smart guys. (00:04:14) They do have a great product. (00:04:15) I've certainly acknowledged that. (00:04:17) He also said that they were very left wing, but that was something we could work through. (00:04:22) Didn't have to be a deal killer. (00:04:24) He said they tried to tell the Pentagon what to do, which the Pentagon didn't like. (00:04:27) But in any event, I mean, look, he wants American companies to be successful. (00:04:31) And he, I think, genuinely really does like high IQ people. (00:04:36) I mean, he says it all the time and people think he's joking, but I actually think it's like one of his core convictions is he just (00:04:41) He really likes smart people. (00:04:42) He likes being around smart people. (00:04:44) Loyal people, smart people, people who are good on cameras seem to be the three circles. (00:04:49) And hey, Sacks, you fall into two of the three. (00:04:52) All right. (00:04:53) Let the audience figure that out. (00:04:54) Topic one, SpaceX has signed a huge deal with Cursor. (00:05:00) You know Cursor, that's the AI coding startup. (00:05:02) Really, they define the category XAI and Cursor are building and collaborating on a new AI coding model that would, quote, (00:05:11) the world's best coding and knowledge work AI. (00:05:14) Here's the deal as it's been explained. (00:05:16) SpaceX will either buy Cursor by the end of 2026 for $60 billion, that's $10 billion more than they were rumored to be raising at, or they will pay Cursor $10 billion for their collaboration together. (00:05:29) Bloomberg says you can think of that $10 billion essentially as a breakup fee. (00:05:34) So I think it's fait accompli that this deal is going to get done. (00:05:38) Cursor's run rate, 2 billion at the end of February. (00:05:41) This is a money printing machine. (00:05:42) They expect to end 2026 with a $6 billion run rate. (00:05:46) They're going to triple it. (00:05:47) SpaceX projected revenue between 22 and 24 billion in 2026. (00:05:53) So this is quite accretive to the revenue story at SpaceX, at the IPO of SpaceX, which is now targeting A valuation of 2 trillion, which would be trading at roughly 80 times top line revenue, which is a, (00:06:06) people would say it's a high valuation, but also a commensurate with the opportunity. (00:06:10) Cursor's valuation would be 30x. (00:06:12) So this is a good deal, I think, for everybody at the end of the day. (00:06:16) Cursor started, I think, built off of Anthropic's LLM. (00:06:21) You could use any LLM previously on it, but in March, Cursor released the second version of their proprietary model, Composer 2. (00:06:29) And here it is. (00:06:30) It's ranked pretty high right now. (00:06:32) It's between GPT-5.4 and Opus 4.6. (00:06:36) as you can see on the screen. (00:06:38) The key part of the story here is that Elon has 550,000 GPUs in Colossus. (00:06:45) He's scaling up to 1 million. (00:06:47) And then, of course, he's going to bring it to space. (00:06:49) So if you believe that infrastructure matters, and it's pretty clear it does, this is incredible for Cursor, who has been compute constrained. (00:06:58) So this is peanut butter and chocolate. (00:07:00) If you put these two together, I predict that this is going to move (00:07:06) SpaceX, X, XAI and cursor to the front of the coding leaderboard within 12 months. (00:07:12) That's my prediction. (00:07:14) Chamath, shareholder in SpaceX via the acquisition of the Starlink company that you were a backer of. (00:07:22) What are your thoughts? (00:07:23) The acquisition was essentially negotiated, and the way that it's structured is so that the S1 doesn't go stale. (00:07:30) So I think the way that it was (00:07:32) announced has more to do with the fact that they don't want to slow down and have to rewrite parts of the S1, have to redo the disclosures, have to redo the risks. (00:07:43) And so I think what you're going to see is that this will get done. (00:07:47) In fact, the deal is effectively done. (00:07:49) But what's so smart is that where is SpaceX today? (00:07:52) Let's call it a trillion. (00:07:56) Where could it be, just for the purpose of this argument, let's say 2 trillion. (00:08:01) So when the deal gets done on a stock for stock basis, it's going to be, if again, if it's 60 billion in tomorrow dollars, effectively Elon's gotten a 50% discount. (00:08:12) And what has he bought? (00:08:13) He can issue $60 billion of stock at a $2 trillion valuation and get a model and a service that I think is extremely compelling in coding, which is where we know all of the immediate and short-term revenue gains are. (00:08:27) It's also patterns that are hard fought (00:08:29) and are really valuable in reinforcement learning. (00:08:32) He gets all of that. (00:08:33) And then he gets a very crack team, which we've known for a while that the Cursor team is absolutely excellent. (00:08:39) If you look at the Grok usage, it shows why he had this excess capacity. (00:08:44) There was a moment where Grok had a very steep and very aggressive discount on their output tokens. (00:08:50) And in that moment, there was just a lot of experimentation and usage. (00:08:54) And over time, that sort of (00:08:57) went away. (00:08:58) So there was a lot of capacity and relatively low utilization, I think, inside of Colossus that he was able to turn around, jiu-jitsu move the whole thing, and basically acquire the most interesting and valuable third-party wrapper service in AI right now. (00:09:14) And the fact is that they got it effectively, I think, at this price for 30 billion. (00:09:20) So I think it was a really good deal, really smart deal. (00:09:22) Sacks your thoughts if you want to unpack it a bit. (00:09:25) Under the framing, I think it'd be interesting for you. (00:09:28) If we were sitting here three years ago, the Biden administration didn't invite Elon to the EV summit and the SEC and other organizations, Delaware, they were explicitly involved in lawfare. (00:09:43) They were trying to put Elon in prison. (00:09:44) And here we are, the most important company in history. (00:09:48) of the United States, SpaceXXX AI and Tesla now on the verge of just creating the greatest products in the history of humanity between SpaceX, clusters in space and Optimus. (00:10:01) Your thoughts? (00:10:02) Well, you're right. (00:10:02) I do remember a press conference where Biden said, we got to look at the sky. (00:10:06) And so on the heels of that, the DOJ brought a lawsuit (00:10:10) Yeah, exactly. (00:10:12) SpaceX, which they can't under ITAR. (00:10:17) They couldn't exactly under ITAR. (00:10:18) Anyway, that's all ancient history. (00:10:20) So let's put that behind us. (00:10:23) Look, I agree with your guys' analysis on this. (00:10:26) I think these two companies are very complementary. (00:10:28) Cursor obviously is very strong in coding. (00:10:32) That's what it brings to XAI. (00:10:35) XAI brings compute and they bring (00:10:38) a foundation model. (00:10:39) And the problem that Cursor had is that even though coding is kind of like the white hot area of AI right now, when it got started, it was really competing against generalists in the form of OpenAI and Anthropic. (00:10:52) But now those generalists have decided to vertically integrate in this area of coding, right? (00:10:57) And so Cursor's now competing against Cloud Code and OpenAI's Codex. (00:11:02) And so they were dependent on foundation model companies that were getting in the business of competing with them, which was just not, (00:11:08) a good place to be, right? (00:11:09) So now they have this new alliance with a different foundation model company, which also brings the compute. (00:11:14) It just makes a lot of sense. (00:11:15) And then they bring, Cursor brings to XAI the training data, a lot of enterprise clients and the experience encoding. (00:11:25) And I think this will accelerate XAI in this area. (00:11:28) Sax, you think they're going to dump KimiK2.6? (00:11:33) Because I think Cursor Composer 2 uses the Moonshot model. (00:11:37) There's no reasonable way that Elon's going to pay $60 billion and not run on top of Grog. (00:11:41) I got to think. (00:11:41) It seems like it, likely, but I don't know. (00:11:44) I think it might be tough depending on the users. (00:11:46) One of the things that makes Cursor so good is... (00:11:49) Wait, say more on that. (00:11:49) What do you mean? (00:11:50) So, I think that the different developers... (00:11:54) want to have choice in that sense. (00:11:55) There's A toggle. (00:11:56) So one of the things that's really good about Cursor is they've got this very well built out IDE, this application layer that puts them probably from a UX perspective, meaning developers are using the tool above Codex, above Claude, above anything else. (00:12:12) You can use a third-party IDE and integrate the models or integrate whatever other third-party service you're using. (00:12:19) But I would imagine that the developers are going to want to (00:12:23) continue to have at least some choice on what's actually writing the code for them. (00:12:27) The thing that people are waking up to in the last 120 days is just how much of the value of AI is being realized by writing software. (00:12:38) And we've kind of got this wrapper term, we call it agents, but agents are fundamentally just quickly spun up applications. (00:12:45) But for all of them, as we're realizing very quickly, you end up making too many agents, they end up being super inefficient, they need to be engineered, and you still need to have a strong software engineering capability and competency to fix all the agents, to build all the harnesses, to make everything work well together. (00:13:01) And that's why having a strong developer environment, a strong IDE, actually solves that biggest problem. (00:13:06) So eventually all the enterprises that are getting hot and heavy on agents are going to be like, whoa, wait a second, we've actually got to fix how this is all being done, as we saw this week in that story with Amazon. (00:13:15) where there's like a million agents being spun up inside and everything's wasting resources, redundant data creation, redundant data stores, redundant API calls, et cetera, tons of money being wasted. (00:13:26) So you have to centralize still. (00:13:28) You have to have good software engineering talent that's making good infrastructure and good use of these agents. (00:13:33) And that ultimately will require an integration of the AI tooling with a standard software engineering front end, which is the IDE that Cursor has. (00:13:41) So I think that that's probably where everyone's waking up to the fact that (00:13:45) Having the software engineers may end up winning you the arms race here, and it seems pretty smart for Elon to buy Cursor. (00:13:53) One other piece of it, you mentioned Kimi K2.6. (00:13:56) 2.6. (00:13:56) Yeah, I mean, so I think that one of the things that's going to become a priority over the next several months is this idea of optimizing, because enterprises token bills are going through the roof right now. (00:14:07) I mean, (00:14:09) Because month over month, they're spending increasingly large amounts because their employees are just building more and more software. (00:14:15) But I'm not sure that anyone's been incentivized yet to be efficient about it. (00:14:19) And it really only makes sense to go to a frontier model for a frontier task. (00:14:24) But more mundane things could be done using an open source model or a less expensive model. (00:14:29) And I think, like you're saying, whether it's the ID or something else, there needs to be some sort of (00:14:34) middleware that determines which model you go to and how much you're willing to spend and what the most efficient way of getting the tokens is going to be. (00:14:42) I am deep in playing with XAI's suite of products, and I would predict (00:14:48) We're going to be sitting here in 6 to 12 months, and they are going to be dramatically, dramatically improved. (00:14:54) Let me just flag one other area that I think is maybe the white hot center within this red hot area of coding, which is cyber. (00:15:03) And I think Mythos has kind of woken everybody up to the potential of frontier models to be a weapon that can be used by either cyber offense or cyber defense. (00:15:16) Now, the issue with Mythos is that it's very large and expensive. (00:15:19) It's something like a 10 trillion perimeter model. (00:15:22) And there's a lot of reports that Anthropic just doesn't have enough compute to be able to serve it. (00:15:25) I'm not sure it was ever built to be a commercial model, to be honest, because I just think it's so big and expensive. (00:15:31) But I think what will happen is these companies will start training dedicated cyber models. (00:15:37) Let's say Mythos comparable models, but with a lower token cost. (00:15:41) And (00:15:42) I think there's a real race on right now to get those products to market because I think IT departments and CISOs are very worried about the risk of hacks right now, AI-powered hackers. (00:15:54) So this is something I think over the next three to six months will be, again, maybe the hottest part of the market. (00:16:00) Polymarket says all of this is fate a complete. (00:16:04) SpaceX acquiring cursor, 74% chance, SpaceX IPO. (00:16:09) By the end of August, 80% chance. (00:16:11) So this is happening, folks. (00:16:14) All right, let's keep going. (00:16:15) By the way, I think that deal structure is smart because, I mean, to Tamas' point, yeah, it prevents the IPO process from being disrupted. (00:16:21) Also, it kind of gives a huge motivation to these cursor guys to bust their ass and make it work over the next, I don't know, six months. (00:16:28) Yeah, they have a $10 billion breakup fee, but I'm sure they want the deal to be successful. (00:16:31) Well, the $10 billion breakup fee will go back to SpaceX anyways, because if they actually run the compute and they're not owned by SpaceX, they're going to have to pay for it. (00:16:40) That is not cheap. (00:16:41) I mean, we saw a bunch of these XAI co-founders leave after the acquisition by SpaceX. (00:16:46) I don't know if that was the reason why, but all of a sudden they're sitting on SpaceX stock and they may have felt like (00:16:53) They had it made, which is always a problem. (00:16:56) It's always a problem with M&A. (00:16:58) This cursor thing came about pretty quickly because, let's just say friends of ours who were supposed to wire into that round were like, where's the wiring instructions? (00:17:08) It all just evaporated. (00:17:10) Here's a tweet from Elon. (00:17:11) We don't have to speculate too much here, Sachs. (00:17:14) He was very clear that XAI wasn't (00:17:18) built right the first time around. (00:17:20) The quote, XAI was not built right. (00:17:22) First time around, so is being rebuilt from the foundations up. (00:17:26) Same thing happened with Tesla. (00:17:28) And that tweet is from about 5 weeks ago. (00:17:31) How crazy is it that when he tweets, he gets 50.8 million views? (00:17:36) It takes the four of us seven months to get 50. (00:17:39) I mean, that's probably our collective. (00:17:42) It's unbelievable the distribution he has. (00:17:44) Well, also, how many CEOs would just fess up like that and say, yeah, we didn't do it right the first time. (00:17:49) Now we're rebuilding it. (00:17:50) I mean, most of them are not willing to say that. (00:17:52) He's a magnet for talent. (00:17:53) He's a magnet for the right kind of talent. (00:17:55) And the SpaceX talent has his philosophy. (00:17:58) He inherited, I think, a lot of (00:18:01) maybe people for XAI or for Twitter that were not in his mold and they're clearly getting aligned. (00:18:09) And it's also going to make his day-to-day life much easier when all of these things are occurring in the same building with the same team. (00:18:15) The continuity of not having to task switch between companies is going to be great. (00:18:19) We talked a little bit about the possibility of Tesla and SpaceX merging. (00:18:23) Even Walter Isaacson now is on the Tesla-SpaceX merger train. (00:18:28) There you go. (00:18:28) He just did a pod. (00:18:29) Everybody's confirmed it. (00:18:30) It's going to happen. (00:18:31) We called it here first. (00:18:32) Okay, topic two, is there a SAS debt bomb in private equity? (00:18:38) Thoma Bravo, we had Orlando Bravo at the 4th All-In Summit last year, is nearing a deal to hand its portfolio company, Medallia, over to its creditors. (00:18:50) This is a SAS for customer experience company. (00:18:53) TB acquired them in 2021 for $6.4 billion, all cash at the top of the market. (00:19:01) As part of the deal, they incurred 3 billion in debt. (00:19:04) And for background in 2021, this company had 470 million in revenue growing 20% a year. (00:19:11) Earlier this month, Bloomberg reported that TB's debt servicing costs for Medallia were about to triple from 100 million a year to 300 million a year. (00:19:21) Blackstone and other firms refused to extend a lifeline to the company, to the SaaS company. (00:19:28) So it looks like Thoma Bravo just handed the keys back and wiped out $5.1 billion in equity. (00:19:33) Chamath, your thoughts. (00:19:34) We've been talking about the SAS headwinds for a bit. (00:19:36) You've been quite vocal about it. (00:19:39) Well, first of all, I think Thoma Bravo is an unbelievably well-run organization. (00:19:44) Their returns are bonkers, and Orlando is a really, really, really good investor. (00:19:50) So what do I think happened? (00:19:53) I suspect that they probably got enough of their equity, if not all of their equity. (00:19:59) There's probably... (00:20:00) probably a decent chance that they did at least one or two dividend recaps in the last five years. (00:20:05) And if I had to guess, I suspect that they are positive return. (00:20:10) It may not be the return that they would want. (00:20:13) And so turning the keys over becomes easier. (00:20:15) Because you have to remember, in private equity, the entire playbook is for transformations of assets that are at some point not working. (00:20:25) right? (00:20:25) It's very rarely that they're buying the same kinds of businesses that the four of us would buy, which is just sort of this, clean white sheet, de novo, grow at all costs kind of business. (00:20:35) So they have operating partners and all of these other people waiting in the wings to unf**k situations. (00:20:42) That's the whole playbook. (00:20:44) So to turn it over, I suspect means that there is a core rot that people couldn't fix combined with (00:20:52) the fact that they have probably gotten enough downside protection that it's not a huge thing for them. (00:20:58) Now, this is an issue for the bond holders, and then that'll maybe flow through to the borrowing cost that Thoma Bravo has to pay maybe for a subsequent deal. (00:21:06) I don't know, but I doubt that they would just walk away from a business. (00:21:11) So I suspect they probably got most of their money out. (00:21:13) I don't know if that's true. (00:21:14) There was someone that published some internal data showing that the sales team was like 18% of target at Medallia. (00:21:22) You guys know what this company does, Adalia? (00:21:25) Customer support is the general arena and customer experience. (00:21:28) Customer experience management. (00:21:30) I don't know what that means. (00:21:32) Yeah, they'll basically send, like you go on Caribbean cruise ships and you get a survey afterwards, and then they use that survey data to provide management insights and operational insights to the leadership team and the operating team on how to improve the quality of their product or their service. (00:21:45) So it's sort of like this feedback surveying loop. (00:21:48) So if I were to tell you guys, hey, you want to (00:21:50) build a feedback surveying loop to run your business better, are you going to buy SAS today or are you going to ask your AI to spin up an agent for you to do that? (00:21:58) And I think that's a big part of what's happened is all these sorts of companies where the alternative to buying a SAS product is to spin something up internally. (00:22:07) and it's much cheaper and easier to spin it up internally. (00:22:09) You get a custom workflow. (00:22:10) No, I agree with that. (00:22:11) I'm just saying in the last five years, you think they sat on their hands and didn't take a dollar out? (00:22:16) They're not that dumb. (00:22:17) Maybe they took cash out, maybe they didn't, but there was still a big debt overhang and the debt's clearly (00:22:25) The debt holders are clearly screwed here. (00:22:28) The question is, Doma Bravo screwed? (00:22:29) And I would say if you sat around for five years, that's not their style. (00:22:33) They generate too much money. (00:22:34) They're too good. (00:22:35) So they may have taken cash out and covered some of their costs, but the equity got fully impaired and then the debt is clearly impaired because you can see how the debt and (00:22:43) the CLOs are trading, which indicates that this business is just not doing well. (00:22:47) And then someone else on Twitter posted some internal information from Medalia saying the sales team is just not hitting their targets. (00:22:54) They're like way, way off their sales targets, which I think speaks to the underlying problem here. (00:23:00) Yes. (00:23:02) So the underlying problem is that these businesses in the SaaS space, where you're driven by net new sales every year, how many new customers are you signing up? (00:23:10) And then you're trying to manage retention and you're trying to increase sell-through and retain customers. (00:23:16) They're just having a really hard time sourcing new customers and there's probably higher than modeled attrition. (00:23:20) That's right. (00:23:20) And when you have a very kind of typically historically predictable business where you can say, hey, I've got a net revenue retention of 118% or what have you, meaning I'm selling into my install base by 18% over what I'm making last year and then I'm signing up new customers. (00:23:36) You can lever that business, right? (00:23:37) You can borrow money against those cash flows because it becomes predictable. (00:23:41) And what's happened in the last year in particular is agents have become so good and so fast and so cheap that many enterprises can simply spin up an alternative to a vertical SaaS solution. (00:23:52) And that's crushing the sales team's ability to sell in. (00:23:55) That's who you're competing against. (00:23:56) Now, I want to make one point and just link this with something else that happened this week. (00:24:00) And that's Kevin Warsh's hearing for Fed Reserve Chair. (00:24:04) Kevin Warsh went and talked a lot about the deflationary effect of AI. (00:24:08) And I actually think we all talk about the SAS apocalypse as if it's this sort of like isolated business phenomenon where these SAS companies are getting blown up. (00:24:17) I think another lens to look at what's going on is the incredible deflation of how much it costs to successfully run a business (00:24:25) and you don't have to pay a premium price for SaaS products anymore. (00:24:29) Meaning that piece of the business can suddenly get much cheaper. (00:24:33) That AI is delivering on its deflationary promise. (00:24:36) I'll just say one thing about what Warz said. (00:24:40) Warz spoke a lot about the deflationary evolution promised by AI and that he expects that it will drive productivity growth like we've never seen before. (00:24:47) But he said, I don't know what that's going to do to the job market, that there may be a dislocation between that productivity growth being realized (00:24:54) and how the labor markets are going to be able to respond to those things. (00:24:58) But fundamentally, he's saying that we're going to see economic deflation. (00:25:03) The problem with economic deflation is that when it occurs, it means some business is seeing their revenue go down. (00:25:10) And if that segment of the economy is levered, if they have debt sitting on top of that piece of the economy where it's supposed to always, always, always grow, like a SAS company's top line is always supposed to grow, suddenly that debt gets impaired and that can have an economic ripple effect (00:25:24) that is adverse. (00:25:25) But what he's pointing out is that as a result of deflation, because it's not coming from some cost cutting or economic contraction, what he's saying is that the deflationary forces ultimately lead to economic expansion because other parts of the economy will now grow. (00:25:40) So if I can suddenly cut, you know, call it 50% of my SAS budget, and I can reinvest that capital in other ways of growing my business instead of managing my expenses, (00:25:50) all of a sudden, my enterprise will grow and the economy will grow. (00:25:54) He also said, just as an aside, and I want to make sure I cover this so that we're really clear, he said, the way that we've been measuring inflation is wrong. (00:26:02) And that he doesn't agree with the way the Fed has been measuring inflation because you can do a survey of any household and they'll tell you, my God, everything's so expensive. (00:26:09) So all of the indices and bullshit that are being used to calculate an inflation index is completely misrepresenting what the average American is actually feeling. (00:26:17) And so he wants to rethink how the Fed is addressing inflation from an interest rate perspective, but he does think that the overall kind of economic picture is one of deflationary pressure and productivity gains coming out of AI. (00:26:28) Sachs, I'll drop this off to you. (00:26:30) I think it's pretty clear what's happening here is that the loss (00:26:34) SAS's loss is the token dealer's gain, right? (00:26:37) And startups are now, and we always see they're the tip of the spear. (00:26:42) They're writing their own tools. (00:26:43) They're making their own dashboards. (00:26:45) I see that every day. (00:26:46) And if you look at the SAS product index, here it is. (00:26:50) Salesforce down 32% in the past six months. (00:26:54) Shout out to Bestie Benioff, best guest we've had on Sacks at the summit. (00:26:58) Service now down 54%, Snowflake down 43%, Adobe down 33%, Figma, which (00:27:03) had a huge IPO pop and is now down 67%. (00:27:07) So what is the role of venture capital and then private equity in addressing the software market? (00:27:16) Software was eating the world. (00:27:17) Now tokens are eating the SaaS business and the software business, yeah? (00:27:22) Well, I'm of two minds about this. (00:27:23) I'm going to talk about the opportunity for private equity. (00:27:27) Let me just say, backing up, that historically, we only had two good exits for software businesses. (00:27:33) One was IPO, the other was M&A. (00:27:36) And then these big private equity shops came along and gave us a third potential exit, which is you would sell to them, and then they would raise the capital based on, I don't know, one-third equity and two-thirds debt. (00:27:46) So it was debt-financed buyouts, which is something that's been around in, let's call it the non-tech part of the economy for a long time, but was a relatively new entrant into the world of technology. (00:27:57) And the reason for that is that if you're going to debt finance a purchase, you need to have very stable cash flows. (00:28:02) Because if you miss, if your cash flows miss and you can't pay your interest on the debt, then you're going to lose all your equity because the debt holders will foreclose. (00:28:11) So in order to do a debt financing of any kind, you have to have very predictable cash flows. (00:28:15) And it was believed for a long time that software did have those predictable cash flows, at least for the mature businesses, the ones that (00:28:24) were at the stage where they could IPO as a potential alternative. (00:28:28) So it was a very attractive thing. (00:28:29) Like I said, I think it was great to have that third option. (00:28:32) I'm of two minds about where the private equity business is today. (00:28:36) On the one hand, (00:28:38) The pricing now has got to be super attractive for them. (00:28:41) I mean, we're seeing public SaaS companies that are doing a billion of ARR, with 20% growth rates, 80% gross margins, and they're trading at three times ARR. (00:28:53) You can buy a dollar for $0.50. (00:28:56) So is that an opportunity, Sachs? (00:28:58) Do you think that's what I think we can hit rock bottom, we should do a roll-up? (00:29:02) on the one hand, I do think that the pricing has never been more attractive if you're a private equity shop looking at a business like that. (00:29:09) I mean, those companies used to be valued at 13 times ARR, now it's three. (00:29:13) I'm talking about like a category leader. (00:29:15) Now, the downside of that... (00:29:17) By the way, Salesforce is off 9% today. (00:29:19) I don't know if you guys saw this, but the market's absolutely tanking today after the Medallia announcement came up. (00:29:26) Right, okay, so that would be like, you know, I said I'm of two minds about it. (00:29:29) So (00:29:30) I would be bullish for private equity just based on pricing, but the bearish part is that in order for their business model to work, you have to have predictable cash flows. (00:29:38) You can't have a SaaS company go from, I don't know, 120% net dollar retention one quarter to 80% net dollar retention six months or a year later because a big part of their customer base is attrited to using tokens, right? (00:29:52) Or to basically creating some bespoke software. (00:29:54) You just said the absolute critical thing in all of this, which is you have to have predictable cash flows. (00:29:59) I think what happens is when you're a startup, you typically have to figure out how to disruptively price to enter the market. (00:30:07) So you're like, okay, if I deliver $10 of value, I'm going to charge a dollar. (00:30:11) And that's the normal playbook, like a 10% ratio, right, of price to value. (00:30:16) The problem is when you start to stack venture capital into it, and then you stack growth equity into it, what you're effectively creating in the preference stack of your company (00:30:28) is that you are creating a higher return hurdle, right? (00:30:32) You got to clear 300 million, 500 million, a billion of pref, and then you have to return 15 or 20% on top of that. (00:30:39) So what do people do as they raise more money? (00:30:42) They increase price. (00:30:45) But the problem is at some point when you increase price, you engender a ton of competition and you put a huge target on your back. (00:30:52) Private equity is the last stop (00:30:55) Because when they come in and they layer in billions and billions of dollars of not just equity, but also debt, and that has to then be completely predictable and paid back, their only lever is to raise price. (00:31:08) They can never cut price to take share. (00:31:09) They can't underwrite that to pay back their debt holders. (00:31:13) And so Sachs, part of the big problem here and why nobody wants to touch these companies is that they are overpriced. (00:31:19) Yes, they're making a billion dollars of ARR, but the unit cost has gotten out of control. (00:31:24) It used (00:31:25) to be 10% of value, it's probably now 30% of value. (00:31:29) And everybody's looking at their contracts thinking, well, when it comes time to a renewal, I'm going to just cut this in half, or I'm going to cut this by two thirds, or I'm going to cut this by 75%, because the value isn't there anymore. (00:31:42) Or they can threaten to and negotiate a better deal. (00:31:45) And it becomes even worse, because the minute you make these products headless, right, and you say, I'm just going to communicate with these products via MCP and with agents, (00:31:54) You can't charge on a per seat basis. (00:31:56) What do you do then? (00:31:57) Freeburg doesn't need 50 seats of, you know, Workday. (00:32:02) He needs two seats because the agents act as the way to write in and out of Workday. (00:32:07) So he wants to pay for two seats, not 50. (00:32:09) And then if you multiply that by a million companies, that's what gets us to this place where it just feels like a falling knife. (00:32:17) And I think it comes down to these unit costs. (00:32:19) The unit costs and the price to value of these products are out of whack with what the market needs and wants. (00:32:25) And until they reset that, or you find new products that can do it cheaper, we're not going to get a cleansing and a clearing here. (00:32:32) By the way, Salesforce today is down 9%. (00:32:35) 140 billion enterprise value on 15 billion of free cash flow. (00:32:40) This thing is trading at less than 10 times free cash flow. (00:32:43) It's unbelievable. (00:32:44) I think it might be a bargain, to be honest. (00:32:46) Yeah. (00:32:47) It sounds like bargain hunting. (00:32:48) And what if Benioff was the king of acquisitions? (00:32:51) What if he just starts cleaning up and buying? (00:32:54) He's been buying his own stock. (00:32:55) Yeah, and we didn't put this on the docket, but did you guys see his kind of headless product announcement? (00:33:00) Yeah. (00:33:00) Do you see this? (00:33:01) It's actually very. (00:33:03) It's very smart. (00:33:04) Yeah, I mean, I think it's very smart. (00:33:05) There's ways that business can maneuver, right? (00:33:08) And I think they're pretty unique. (00:33:09) It may be that of all the businesses in the scape, like that, the ones that have that scale, that have that multi-product platform, that have a lot of your data, there's a lot of opportunity for them to, you know, maneuver their way into an evolution of the business. (00:33:25) And because like if you look at it and you compare it, for example, to other companies, I think the Workday response was to say, you can't have an AI interact with us without paying some kind of like toll. (00:33:35) You're exactly right. (00:33:35) That's the exact opposite, which is he's like, okay, we're going to go headless for the whole thing, which is brilliant. (00:33:40) You're exactly right. (00:33:41) You're exactly right. (00:33:42) I think that's going to be the distinction of the winners here and the losers. (00:33:45) And are you on the wrong side of this? (00:33:47) The problem is that we have to figure out what is the bottom clearing price, and that has nothing to do with business quality. (00:33:55) And so is Salesforce a good buy at 10 times free cash flow? (00:33:59) Historical artifacts would tell us a screaming yes. (00:34:03) The problem is that if you cut everybody's cash flows off at year five or six or seven, then all of a sudden I think you see the natural compression to between three and five times free cash flow. (00:34:15) Dude, that's crazy. (00:34:17) That has nothing to do with business quality. (00:34:18) That just says you literally mathematically take years seven through N of the future and you discount it to zero. (00:34:26) And having free cash flow in a war chest gives massive optionality. (00:34:30) We've seen this with Salesforce, we've seen it with Apple, we've seen it with Meta, with Google, with Uber. (00:34:35) Just having massive free cash flow, and you've got 10s of billions of dollars, you can put it to work and you can weather these storms. (00:34:42) J Cal, I think another way to think about this. (00:34:45) is, to the question about maneuverability and who has the gumption to make the hard choices right now. (00:34:52) Look at Benioff. (00:34:53) He's the founder of the company. (00:34:55) He's run this thing since its founding decades ago. (00:34:58) He is willing to bet it all. (00:35:00) He's willing to make the change. (00:35:01) And it may be that the index you buy in this era of AI transformation is the index of founders, that the founders who are still running their businesses are going to be the ones who are most likely to see the future. (00:35:11) They can burn the boats. (00:35:12) They'll burn the boats. (00:35:13) They'll maneuver the boats. (00:35:15) All of the guys who have hired managers to run the business are going to do the things that Shamaf's talking about, which is try and charge fees and try and maintain the old way of doing things as opposed to reinvent for the new future. (00:35:25) If you look at the 10-Ks, if we could figure out what the unit price cost and the trend and the inflation is of a per seat license for these products, I will point to the ones that are going to die first. (00:35:37) Can I make two quick points? (00:35:38) Yeah, wrap us up. (00:35:40) One is, yes, I would fully endorse what you said about Benioff. (00:35:43) He's made every previous wave work to his benefit, whether it was social, whether it was mobile, whether it was big data, all that kind of stuff. (00:35:52) What are the odds he's going to make AI work to his benefit? (00:35:54) I'd say pretty good. (00:35:55) So his stock might be a bargain right now. (00:35:58) So that'd just be point number one. (00:36:00) I want to say just a quick thing about venture debt, which is, look, I think it's fine when private equity guys use it because they know what they're doing, but I've always hated when founders take (00:36:10) on venture debt. (00:36:11) I know, Jake, how you agree with me. (00:36:12) Part of it is that founders forget that they have to pay it back. (00:36:15) They treat it like venture capital and they forget about that. (00:36:17) And then they get surprised. (00:36:19) But the other thing I've never liked about it is it makes you more fragile. (00:36:23) It basically subjects you to a bunch of business covenants and it makes it harder for you to do an abrupt shift in your business because now you've got a bank looking over your shoulder and they want to make sure they get paid and they have to review your financials and all the rest of it. (00:36:37) And to your point, J. (00:36:38) Cal, the companies that have free cashflow right now are the ones that have the most maneuverability. (00:36:42) I hate taking away maneuverability from founders, and that is what debt does because it subjects you to a fixed schedule of payments. (00:36:52) And so this is always a thing to remember, whether you're a business or you're an individual, when you put on that debt, it makes you more vulnerable to big disruptions in the market. (00:37:04) Yeah, you become incredibly brittle and founders who are listening, when you get that in peak markets, peak ZIRP, you're going to have venture debt people offer you tons of cash. (00:37:14) And then the problem Dave and I saw up close and personal in many different companies, where the founders would look at it as like, oh, I'm extending my runway. (00:37:21) Well, if you're a hot startup, there's always more venture capital. (00:37:23) There's always more people who want to own equity. (00:37:26) The equity sale gives you optionality and you have more people on your team, more people rooting for you and aligned (00:37:33) with equity interest, as opposed to now you have a debt instrument, they have a different goal, they have different downside, they're trying. (00:37:41) No bank wants to be your last three to six months of runway, because that means that in a high percentage of cases, (00:37:51) I've never seen venture debt work well to improve the quality of a business. (00:37:56) Never, it doesn't work. (00:37:57) I've only ever seen venture debt that damage companies. (00:38:02) And if you get the venture debt, you can never actually use it. (00:38:05) So the venture debt investors that ultimately make money, it's because they put money in a company and the company never actually used the money they gave them. (00:38:12) I hate this business. (00:38:14) I think venture debt's like the worst (00:38:16) If you get down, if the last money in the bank is the debt you owe to the bank, they're going to rug you. (00:38:24) That's when you get rugged. (00:38:26) 100%. (00:38:27) You think they can afford to lose 100% of their money when they're getting an 8% return or something like that? (00:38:33) No way. (00:38:34) That's not how it works. (00:38:36) VCs can afford that because we have the opportunity for (00:38:40) a 10X or 100X or 1000X for that moonshot. (00:38:43) So we can accept a bunch of zeros. (00:38:44) The bank can't accept a bunch of zeros. (00:38:46) Well, and then when they do get scared and when they do think they're going to lose their money, (00:38:50) Wait till you see what they extract in terms of value, what they ask for. (00:38:54) They will ask, they'll double the interest rate. (00:38:57) They'll ask for warrants. (00:38:59) It's basically like being in debt in prison. (00:39:02) Tamatha, you could talk a little bit about your experience when you were in debt in prison. (00:39:06) It's not going to be pleasant. (00:39:08) I've been in debt. (00:39:09) I mean, I've had a $420 million credit line. (00:39:14) And I had a moment where it was reflexively kind of collapsing inward because the assets that I was using to secure it shrank in value in a moment of market disruption. (00:39:24) I was scrambling. (00:39:25) And then at the same time, there was a risk. (00:39:29) It was the worst moment of my professional working life. (00:39:31) I had like a couple hundred million dollars sitting at Credit Suisse and they were about to implode. (00:39:36) And so on a weekend, I was trying to figure out whether my money was still there. (00:39:40) I had always had this rule, don't have debt. (00:39:43) And then I violated it to try to run the number up. (00:39:46) I almost got run over. (00:39:48) I almost lost everything. (00:39:50) I will never do it again. (00:39:51) And if I ever do it again, if you guys ever hear me do it again, please just come and punch me in the face. (00:39:57) We will. (00:39:57) We've been waiting for an excuse. (00:39:59) Can we punch you in the face for us?