The consensus is in law, Anthropic will probably win a good slug of this case, which is different than saying they're going to win the fight. The idea that you can just have all this for free is at some point going to stop because someone's going to have to cover their nut.
This week, Anthropic sues the US government. Oracle and OpenAI end plans to expand flagship data center. Meta moves in to absorb the surplus AI data center capacity. CrowdStrike, they beat expectations but trade down. And then finally, we have our stock picks. What to buy, what to sell, public market predictions coming up.
I think getting rid of juniors is where we get budget for these data centers in part.
This goes really cold. You can have dispossessed urban poor forever and nothing happens. But if you piss off the twenty-something-year-old middle class, the overeducated elites, they tend to cause trouble. The truth is OpenAI has the most amount of surface area to attack of any company in the world right now.
The era of gentle deceleration has ended. It's dead. Ready to go.
Boys, we have a big week of news. As always, we're going to start with this week in Anthropic. So, we have gone from a couple of hundred million dollars in lost contracts with the Pentagon and with the DoD to potentially billions of dollars at risk. If we start here and the billions of dollars at risk, how did you analyze that? And maybe Rory, the wonderful context giver, if you want to provide some context.
Yeah, I will. I think that Anthropic, in my view, correctly sued the federal government in both California and subsequently in DC just for procedural reasons. Basically saying that the government's designation of them as a supply chain risk was incorrect, badly done, and would not legally stand up. And the reason they're suing immediately versus talking is they're stating that quite a lot of revenue could be at risk for them because the Department of War and the government, I think I should say that the administration, has articulated a very aggressive definition of supply chain risk. If you recollect last week, I was not entirely pro the Anthropic position and I stand by that, but I think the government is now overreaching as well. So, like all good disasters, both sides are doing things wrong. But it's totally rational for the government to say, 'You want $200 million of revenue from us and you want to get in our shorts and tell us what to do, and we're the Department of War and we ain't going to do that. So you're not getting your 200 million.' That's totally fair. But designating a supply chain risk has a series of escalating consequences. At a minimum, it implies no one can use them. No one can use them selling. No other company selling to the Department of War can use them for those contracts, which is one step more. But again, I see how the Department of Defense might get to that place. But then you're seeing the government articulate these really expansive definitions, which is throw them out entirely of every government contract, which seems an overreach, and then even more overreach. And it's actually not happening because the various cloud providers have pushed back in it. There was one version that said if you use Anthropic at all, Microsoft or Amazon, then we won't use you at all in the US government, and that's an element of frankly almost ludicrous overreach. So what's happening here is basically Anthropic and the Department of War are trying to figure out, you know, where in the spat, how big is the blast radius of consequences from this $200 million contract. And my perspective is there should be consequences, but having it be this big a blast radius doesn't really stand up. And I think the consensus is in law, Anthropic will probably win a good slug of this case, which is different than saying they're going to win the fight.
Well, look, I mean, there's something to me that's minorly interesting, and then more interesting. The minorly interesting thing is how hard the hammer came down. I mean, they're going to lose like the, it's like tariffs. Even if for some reason Anthropic wins their case, they'll just, the DoD will just come, the Department of War will just come back with another issue, just like tariffs. They'll find another justification to block Anthropic. Or even at a minimum, I think the real panic, I think maybe what they didn't realize, and then I'll talk about what's more interesting, but I think what they didn't realize is what a blocker it would be with their customers. This is just a classic B2B sales issue. You know, you walk into these big deals and they said it in the complaint, and we're struggling to close deals because prospects are worried that just some of their business has exposure to the federal government and the Department of War. And then their competition, OpenAI, said we don't have these problems, and they may steal the deal. Like they literally said in the complaint, this is an issue, and this is classic B2B stress. Like you don't, it only has to be, the risk only has to be ambiguous in a B2B deal for you to lose and go with another leader. It just has to be, it's a little bit less secure, or they went down for two days, or the database was left open. And so there was a classic B2B panic. And I don't want to get into politics, but I honestly don't. I mean, the government is saying, listen, we need you for war. And in fact, it's more than that. The government is saying we used Palantir without Anthropic's knowledge to conduct a war in Venezuela and depose a government. We did that without Anthropic's knowledge. So it happened. But the government's going to say, 'You're either going to agree to this or we're going to freeze you out of the US economy.' And it doesn't matter what a court says because they'll come up with another justification. There'll be another one. There'll be a different type of supply chain risk. It'll be endless until they bend to the will, I think.
Yeah, but you may be right in terms of what the government will do. But I just, as I say, having been critical last week, I want to say it's wrong to try and drive a widely successful American tech company out of a good slug of the economy simply because they don't want to serve this one particular need. And it's interesting if you read the complaint, a lot of what Anthropic is talking about is literally their First Amendment rights. This is not a contract. What they're basically saying is we on Anthropic articulated our less than love for some of the things that the Department of War might use this technology for, and you're trying to ruin our business simply because we use those words, and that's our First Amendment right, and that's un-American. And so while I think you are entirely correct, it's a really tough position for Anthropic to be in. And as I said many times, I don't know if they should pick this fight. I also think it's particularly unfair of the US government to try and be this sweeping in their actions against what is one of the great American success stories. If AI really is all about the leading edge of the economy, right, it's probably a mistake to pound the crap out of one of the two leading companies in that space simply because at the margin you find them slightly sanctimonious, which many people do, and you don't like their politics. You don't have to buy from them and put your $200 million in another place. Maybe don't let them into the Department of War at all, but stop trying to cut them off at the knees across the entire B2B infrastructure. I think it's a little bit of an overreach, but I think you're right, Jason. Me thinking that is very different than them changing their mind. There's a style in the administration that's prone to overreach, and I'm sure Anthropic deeply regrets getting down in the mud pit wrestling with this particular opponent because it sucks.
What actually happens here, like realistically, years time, does one cow and give up play out the realities for me?
I think the interesting thing, I think on the merits of the lawsuit, that I think Anthropic will probably prevail and may even prevail quickly on some of the, you know, I think if I read the lawsuit last, I think they've actually moved for immediate redress. I think, you know, to basically get it reversed. I think they might prevail on some of the things. If I was to be logical just based on prior, they might prevail on some of the law. They won't clearly get a $200 million revenue deal with the US Department of War. But in a perfect world, if they bend the knee, prevail on the law, what we've seen in multiple other situations is you end up with some kind of, you know, acquiescence to the administration in return for leave us alone to pursue the rest of our daily business, our B2B business. We're not going to say, the administration is not going to be saying to Microsoft, hey, if you use Anthropic anywhere, you don't get any government business, which is the most overreaching version of this. And in return for that, Anthropic will probably have to bend the knee and be slightly supplicant. That's my gut. They'll win the law just like many of the other institutions have won some of the cases, but will find that they've picked the fight they regret and just want to settle it.
Does this impact their going public later this year? Polymarket predicts Anthropic no longer does.
And Polymarket and Kalshi, no. I mean, Polymarket and Kalshi were able to tell you the date at which we were going to attack Iran. I mean, there's the amount of, but Polymarket knows lots.
The only thing I, it does, they are tied, right? The only thing I would say, maybe there's a related point that's broader than this I'd like to make, but I do think we overstate IPO risks in general. Rory's been through this, right? We all, and listen, it totally makes sense that people would react, oh my god, you know, Anthropic has to work to perfection to pull off the IPO they want given the amount of capital they want and the valuation of a trillion, and being a supply chain risk would be risk factor number one and freak out the public markets. I think we've all seen every startup we've ever worked with IPO with several things that had to be highlighted in the prospectus, and it never seems to be the end of the world. IPOs are very binary, right? Either folks get super greedy and they're genuinely massively oversubscribed, or they barely get done at all. So I just think that is overstated by the media that there's existential IPO until it shows up in the numbers. I don't think fear overcomes greed.
And so we didn't actually get to that though, which I'm just going to push you on. Does it show up in the numbers? Yes or no?
At the level of growth we're seeing, it can't show up in the numbers yet. The beauty to numbers, and this is why I think all 80% of public B2B companies are in much worse shape than they look, is they're backwards looking. Every set of financial statements is backwards looking. And it's even worse, unless you do a SPAC or something, you can only project so much. So you're stuck in backwards looking land. The closer you get to the IPO, the less you can say about the future. And the more you have the benefit of hiding in last quarter. I'll be clear. I don't think it shows up in the numbers provided what appears to be the consensus legal outcome happens relatively quickly, which is you can't design, first of all, the process of designating a supply chain risk has more steps than if you read the actual legislation than just deciding at five minutes' notice. There's a review, like all government actions, there's a review process, there's a comment, all those kind of steps. And then the second thing is, like I mentioned earlier, the interpretation from the hyperscalers, who remember do a huge amount of distribution for these guys, and the developers who are clearly flocking to the app right now, clearly is I can use this for non, definitely non-Department of War and possibly non-government use cases without any supply chain risk whatsoever to a rounding error, right? If that version of the world, which appears to be the correct legal interpretation, prevails, then yes, if they went to court and they started to lose those cases, then they would be on the back foot and then they would have to probably settle pretty quickly. But the consensus appears to be that from a legal perspective in this case, the administration is overreaching. So right now, no, it's not showing up because yes, you're clearly down 200 million in license revenue. And if you're the sales rep for the department, if you're the sales rep for the Pentagon and Anthropic, it's going to be a slow week, right? Conversely, if you're the sales rep or the guy running, you know, PLG growth for the Anthropic Cloud app or for the Anthropic developer products, you're exploding. Jason's right, against that wall of growth, there's nothing showing up. If those numbers are correct, let's do the math here. 200 million, let's call it 18 million, actually 17 million a month. So they're doing 1.5 billion, 17 billion a month is 1%. So this is a company that's 10x-ing, so it's going to drop 1%. It's lost in the noise. I mean, I'm sure they deeply regret ever taking on that $200 million contract because it's in the noise compared to the business, and now it's produced this contingent risk. One of the things I reflected on this a little bit was on the one hand, it can seem very localized to Anthropic, right? Its principles and a DoD, DOW, Department of War deal. But I think a lot of us are going to have to wrestle with these issues. I'll give you a very small example. I've basically worked with a ton of next-generation CRM products, okay, that everyone's building. We talked about Dad VC, right? Everyone's building. One thing that the, and the approaches to next-generation CRM are very different. Some are very agentic-focused. Some are redoing the classic core lead contacts opportunities. Some are building together with existing folks. A lot of approaches, but there's one thing that all of them seem to have in common. Every single thing that a human does is fully tracked and logged. Everything. Now, if we go back in time when things like Gong blew up, at first we had a pause. We're like, oh my god, my business is tracking not just a random call, but every single call. And then every, like we had to pause for a moment. Is that okay? Right? And this became the way we worked, right? Gong, Gong at all. Then, you know, even during this pod, we kind of granola, is that okay? You know, is granoling people in meetings. Okay, now every app is, this is becoming necessary core CRM functionality to record and manage every keystroke, every interaction, everything you type, everything you do because otherwise you can't automate a CRM. Like the next-generation CRM just does not work if you have any privacy in the workplace at all. And so it's just an example of you wouldn't think that a next-generation CRM would have to deal with the same issues as Anthropic and the Department of War, but I think we're, many of us are going to have these issues. Are we going to lower, change our morals? Are we going to say a 40%, a 50% layoff is no big deal at Block because it's a part? And I think we're all going, as we chase these faster and faster growth companies, I think we're going to, we're going to discard more and more of our previous moral standards. And maybe it's like that's the AI life. But I have a little bit of anxiety here. Just a little bit.
A technology that's so important that it's consuming 50% of the capital investment of the entire United States probably should have a significant number of pretty significant societal changes.
Two signs that like maybe we're facing or entering an age of more realism. One is Oracle and OpenAI and their plans to expand their flagship data center. So Stargate Texas data center expansion to 2 gigawatts potentially being axed, capping in at 1.2 gigawatts. Is this early signs of the end of the capex cycle and hype cycle or not?
More no than yes. Look, and because I'm sitting and going, I think we are overinvesting and I think at some point there's going to be a reckoning and a bunch of companies are going to realize they've massively overinvested. But I don't want to be that kind of guy who's trying to call the turn, you know, three years before it hits, right? Because I think the proof that it was, look, it speaks to Rory. How can we be overinvesting when Jason tells us that this is the year of inference and we'll have inference running?
Can we come back to that? I will come back to that.
Yes, we can't possibly be overinvesting.
I mean, the point is this on the question you've asked. No, because instantly Meta said, no, if you guys don't want this data center because you're having some issues, we'll take it. So I think if you just look at all the comments from the hyperscalers, if you look at Jensen's comments right now, at least to Jason's point, demand is insatiable. So demand right now, if demand is insatiable, then you can't use this kind of one-off set of dynamics around Oracle as a sign of, oh, it's the capex cycle turning. I think that would be over-extrapolation.
I think that Meta is playing a game. Google's playing some of it, Amazon's playing some of it, that it's hard for Oracle and anybody else to play. Which is that ultimately, and I think Meta is the only one that I've known that's explicit, is they're betting on a world where your AI is 24/7 persistent and infinite. Right now, most of us are using a little bit of ChatGPT, a little bit of Claude. Maybe we're using a couple hours of coding compute, which is quite expensive if it's not bundled, and that's it. But none of us are talking. I mean, we made fun earlier in this day of the OpenAI pen with Johnny IV, right? We made fun of it, but I think we missed the point. And at least I did, I missed the point when we talked about it, which is I didn't fully understand, because this is a few weeks later, right? This is six months later. I didn't fully understand what the world would look like when our AI agents run 24/7 and the amount of compute that we will need to do that is obviously orders of magnitude. Then take multiple agents running in parallel, which is everything Cursor is about, everything Claude Code is about, is multiple agents running in parallel. Replit v4, it's all about running five, ten, twenty, fifty agents in parallel 24 hours a day. I mean, I'm not even smart enough to do the math, but we're talking about multiple orders of magnitude of compute that we need today. Does it get more efficient? Maybe. On the RAM side, it's not. We've already run out. So anyhow, my point is I think one thing that, and maybe because Meta is so consumer and also in some ways so far behind, but I do think they're making this bet of 24/7 persistent AI in your life at a consumer. And so they will buy all the compute that is available that they can afford to make this dream happen because they're well positioned for it. As silly as it sounds, they're probably the, well, unless Claude, which is playing its own endgame, or ChatGPT can figure out its social side, they're pretty well positioned to be the AI that lives with you 24/7. Them and TikTok, they're pretty well prohibited. And so the world's going to be so different, right?
Genuine question, Jason. What is that thing that they're shipping that you're talking about? What I don't, what is my AI that's 24/7 that I can get on Facebook or Instagram? I don't see it.
It listens. It actually can see. It can see everything you do. It sees everything you do.
What tense are you using? Are you using the present tense or the future tense?
What should happen today? What is entirely possible today? It's just we don't, we can't afford it.
No. Which will be is that as soon as this call ends, this Zoom ends, my AI talks to me. Hey, Jason. That was something pretty dumb you just said with Rory on the call. Let me explain where you got it wrong and let me explain how this ties together. I don't even do anything. My AI is already ahead of me. The second this Riverside ends, that could be true today in everything you do in life. Every partner meeting, every pitch you make, it just we cannot afford for the moment the level of compute it takes to have that.
Right. I'm willing to stipulate partly in humor that the amount of compute it would take to correct all the stupidities that I say, you say, and Harry say might well require data centers in space. So you might well be right, but I'm not sure we can afford that. I mean, so but what about your AI every minute of your at least printing out your personal life, your work life, helping you be better on every single thing you've said or thought? Every single decision you, how many micro decisions do you make all day long, not even the big ones of a hundred million check, how many checks do you make every day, every week? It's impossible. And if your AI has full context, full history, has every deal that's ever been done at scale, every deal in the last year, every deal any VC has done, every article, every interview, is it hard to imagine that that AI immediately after every conversation you have won't enrich your work life? Like of course it will, of course it will, it will be instantly better, right? And so that's, but I don't know that the problem is in the short term. I don't, Oracle doesn't have the cash to keep up going to Harry's point, it just doesn't have the cash.
And that is it does. But one of my bigger asks on this whole is it all going to go wrong question is the zoom out comment is the people who say it's not talk technology and the people say it is talk economics and we kind of just don't quite overlap, right? And that's the core of the challenge in this discussion. There's a million things you could envision doing with quote unquote AI. The question is, are they worth doing at what price? I mean, I was just doing the math in my head. There's 600 billion dollars being spent on capex. There's 150 million people working in the United States of America. 150 million. That's 600 bill. That's roughly 4 grand per head, right? I don't know if an employer wants to spend four grand per head on automating with AI. I mean, if you run through all the people, there's a bunch of people, you know, serving coffee. They probably don't need four grand worth of AI. You start doing that math and you say to yourself, is the return there? So I stand by my comment. I think we probably are overinvesting, right? And at some point, you know, those chickens come home to roost. I don't think it's a controversial statement, Harry, pushing back because Zuckerberg, who I think is wildly smart, has even said it. This is pro and everyone said it. Bezos has said it. This is probably going to go wrong, but I'd prefer to ante up and be in the game and have a chance of winning than not ante up like Apple and know I'm going to lose. Right? By definition, if six or seven people articulate that that's the game they're playing, and only six or seven people matter in this discussion, then it is almost total logical to say we're going to overinvest. Because that's what game theory says. Until such time as someone realizes, oh, I'm in the poker game, but I don't have the right hand. I got to fold. Until that happens, we're going to overinvest. And I don't think it's happening in 2026. The only person that's even struggling slightly to cover their nut is Oracle, right? Because they have the weakest balance sheet and the least compelling use case, which is why at some point they're going to throw 20 or 30,000 people over the side to make the P&L work, right? Meta can borrow. Google actually has a business and Microsoft is quietly stepping back from the table. So, and so this game goes on for another year or two, but I do stand by my comment. There's a level of overinvestment going on.
I thought it was interesting. It was rumored, and this is rumors so Rory does not like rumors, but that Alex Wang is being sidelined and that his position is no longer as superior as it once was with the creation of a new lab and him not being in charge of it. That lab director reporting directly to Bos and it appearing like the Scale acquisition was bluntly hastily done and a mistake. And to your point, Jason, on Meta being behind, that is the kind of consensus that's being shared that would be overinvestment. I paid $15 billion for an asset that 24 months later, 12 months later I put on the bench. That is the very definition of overinvestment. So I actually think you made my case for me, right? And the fact that the response of Meta would be, oh well, we'll do something else, shows that we're not yet at the overinvestment being terrifying stage. We're simply at the I got to win, try something, if it doesn't work, try something else.
Can I, can I, I want to come, I'll add, I think this Moltbook is an interesting micro story of Meta's very interesting AI strategy, AI M&A strategy. Before I get there, this overinvestment thing, I just want to bring up one thing that happened this week which I think people misunderstood and is so important to my point of we're not even like the amount of AI we want to use when we're running 24. We need massively more compute than we have, which is Anthropic launched a true cloud code review. Okay, great. Hooray. To find bugs, to find issues, and they said it's $15 to $25. And the internet blew up. Oh my god, this is so expensive. Like I only spent $200 a month for Max. I spent $20 and now you want 20 bucks to do a code review to check for bugs and issues. And there certainly are folks who are trying to manage their cloud code cost to the bare minimum, right? They're vibe coding from a cafe in Thailand and they got to keep. But the response back from the head of cloud code or at least whoever built the feature was, my god, we're spooling up 10 plus agents in parallel to run for 20 minutes to find every single bug in your product. This, like you could do this manually and I've been doing it manually in Replit for months, but to do this in one click, it's profoundly valuable. Like humans can't even do this, find every single bug in 20 minutes. And so the amount of compute now they need, even Claude Code needs an extra 20 bucks of compute to do this. And you know what you'd really like to do? Run this after every commit, not episodically, not when not getting up your nerve to spend $20 worth of tokens, which is a lot because this isn't subsidized, right? This is Anthropic's. You'd love to run this 10 times a day after every commit you make. This is, that is orders of magnitude more compute being used for this code review than we used before. And so we're just between that, everything running 24/7, we've just scratched the code because if you've built anything in cloud code or Replit or Lovable or whatever, what you realize is it's just the beginning. Like who's doing all the QA? Who's doing all the code review? Like we've only scratched the surface. We need data centers in space so that I can keep coding. But I think the point is interesting from the news. I think people miss this point of this code review thinking is $20 a lot to review your entire codebase and find all your top bugs. Is that a lot of money? It's 20 bucks.
Of course it's not. You know, there's, we're happy investors in CodeRabbit, standalone company doing it across multiple tools. And yes, now we have exciting and interesting competition from cloud. That's bad on the other hand validates the space and says you need to do this. I think the description of the product, of the need, is exactly right. People, if you're generating infinite amounts of code, right, using all these agents, the idea that you're not going to kind of do code review is absurd. So you're going to have automated code review. You're exactly right. So, and I think that whole comment about your comment about spinning up all those agents was exactly the tech you illustrated my point. From a technology perspective, there's infinite need for large amounts of agents to large amount of code review. From an economics perspective, you have people complaining about the fact that they're being asked to pay 20 bucks. And you're exactly right, Jess. The correct answer, which Claude was too nice to give, because we know they're nice people, despite what Pete Hegseth thinks, was for Jesus's sake, I'm giving you, you're automating an entire developer, and we're giving you automated, we're letting you generate infinite code for 200 bucks a month, and you won't even pay 20 bucks to check it before you put into production. For God's sake, man, pony up. That's the correct response. And that's where technology is meeting economics. The idea that you can just have all this for free is at some point going to stop because someone's going to have to cover their nut, right? And then we'll discover the really interesting question is how many of those users are prepared to pay 20 bucks per code review or 200 bucks per big code review or whatever, right? So I do agree with you. The demand is infinite. The number of things you can do is just freaking amazing. The question is how much, what are the things that people are prepared to pay full boat for, right?
And it was just, it was interesting by the way, and it kind of validated a little something we were thinking. Claude, it's very, I would love to be the pricing person for Claude. And you've done pricing, Jason. How do you decide what things do you kind of, because they've got some really interesting things going on. The API is expensive and metered. On the other hand, Claude called the subscription, as you point out, you can get a whole ton more than you should be at 200 bucks. So they're kind of not metering that. And then they decided, and in my view wisely, hey, maybe code review is something that's kind of feels managerial and maybe there is budget for that. And you know, you and I both did deals where they were, you know, low in SaaS land, which were PLG for individual users. And people were like, oh, how are you ever going to make money on, you know, small low-end signatures? But the more you go to management of signatures in the enterprise, the more there is a propensity to pay. And I think this is the beginning of the same thing. You, Mr. Hacker, you could have all the free codegen you like for 200 bucks, but if you're going to be rolling this out into your Bank of America systems, you're going to want the deluxe, not the $25 code review. You're going to want the $250 security-proofed code review, and you're going to see that kind of push to make money here. I don't know if that makes sense or not, but I think that's the whole dynamic is the whole ton of stuff you can do, and then at some point someone's going to have to pay for those data centers, and it ain't going to be Pete at the Department of War.
I think related to that, the thing that, you know, sorry Harry, you can keep us on track. You're the boss. But I think one of the, there's certain things we've talked about in the past that are true, but we're also willing into existence. The enterprise is willing its desire for AI to replace humans. So it's going to accelerate because they want it to accelerate. I mean, my god, the one that is being accelerated is hire no junior developers. Hire no juniors. This is being willed into existence. No one wants to train anybody. It ties to this code review because listen, if you're going to hire no juniors, of course you can spend 200 bucks or an extra $500 a month to do code review for your senior developers. They're already insanely competitive, productive, right? And I think the death of the junior, it worries me. And I had dinner last night with my son who's off the charts smart, but he's at Penn State, which is not a top 10 Ivy League school, and he said there are just zero jobs for anyone in CS or math. There's just zero. There's no one even coming by to hire them for C-tier. There's just, he has job offers because he's like publishing things on Jensen fine string theory that I don't understand. So great, six people in his class at his large state college have tech offers, but it has become not only in our world, but in his world, it has become we just don't want juniors. We don't want to train them for three months or six months. They don't know the tools cold. They don't know it. And that's where the budget comes from when you hire no juniors. When we're just a world of the middle of the bell curve in terms of experience. We don't want folks that don't want to pick up the tools and we don't want folks that aren't experts. And it's going to ripple. This is my big worry for 2027 is it happens. It's going to be true in sales. It's already true in support. Like you want humans in support, but you don't want juniors. My god, I don't want a, like I want a human that knows CodeRabbit cold that can handle the escalations. I don't need junior marketers. I don't need junior nothings. And Harry, not to get us off track, but I think that's, I think getting rid of juniors is where we get budget for these data centers in part. We're just getting, they're just dying faster than we ever thought. It's happening in front of us. The death of the junior. Bad for society, but it's a reality.
I think it ties to two fundraisers that we saw. Rory, you may disagree as an investor in one of them. I would agree as an investor in another, but we saw Intercom raising $250 million and we saw Lora raising $500 million. And I think both tie to your point there of the replacement of juniors and lower ranking death of the juniors.
Why would I want, why would I mean, God, I mean, I know I'm not a Lora or Harvey expert, but I do know the dated space. I mean, who the hell wants to wait for an associate to scale up for two years, right? That was the classic legal hiring. You hire someone real from a top school ideally that was top 10% of their class. They have the IQ and you basically subsidize them for two years until they cross the line and become senior enough that you can give them a, you know, they can work with a top client or take a case. Who God, who wants to do those two years now unless you have to, unless you have to.
This has been one of our long-running discussions and it's good that I'm evolving my position. J, I said this, I've kind of come more to your perspective that there clearly is something going on here. And even though I don't believe in the catastrophic, you know, mass unemployment scenario, I think you're exactly right for a couple of very targeted demographics. One of which is entry-level commside type jobs. Other would be actual customer support jobs, actual lawyer, actual legal associate jobs. There is a meaningful impact on our employment right now. So I just want to acknowledge you, right? Well, let me finish. It is interesting because, and I think that's true. On the other hand, yeah, part of you says typically when there's new technology, right? The best adopters are young people because they come in with less prior. And there's a part of me that says, are these schools doing their job, especially now that it's what, 2026? If you're graduating graduates who aren't AI first and totally awash in using these tools, you're just not doing your job. And I think there's an element of not.
You're unable to alter curriculum fast enough to update Sinclord. Curriculums take time to change.
They need, well, they need to change that because at a, I mean, because if you're looking at investing, you know, 100 grand a year for four years at a four-year college education to get a computer science degree that turns into unemployment and a barista job, you should be pissed. And you will be pissed.
But I'm with you. But when you look at Andre Karpathy's journey in six months' time from bluntly using it for 20% to 80%, and this is Andre Karpathy that I think it's unreasonable to expect an educational institution like a university to alter the entire curriculum on Claude's ability to progress throughout the kind of skill journey. It certainly doesn't appear to be happening whether Rory's point, should we be expecting it from these institutions? Yeah, they're becoming, you know, they're becoming just babysitting for expensive babysitting for four years to.
And they shouldn't be. And you know, I think give him credit. I think 15 years ago, Peter Thiel articulated the perspective that the ROI on higher education is going down and at the time it was still high. And it's worth pointing out even to this day, unemployment among college graduates is lower than unemployment among high school graduates, which is lower than unemployment among high school dropouts. So you're still quote unquote better off going to college in terms of employ. But the point is now you take into account the net present value of that 400 grand. You're not getting your money back on anything other than the best degrees. So it's not, you know, these are not the most deserving people on the planet. They still have a lower 3 point something percent versus 4.4 average. But there is still, but there definitely is a trend here that would be troubling if you're a recent graduate. So I'm kind of with you on the trend.
Roy, when you say you don't see the mass unemployment case or you don't quite think that or think it's an overreach, when you look at legal, computer science, customer support, bookkeeping, those alone if it had no impact anywhere else would be mass unemployment.
I don't know if it would be. Again, look at the numbers. I actually was looking at it today. Entire tech industry excluding tech people employed in non-tech companies. Technology NIC classification 51 tech companies, 3.1, 3.2 million people in the United States of whom roughly 800,000 are software developers. Right? If you go down by half, that's 400,000 software developers, right? It's a lot of people. It's, but it's, you, it's 2% unemployment, right?
And how many are in customer support?
Much bigger number, right? You're exactly right, but also way more fungible in the sense of a lot of the low-end customer support, there's lots of other service-type jobs folks can do. I'm just a believer, Harry, that over the, again, I want to be over the medium term, the technology takes longer to diffuse than we in Silicon Valley allow and people are more adaptable. So over the medium, I'm not, I don't see any kind of 10% plus tech unemployment, which is, and I know you're going to disagree and the point is we don't have to argue about this because in the end we'll know, right? I just think I have 200 years of facts on my side and you have, but maybe you're right. That said, I do believe you are right in the short term in isolated pockets of very vocal people are really struggling. Look, I have, yeah, three recent college graduates too over the last five or six years, so I'm totally aware of how hard it is to, you know, get a job as a college gradu. And I think it's a real issue and we better get on top of it because one of the other things the French Revolution teaches you, right? And actually, history in general says, this goes really cold, you can have dispossessed urban poor forever and nothing happens. But if you piss off the twenty-something-year-old middle class, the overeducated elites, they tend to cause trouble, right? And I actually think the continued production of masses of college graduates are fused with their sociology and English degrees facing income science degrees where they've done played it by the rules for 20 years and then given them on mass unemployment is going for that cohort is going to be problematic. So I'm kind of in, I'm not, I want to be clear just because I think the macro trend is not there doesn't mean I don't think the micro trend is a real issue. And that to Jason's point, I know if it's a moral issue, it's an economic issue. And I predict in 2026 it's going to be a political issue. I mean, it's so funny to watch the NPS survey, right? The NPS for AI polls really negative, right? It's like, and as someone tweeted, you've been telling us you're going to blow up the world and make us all unemployment, and you're surprised to discover we don't like you. I mean, it's almost.
Yeah. At the same time, if you look, I'm sure Intercom data says this, everyone's data that works in a heterogeneous human AI agent environment. The CSAT is always top 10% for the agents. It's never number one. They never say that the number one most popular person is always a human, but the agent's always in the top. That is more telling than most people realize because I think the theme that's like, I'm, it's early but it's accelerating is we just rather work with agents. We'd rather hire agents. We'd rather buy an agent. We don't want to, I mean, Harry's got 3,000 applicants for his EA position because he needs a human. But I tell you, he'd rather have an agent if it worked. He don't want to.
interview 3,000 people. If he could have an agent, he'd pick an agent. And we're all going through that. We're all... We just... And it doesn't even matter if it's cost effective.
We just would rather have an agent. I've been thinking a lot. I know this is Captain Obvious stuff, okay? But my god, when the... why do some of the AI B2B startups grow so quickly, right? Um, the simple answer is product market fit, product-led growth, agent-led growth. But um, as you dig deeper and as the year goes on and as we roll into late '26 and '27, and I really like... people just want to buy an agent. They don't want a human. And there is... and if you can deliver perceived massive ROI, that is so appealing. It is not layoffs. It is not this. It is, oh my god, I could get customers. I could run campaigns. I could do support without humans. My god, I want that. And that's what our... and so when I look at so many of the mediocre products the public B2B companies have launched, they do not service that demand.
I don't want humans at Atlassian or HubSpot or other folks we have. I don't want them. I don't want to make my humans 8% more efficient with your AI. I don't want these humans. I want an agent to do this work. And when you do that, whether the Lora, Harvey, whatever, a million on the show, whether they fully can realize that potential, people will line up at your door when you can give them an AI instead of a human, an agent. They just... that's what they want. These are the companies we want to build and run, and they will keep shrinking as a size of revenue. They will keep shrinking for this reason.
Actually going to... as is so often the case, I agree with the broad direction and might disagree with the... not so much the sequencing as the narrative. I think a lot of the B2B adoption to date, up until... with the exception of customer support and in the last year with the exception of coding, I think folks like Harvey, Lora, and GCI, they're all about... they're not actually about human replacement except very much at the margin, at some vague level, maybe I don't need the 17th associate at Linklaters, right? They've actually... because so far it hasn't been agentic. So far it's been primarily, yeah, let me help you, Mr. Lawyer, be smarter, work more efficiently. Yeah, at some level not ask an intern to do something, but it hasn't been a wholesale replacement story. I think you're right. It's been at the task level. I've been able to increase the number of tasks that I can do and therefore decrease the number of tasks I have to pull in an associate to do. Right. I think you're right, Jason. Now with agents, and I think we've seen it first in software, it's stunning what's happened in the last six months in terms of the ability of these things to genuinely just code start to finish. And therefore, you know, this is where I think you are right. We might be at the start of some of this identification, and the next 12 months will be B2B companies saying, now I finally can automate this task entirely, and then we'll find out what the demand is.
I like to think through a frame of what will my children look at me and how I used to live, and god, I can't believe you used to do that. And I think one of the things they will say that to is, I can't believe you spent years training people and then they left. Like all those years wasted and then they left. That's so nuts. It's crazy. I think that will seem incredibly archaic. I look at my EA today, we've mentioned it, I've spent years training her. She's amazing. And then she leaves. It's nuts. An agentic world which compounds education and learning and they never leave you. Jason, would I prefer it? Yes. All of that time was wasted. I've got to start again.
Yeah, that's... I just think if you're founders in B2B, this is what you should be building, is something where folks will line up the door because you're automating, because they'd rather work with your agent than a human. Lining the hell... like, and most of the public companies are not even thinking this way, but the ones... the startups that do blow up, they line out the door.
Example of the agentic ones, please, as distinct from the automated task ones. Give me a... what level of agent are you talking about?
I mean, literally, I mean, listen, there's a bunch, but I've done so much in these GTM agents, these go-to-market agents. Okay, so this last one that we deployed, this Monaco one that Sam Blonde did, I mean, they closed seven figures in their first five days and they have 60 days of people lined up. And why... and look, I'm not hyping it. They're all good, but why are they lined up out the door? Okay, this is the more interesting question. Why are they lined up out the door? Because it... whether they're fully successful on all their journey or not, it's worked for us as a tool. It talks to the prospect. It texts them. It pitches your product and it sets up the meeting and the meeting is fully ready to go and close. If you can deliver, if you can deliver that, okay, and it's a high bar and I don't even think it was possible 60 days ago, okay? Okay. And all the other tools we use, just before you crit, we can all poke holes at it. It's early, but people will line up the door for this stuff. They'll line out the door for this stuff. And so that's what you... whatever category you're in, you want to build this agent that is sufficiently better than 90% of humans that everyone will pick it. And the budget magically comes out of nowhere. It literally comes. That's why we're just shocked at the TAM. Like the TAMs for some of these products don't even make sense to us. Like they're not consistent with the past. But people will line up the door if you can do this without humans. I mean, we built our own VP of marketing. Literally yesterday, this week was the first time it led our human staff meeting. Our AI VP of marketing led the team meeting. It led the meeting. It summarized every single metric, every single thing, every single thing the humans on our team had to do. 10k ARR if it led the meeting. I mean, as great as Clay is for enrichment, it's great. If it could do that, I'll give you $50,000 tomorrow. An extra 50 grand tomorrow.
You're doing the Intercom deal. Are you betting that they beat Sierra?
You know, I think that again, I prefer to let Owen speak for himself on what he wants because he's a very ambitious man. I'll tell you what I am betting on. I'm betting on that they can carve out a significant space in that market for that product and that they've had explosive growth in it. And coming from the SaaS, I'm betting that... I'm betting that that team coming from the SaaS background but being grounded in what it takes in AI can take a significant portion of market share. They tend on average to be slightly smaller in terms of customer size at the high end of Sierra and they compete more with Decagon. One thing that I think in a lot of these categories that have traction, but if you want to say, and I'm only a partial expert, but if you want to say Intercom versus Decagon versus others, right, Zendesk, Salesforce, one thing that people miss a little bit in terms of the winner, which you kind of asked Rory, right, is FDEs, forward deployed engineers, for the moment are a limiting factor. And let me explain what I mean. What it means is no one has enough fully trained FDEs that can get a customer up and running fast enough. No one has enough of these resources. Okay. And what it means is that almost every vendor has to pick lanes. They have to... you can't do as much as you'd like to do if it was all software. You've got to pick where you're strong, right? And if you're best at digital transformation for enterprises, maybe that's what Brett Taylor is, you got to pick that, right? And so not that these folks... I know the sales team and folks think they're direct competitors, but you don't... everyone has more demand than S-tier FDEs to service that. So they're all picking lanes today and that is a frontier we haven't crossed yet. The need... and so in that sense they're all services businesses in a sense that this VC meme is true. They're not long-term services businesses, but if you need humans to train it, if you need humans to manage it, if you need humans to iterate it, you have the same constraints that services businesses have. And this FDE thing is a limiter. It's exciting because I think the losers, a lot of the losers out there can't even get good FDEs. They can't even... they can't... if you look at folks that are... a lot of folks that are struggling in AI B2B, if you peel the layer back, sometimes their agent isn't competitive. Sometimes it's too slow, right? A lot of issues. But if you really dig down, how many great FDEs do you have? How many folks do you have that can spool up a customer in 30 days? Well, I've got three. Well, you're going to lose.
Just staying on banter and private and just covering like the news item so people are up to. As I said, Lovable raises 500 at 5.5 billion. Excel led that round. Obviously, Harvey are better funded at I think 11 billion in the latest round. Decagon did a tender at $4.5 billion. Intercom raised 250 million at I don't know the price, Rory. You might know, but they didn't announce it.
Yeah, they're a large company. Yeah, they're... Yes, it's a debt round.
Great. Founders Fund... Founders Fund are closing in on $6 billion of new funds. They invested the last 3.3 in 11 months in amazing businesses like SpaceX and Stripe and Anduril. I'd be very happy if I was an LP. Base44 hit 100 million in ARR. Don't know if you have anything to say about that, Jason. Given Replit, does it save Wix?
Well, you know, I was thinking, um, we had on the agenda CrowdStrike crushes the quarter, trades down, right?
Yeah, I was going to move on to that with our prediction. No, but it's related to the Wix thing. And then I was looking at Cloudflare, which is on my top four, right? Cloudflare, uh, 27% growth a year ago, 34% growth last quarter. I mean, that's pretty good acceleration. Okay, that's pretty damn good acceleration. Um, 40% year-over-year increase in net new customers. Um, I think the public markets, you know, we're wondering what's happened this year and just the... and you had her on from Monday, a week ago on 20VC, and he said it, we've got to accelerate to get back credibility with the public markets. Accelerate, not manage a gentle deceleration, which was '22 to late 2025. Not manage a gentle deceleration with higher net margins, which everyone thought was the job at scale. Now the truth is you've got to be Cloudflare... it's not just Palantir, you got to be Cloudflare or better. And so I wouldn't want to be running Wix because 100 million from Base44, it's great and it shows you can cross-sell, right? And it shows a lot of new things, but their larger customer count is flat to down. So they've got the gravity of the decline of the core business. Like they've got to get back to so much growth that 100 million just... it's not enough. It's not enough. And we could... Rory could help trail out the math, but I think Base44, and there's some cannibalization there, it's got to be doing 500, 600 million to move the needle for Wix with its core growth decelerating. It's just so hard. And you know, for startups, you know, you got to start wondering this year when you should just give up on your portfolio. If they're not accelerating, at what point do you just give up? Because if the public markets won't tolerate gentle deceleration, right, how will the private markets tolerate it, right? It's just the era of gentle deceleration has ended. It's dead. We didn't think it was such a great period '22 to '25, but it was pretty nice that gentle deceleration was tolerable.
I mean, again, so much to unpick. First of all, to be clear, gentle deceleration is like entropy. It is the end state of the universe because everything decelerates to GDP growth, right? I mean, entropic is... but like the universe will be dead before some of it happens.
Agreed. So I just want to be precise because you know me, I'm that guy, right? What you're really saying is at 10x growth a year, Anthropic can afford to decelerate at 40% growth like Figma had it a couple of calls ago. Yeah, you accelerate to 30, you're going to take a little hit. But you're exactly right, Jason. What you're really saying is at 10% growth, you're now so close to GDP that you've got to reaccelerate. So that's really what I'm... just being precise, right?
I don't even know if the public market thinks it's okay that Figma might accelerate, right?
I got push for that because let's take CrowdStrike, right? Slight deceleration, 27 to guidance of 23... 27% revenue growth, guidance of 23. They'll probably beat that by two. You know, it's okay to grow 25% when you have scale. I mean, you know, Microsoft and Google are all going at 13, right? So yeah, like you just have to be objective. When you have enough scale, 25% is enough to matter, right? Not everything's going to accelerate all the time, Jason, because it turns out...
No, but the markets have been brutal this year. And I think it's not that you're wrong analytically or empirically, but the markets have given up on folks not accelerating. They've just entirely abandoned...
Well, yes, but again, agreed. I'm being precise. What they've done is they've said for three years I thought it was temporary and now I recognize it's permanent and now I'm going to value based on this growth rate, take a hit off you for technical obsolescence and the terminal value, take another hit off you for free cash flow, and I'm going to give you eight or nine times EBITDA revenues. That was sweet that I said that, wasn't it? Those are the good... No, you're exactly right. I mean, so they've all corrected... they've corrected to fundamentally what they're worth with no pixie dust. And that's just a tough place to be long term. You... I agree those guys...
The negative spin is they have to reaccelerate. The positive spin is even a small amount of reacceleration, you know, will get you some lift from here. But you got to do it anyway. Wix is a good example of that. It's funny. Let's try and take apart Wix. I hadn't prepared on this one, but I can't remember. The overall revenue is what?
