SpaceX's Financials Leaked: Is it Worth $2TN | Meta Debuts Muse Spark: Are They Back in the AI Race?

Jason Calacanis

I don't buy Dario anymore. He may well be the second-greatest founder of all time behind Elon, but I am just so burned out on the boy who cries wolf.

Starting off on the agenda, Anthropic unveils Mythos but withholds it from public release because it's too good at hacking, discovering thousands of zero-day vulnerabilities. Admittedly, some were quite old. Number 2, public software stocks tumble to new lows, with City saying there really is no flaw. Optimistic. And then finally, Meta debuts Muse Spark. It's Alex Wang's first model from Meta's Super Intelligence Labs. Does it save Meta in the race to catch up?

David Friedberg

So, I'm pretty bullish, actually, on OpenAI in the enterprise.

I think it's a 2-way fight. Anthropic has the advantage of clarity and focus. OpenAI has the advantage of the consumer business.

If your agents are only 60% as good, you're in a slow death spiral.

It appears to be the most expensive IPO at scale of all time.

The Elon discount rate is 0, and the Elon probability of failure rate is 0, to get to $2 trillion.

Ready to go. Guys, I am so excited for this show. As we always have, we're going to start with Anthropic. What else could we start with but Anthropic unveiling Mythos, with the preview withheld from public release because it is too good at hacking, discovering thousands of zero-day vulnerabilities? Admittedly, some were quite old. How did we think about this? Did it deserve the reaction that it got?

Jason Calacanis

Which reaction are you talking about, Harry?

I would say widespread fear, which was then shown in a loss of market cap for a lot of public companies in the US. Let's leave to one side whether it was a marketing stunt or whether they don't have enough compute. Let's focus on what Mythos does in terms of cybersecurity and what your correct response to that would be.

If you read a lot of the coverage, it finds a whole ton of vulnerabilities, including some that have been lying there for years. That's the “Oh my God, that's scary” part. Then you see—and that's why they withheld it and shared it with a bunch of security vendors—a bunch of counterarguments that are basically some version of: using older models and using them well, you can actually get to the same outcome. You can find the same security vulnerabilities. That's the counterargument. There were a whole bunch of tweets that said this is not a big deal, and I'm processing this from the outside. My conclusion is that those people who said it's not a big deal are wrong, and Anthropic is right.

I was thinking about the metaphor here today. It's actually very interesting in the context of the agentic revolution. It's kind of a microcosm that allows us to talk about a lot of things. Basically, the naysayers are right in that you can take an older model, point it at some of these issues, query it, direct it a couple of times, and someone actually did the exercise: “Here's how I found the same bugs.” I have to steer the model a little bit, and you get it. But the comment is that Mythos just kicks off on its own, agentically goes and looks at all the code, and finds them on its own.

The metaphor I was trying to look at here is very simple: it's the difference between a rifle and a machine gun. In one sense, both of them can kill someone, but one shoots 1 bullet and then stops to reload, while the other just spews bullets out. In the First World War, we all tragically learned that machine guns are a very different thing. It might be the same thing, but quantity makes a huge difference.

I think that's what's really going on here. The speed at which this can process and reason across large codebases means that it's just going to find more bullets. It's going to shoot more bullets. The cynical Twitter take that it's not that different isn't true, because it's the capability to do so much, so quickly, with so little human direction, that makes it definitely a quantum-step difference in terms of real capability.

My big aha was that it's not overblown in the sense that it can find stuff. I think AI is enabling every single breach possible, every security hole, to be found. Not a subset of the hottest companies, not just folks trying to attack OpenAI APIs, but everyone.

For example, the other day my fitness pal bought Cali. Cool story. What was it, Harry? $100 million? A 19-year-old kid from Miami, something like that? A great story. 2 days later, it was instantly breached. All the records were stolen: 3.2 million records. All the data on you, all your HIPAA data, every single thing on you was stolen within days, and it became a sport for a hacker. They just stole it all.

The root cause was—and this is surprisingly common—an issue Superbase and others had to deal with: they didn't have any authentication on Firebase. As more and more databases are now built by AI, and as more and more apps are built by AI, the number of issues is going to explode.

If Mythos and friends let bad actors find every site the second it launches with any PII and steal it, I think we may enter an era later where sites get more secure as it flips to the other side. But I think we're going to go through a transition phase where security is just getting worse and worse and worse, because every single website can be instantly hacked and stolen from.

In the whole Mythos run, they said—I think Claude said—it took them, I don't know, I'm sorry, I'm going to misquote the numbers, $20,000 of credits or a couple of hours or something like that. That's enough that I'm not going to do it against Scale's website. But if I could simplify that and distribute it against every single thing with any PII on it, bad actors are just going to hit everybody.

I think it's a big deal. Whether this is a publicity stunt for not having enough capacity, I don't know. Maybe, maybe a little bit. But everything's going to be found—every security hole.

David Friedberg

Agreed.

Which is why the second comment I'll make is that I thought the reaction to it in terms of security stocks going down didn't make sense. I'm like, what this says is there's no doubt that, going forward, part of the security process will be to use Anthropic or another code model to check your code before you deploy, to find these vulnerabilities. This will be a thing.

Someone's going to have to administer that. Someone's going to have to build frameworks and harnesses to do prescreening of code. But then, more importantly, everyone's going to have to operate on the assumption that if you miss anything, they're going to find it, which is different from if you miss anything and you're really strategic, they might find it. To Jason's point, if the other side now has machine guns, then you've got to build tanks.

What security is might change. The vendors who step up and meet the challenge will triumph, and the ones who don't will fall away. If Anthropic is saying that their model now allows anyone to find any vulnerabilities, and they're going to withhold it for 6 months, that means that in 6 months and 1 day every bad guy on the planet is going to be pinging your code and trying to find the bad bits. So you bet you're going to be investing in cyber.

I think the part that made sense was that this is a big deal. The part that didn't make sense is that cyber stocks should go down, because I think you're going to want way more defenses because the bad guys are more heavily armed. And, yes, as I say, it is an arms race.

Do you buy Dario's “It's too powerful; we can't release it to the public”? Is it just great marketing?

Jason Calacanis

I don't buy it anymore. I'll tell you something: one thing that changed for me with the Mythos thing, for what it's worth, is that I don't buy Dario anymore.

What I mean is, listen, he may well be the second-greatest founder of all time behind Elon. Look what he's done in 5 years, right? 5 years to $30 billion. The greatest grudge startup of all time. I mean, it's hard as a founder not to fall on the ground.

But I am just so burned out on the boy who cries wolf. Every job's going to be destroyed. Everything is insecure. Everything—enough already. I've heard it so many effing times. And about Mythos, I have to hear that he's created the spawn of evil if we're not careful. I just can't.

I've rotated back to Team Sam after all this, because I just can't take the endless boy who cries wolf. Even if you're right, there's a point where I can't open the Strait of Hormuz myself. I can't do this. Enough already. Let me just use my tokens.

Seriously, I've lost confidence—not in him as a CEO, but in this endless marketing machine. I'm tuning it out now. I don't care anymore what he says about this stuff. I don't care.

What specifically—what specifically do you not buy? I'm just trying to understand.

Jason Calacanis

Listen, if Dario says 80% of jobs are going to be destroyed in 2 years, we'll need no programmers by next week. That endless thing. Maybe he's right, but what can I do about it? I heard you. I heard you the 11th time. I heard you the 80th time. I heard you on Joe Rogan. I heard you on TBPN. I heard you on Harry. I just can't.

And then the Mythos thing, and they were holding it back, and it's like, I believe you're a safety guy, but if you talk your game too much, I just have to check out at some point. Show me something that's inspiring. I honestly feel like his message is uninspiring. That's the problem. It's uninspiring.

David Friedberg

I’m going to push back a little on that, but in the following way. I think a lot of the doom warnings are wrong, and the doom warnings to date have been wrong. If you look at the unwillingness to release GPT-2, which in retrospect was overdone, the concerns were sincerely held, right? It also is good marketing—I acknowledge that too—but I think the starting point is that there is a belief here that these things could happen.

To be very concrete, I think it’s totally wrong about the economic 50%. I think that’s beyond madness, and I’m not worried about it in the slightest. But I do believe—and I thought about this a lot—because what I realized is, if I’d met them at the Series C, which I didn’t because it was outside our price bracket, but if I’d met them, I would have done exactly what you did, Jason. I would have listened to the doom warnings and said, “That’s all silly and wrong. Therefore, I won’t do the deal.”

Jason Calacanis

Right. What I’ve learned is something more nuanced. I think a lot of Silicon Valley companies have a culture that’s overreaching. You listen and you go, “The grandiosity…” If you’re a grounded person, you reject the grandiosity, but what I’ve internalized is that the grandiosity is a rallying cry, sincerely held, because I don’t believe you can portray grandiosity consistently for 5 years if you don’t believe it—unless you’re really psychopathic, sociopathic, I should say.

Take, for example, Elon and, “We’re going to Mars.” The minute we had to file an S-1 and someone had to say, “You might have to go to prison if you say things wrong,” we said, “We’re not going to Mars. We’re going to the moon.” Right? So you could be cynical.

If I’d looked at that deal much earlier on, I would have said the cynical but incorrect approach would have been to say, “I don’t think they’re going to Mars. Here are 10 reasons. Therefore, I’m not going to do the deal.” The more evolved approach—and this is why I’m pushing back on Anthropic—is, “I don’t think they’re going to go to Mars, but I do think the Mars vision over 20 or 30 years is a rallying cry that will allow them to do amazing things in the short term,” which they’ve done.

And I think the same thing applies with Dario here. I think it’s all overblown. I think half of Silicon Valley, I’m going to say it here, is running around thinking they’re inventing the next thing after the atom bomb, and I simply don’t. I don’t think we’re going to unemploy 50% of white-collar workers. I think it’s madness. I think it has some legitimate dangers in cybersecurity and bioterrorism, but they’re manageable.

I think it’s all overwrought, but that overwrought—I won’t say hysteria—that overwrought intensity has allowed them to build a culture where it had no churn. It’s given them mission clarity, and then it’s given them a $30 billion revenue line and a possibly trillion-dollar market cap.

What I’ve learned—and it’s really hard for me, because I find all this problematic—is like Airbnb. Remember, Airbnb was the sharing economy, and Uber was the—remember the sharing economy? It sounded like a bunch of communism. We’re all going to sleep on each other’s air mattresses. That was a visionary fairy tale. What’s really going to happen is people are going to buy houses and rent them out, right?

Steve Jobs said it was “a bicycle for the mind.” It turns out we’re all just going to sit on our phones and watch Instagram and get depressed about other people’s lives. But the vision-oomphy bit just helps keep the machine of innovation churning, people.

So what I’ve learned to do, which is really hard, is literally listen to the idealism. Don’t say, “Do I agree or not?” Say to myself, “Will it motivate people enough to do something where there is economic advantage to be obtained?” And that’s a very cynical old person’s perspective.

But that’s the context in which I say I think Dario believes all that stuff, and I think it’s useful for them, and I think it’s wrong. But it’s damn useful, and it’s worked.

Jason, I get you. What do you want him to say, then? Like you said, “Oh, I don’t find him inspiring enough.” What would make you happy? What do you think he should say?

Jason Calacanis

I think before things got tougher, Sam was good at teasing at this. I want him to take us to Mars. I want to see the good side. Even Vinod, who is very direct that there are going to be a lot of job losses—right or wrong, Rory disagrees—but Vinod’s very direct. His point is it’s going to be okay, right? We will figure this out with AGI. Everyone will pay more taxes, even in California. It’s okay.

Marc Andreessen’s point is we’re entering an era of deflation and abundance, right? I don’t need too much on the other side. I don’t need the fluff, but I just need a little inspiration that there’s some good. I’m not saying—maybe the third hour of Dario’s speeches has it—but everything I see on social media feels like he’s an inverted Debbie Downer. And I’m tuning out. I’m just tuning out now.

And maybe in the enterprise, he’s got to—we’ll just see. Listen, I don’t run a $30 billion business. It may almost have to change as the years go on, as this works less well with the million-dollar customers, right? You may have to be more positive about the benefits in your workflow.

Jason, the odd thing is we’re actually agreeing on one thing: tune out the noise and, in the words of Haldeman, “Don’t look at what we say; look at what we do.” I always love quoting the Nixon White House as what used to be the most cynical White House we’ve ever seen, but we can come to that another day.

Ignore all the people saying, “Oh my God, this could eliminate white-collar jobs,” and wringing their hands and saying, “This is awful.” Look at what we’re doing. We’re shipping code. We’re shipping software. We’re doing $30 billion in run rate. “Oh, it’s amazing.” Turns out you’re not buying the guilt. I mean, you’re not buying the guilt. You’re not buying the hand-wringing. You’re actually buying the revenue, and the revenue is amazing.

Jason Calacanis

I’m with you.

Rory O’Driscoll

It’s pretty amazing.

I’ve spent a lot of the last year attempting to help founders understand that they genuinely need to move more quickly, that they are too complacent in their approach to AI, and that they have, at best, a 60% solution, a 60% answer to the problem. I’ve tried to vibe-code in public. I’ve tried to build my own apps. I’ve tried to share how we’ve rebooted our teams to 3 humans in 28. I’ve done all this, and I know it’s profoundly helped a lot of people.

I get so many messages. So many public-company CEOs and leaders reach out to me. Even I’m almost done with this phase. I’ve alerted you, okay? If, after me—with my 10 trillion tweets and 2,000 blog posts and 54 20VC pods together—you haven’t heard the message that you’ve got to catch up in AI, maybe I’m no Dario, but even I’m ready to move on to the new world.

I’m leaving the past behind. If we’re all going to live in a world of robots and AI lawyers, so be it. I’m ready to move on to the new world. I’m almost—and frankly, I’m ready to write off a lot of portfolio companies and a lot of public companies. It’s time to move on, guys. If you’re not going to get there in April 2026, then so be it. So be it. Let’s mark it down and call it a day. Good luck to you.

I’ll close with the Oppenheimer quote. All of these founders have their Oppenheimer moment. They want to be Vishnu, destroyer of worlds. They all reference the book, obviously, so they’re channeling Sam and channeling their kind of Oppenheimer moment with the new atomic bomb.

My favorite moment in that movie was when Harry Truman says, “Get that crybaby out of the White House.” In the end, Harry correctly says, “I dropped the bomb.” The equivalent of that is, if a whole bunch of people are fired, Jamie Dimon will fire them. He doesn’t need you wringing your hands with guilt, Dario. It’s okay, right?

Other people—and it was a great moment when Harry Truman correctly said, “History won’t say, Robert Oppenheimer, you killed all those people.” History will say, “You built the bomb.” And history will say, “I dropped it.” Right? In other words, get over your guilt. Ship the product in a methodical fashion. Do be careful. I do think he was right to keep that product back, but in the end, it’s not stoppable.

I appreciate that you—I actually think he’s very thoughtful about it. So I’m on his side on being thoughtful. I don’t think it’s fake, but this is going to happen, and other people are going to own the problem. Onwards.

Jason Calacanis

Onwards.

The final element, which is connected but not the same, is that Mythos was trained entirely on Amazon’s Trainium—

Rory O’Driscoll

I don’t think that’s correct, straightforwardly. What makes you—sorry, I cut you off there because I’ve had too much coffee, but whatever.

Jason Calacanis

It feels like a 3-cup morning for Rory, doesn’t it, Harry? What are you saying?

Keep—sorry, keep going. Keep going, because I’ll let you finish your sentence.

Rory O’Driscoll

Well, apparently Mythos was trained entirely on Amazon’s Trainium chips. Jassy disclosed that it’s now a $20 billion annualized business growing triple digits. Trainium is now nearly sold out, with Uber among their biggest customers.

The question is, if this is the case, are we slightly seeing a loosening of NVIDIA’s stronghold on the market? And does this change how we feel about NVIDIA?

Jason Calacanis

First of all, I think you need to be really precise here, because I checked—I didn’t know this point, so I checked it.

It sounds like you're saying a couple of things. It sounds like you're saying, “Oh my God, they're shipping Trainium chips to others who are buying chips and using them.” They're not. They don't, to a rounding error, have a merchant silicon business that is competing directly with NVIDIA.

What is true is that Amazon, instead of buying NVIDIA chips, is buying its own chips and then offering cloud-hosting services, inference services, and model-training services. So think of it less as, “Oh my God, someone's buying chips and competing with NVIDIA,” and more as Amazon not buying NVIDIA chips and instead using its own chips. Most of that runway is internal purchasing.

So what they're really saying is, Amazon is saying, “We have a capex budget of $200 billion a year this year,” which probably means about half of that typically goes toward chips. It's $100 billion. So where we can, we're buying our own chips, of course we are. And where we're not, we're going to have to buy NVIDIA, just like everyone else. That's true.

In terms of who's “a customer,” all they're saying is that when they're doing training runs for Anthropic, which I'm sure they are, or inference runs, which I'm definitely sure they are because that's a product they offer through Bedrock, they're running it on their Trainium chips. So yes, in that sense, some of the Mythos model was probably trained on Trainium chips, but not because Anthropic said, “Yo, I love Trainium.” It's because, to the extent that Amazon is offering them compute, some of that compute is Trainium. That's all that's happening here.

But, all that said, it's still $20 billion that didn't go to NVIDIA; it went to Amazon. So it is, at the margin, meaningful. It's 10% of NVIDIA's revenue, a little less than 10%. It's not a mega-competitor; it's just an in-house bundled product at some significant scale.

I'm loving this, Rory. Jason, aren't you?

Rory O'Driscoll

You know, 10% is material, right?

The bear case is just everyone's building their own chip or deploying their own chip. Everyone's trying, especially on inference, and that just means NVIDIA is dented. It's dented sufficiently to see multiple compression; it's dented sufficiently that our 401(k)s go down more, right? It's really just that when things are priced to perfection, there's a dent. Fortnite, maybe?

Jason Lemkin

Fortnite. NVIDIA may have its own Fortnite moment, as crazy as it sounds. It just may be the first 30 seconds of the game on a relative basis.

But no one knows this better than Jensen, right? No one is friends with his frenemies, partners, and friends of partners better than Jensen. No one's played this game in the history of mankind—being kind to everybody, pulling back, being right, understanding the dynamics, and still winning. He's crazy good at it, right? He doesn't get his dander up about this stuff like most of us do.

Rory O'Driscoll

And remember, I think Amazon and NVIDIA have famously had a difficult relationship. So they're probably the company most interested in not buying from NVIDIA. I think what you're saying is fair at the margin. It's $20 billion; to have 10% is not meaningless market share.

Yeah. But the big-picture comment is that compute is scarce, chips are scarce, they're pretty much sold out, the stock's at $194, and it didn't super-accelerate when they made that $1 trillion backlog comment, but it didn't go down either. So I think we're up against the constraint limit rather than anything else.

We're going to stick with Anthropic, but meanwhile, Anthropic is now competing with Lovable directly. They launched a competitive product recently, in the last 48 hours.

Jason Lemkin

Are you sure it was launched?

Well, they announced it.

Jason Lemkin

Okay.

Rory O'Driscoll

Harry, see, Harry's a media guy. He thinks when things are announced, they're real. Jason is a software guy. He actually thinks you have to ship product.

Jason Lemkin

No, it's all about the announcement. Surely you've seen that in the last few days.

As the Lovable-Replit guru here, Claude Code is clearly directly competitive with Cursor, and then you have this slightly different segment. I'd love your opinion, Jason, on the Lovable-Replit segment. What do you think the competitive threat from Claude, from Anthropic, is to those players?

Jason Lemkin

Well, look, Eric from Bolt saw these screenshots. Eric from Bolt said, “We all knew this was coming. I was at a dinner with the CTO of Level. It was just a question of when.” Then I asked him if he thought the screenshots were real, and he said it didn't really matter because it was coming, right? So that's a number 3 or number 4 player's view.

The meta-question is, if this were 52 episodes ago, I'd be like, “Well, it could happen, but it's not important enough.” They're going to have to get into databases, hosting, identity management, OAuth, and end-user support—consumer-level end-user support. It's a whole bunch of things culturally that they don't want to do, right?

But the pace of innovation at Anthropic is so intense that, on a whiteboard, it's hard not to want to grab a couple of billion in extra revenue from vibe coding, because you're just a database, OAuth, and a few other things. For them, it's 30 days of work, right?

The classic, old-school VC would be like, “It's distracting. Even if they launch it, they're not going to maintain it. They're not going to have support. They're not going to put in all the resources you need to maintain it.”

But the truth is—and maybe Eric from Bolt's point is that maybe if they don't directly compete at the prosumer level, even if they don't build a Base44, Lovable, or Replit, they might just go halfway there, and that might be enough. It might be something that developers use who just want to get something going. It might be something that more technical product teams use, which is the number-one highest-ROI category for Replit and Lovable: these product teams.

They may only target the nerdier, more technical part of the market, and that might be sufficient to, again, maim the folks. It doesn't have to be 100%; they don't have to replace shopping sites and stuff like that to have a material presence.

But Anton and Amjad and everyone think about this 26 hours a day: How do we stay ahead? How do we stay ahead? It may just be maiming. If it happens, it may just be maiming, but maiming hurts.

But Jason, those product teams could just use Figma Make, right? I mean—

Jason Lemkin

They can't use Make.

Rory O'Driscoll

Don't feed it. Now you're just poking the bear. Harry, please leave the bear alone.

Annie's off. Annie's off.

Jason Lemkin

Well, let's stay on this topic. It actually ties to software stocks and why I've become more pessimistic since the last show on them.

Oh, no. Why have you become more pessimistic?

Jason Lemkin

Because we're kind of ahead in this agentic thing, right? At least in the real world. I do talk to lots of teams—lots of senior product teams and lots of CEOs at massive companies—more than I've ever done in the last 10 years combined. Every week, there are multiple Zooms and multiple calls. There are exceptions, for sure, but I would say here's why I'm pessimistic and why I think the drawdown is accurate.

Even though I don't understand the public markets, I think almost everybody's building a 60% solution, and Make is an example. You look at what Claude did with prompting last November, and you spend the last 4 to 6 months building something that's kind of similar to what these products were like 5 to 6 months ago, but you can't really afford the tokens. You're worried about the cost. You can't build all the features. You're trying to use cheap models to bring costs down, and you're trying to limit it. You end up with Make, but everyone has a Make.

Why is Make so crappy? It's because you didn't care enough to spend the money or put the team on it. Make is a good copy of Replit or Lovable from last summer, right? That's what happens. By the time Make catches up to Replit and Lovable today, they decide it's too expensive, and then they decide they have to lock it down because they can't afford the agents and tokens. Now you're 9 months behind, or 12 months behind.

So what I mean is, it's not just that you have a 60% competitive solution. Here's the meta-problem for the incumbents: you can't charge for a 60% solution. If you could charge 60% of what Claude charges, or what Lovable charges, or what Replit charges, that would be great. That would be enough to charge 60%.

But a 60% product has to be free. It has to be included with your base charge. While we're recording this, for example, HubSpot is launching its next group of AI agents. I'm excited to try them, and I will be supportive. But if they're only 60% as good as standalone solutions, HubSpot cannot really charge for these things. You can't get away with charging another $20,000, $40,000, or $60,000 to a HubSpot customer if your agent is just fine and works in isolation.

When I meet with internal product teams at large companies at scale, they are so fucking proud of themselves.

They show me the agent they built. They show me their vibe-coding thing, and if I didn't use any other products, I'd think they were great, too. Or if this was 51 weeks ago, I would think these products were great.

They're so insular at their 2,000-person company, on their fancy campus wherever they are, with their mugs, bringing their mugs to the meetings. At their pace, they're failing even as they're proud of themselves with their 60% solution, because the market will not pay. They will use your 60% solution, but they're not going to pay for it.

They're not going to pay for it. And so you're stuck in this doom loop of, yes, I have an AI product as a public B2B company, but no one is willing to pay for it—for a 60% solution. They're not willing to pay 60%. And so we can say all these companies have moats, and ServiceNow has the biggest moat of all, so it shouldn't be sold off and this and that. But if your agents are only 60% as good, you're in a slow death spiral.

And that's what I see. I can't think of maybe 1 or 2 exceptions of everyone at scale where their agents are as good as either a standalone company or just what I can do in Claude. Now, I can't maintain Claude; there's a whole bunch of issues. But if it's only 60% as good, there's no way I'm going to pay this AE that just called me up $100,000 for it. I'm not going to do it.

It's not good enough. Checking the box does not work with agents. The check-the-box feature cannot be monetized in the AI era. And this is why I think they're all properly sold down, because none of them—they all have 60% solutions. All of them. All of them.

And they should be—it's do or die, guys, because you can't sell these things. You can't sell these things, right? And there are 2 counterexamples. We could argue over Agentforce, and the Base44-Wix one may not save Wix, which has repurchased like 30% or 40% of its company, but at least they made a bet that got them beyond a 60% solution, right, to 9 figures in revenue.

I agree. I think there's a lot to unpack on what's happening in SaaS, but I think I've internalized that Jason has articulated one of the big truths, which is: unless you have a product that's good enough to charge for independently, you won't have revenue acceleration of any meaningful scale. And if you don't have revenue reacceleration, then you're in a different valuation metric.

I'll talk about how to value mature companies with probably persistent users, but stock-based comp issues and no-growth issues. And you can do that. One of my rules is, price clears all markets. There's a price at which ServiceNow and Salesforce are, quote-unquote, worth something.

So zoom out a million miles: Jason is right. If you can't charge for it, you won't reaccelerate. And if you reaccelerate, you instantly move to another valuation bucket. I actually think I've listened to you a few times, and there's a clarity of simplicity there, because you can talk a lot about, "Hey, we're doing this or the other," but the gut-level test is: can you charge for it?

I really think it's a big freaking comment, right? Because I've been wrestling with what you use to sort out all these public SaaS companies and what the—how do you think about it? If you're in a very workflow-centric world for the last 2 decades, like Salesforce and ServiceNow, you are in an agentic world now. There might be other things that are more payments-related or stuff like that, where I think there's different dynamics. But if you're in that world now, you better have exactly what Jason says: agents that are worth the money, which means they do the work.

It's actually a very brilliant test. If I was literally looking at the horizontal software applications list for Morgan Stanley, Jason, you're right that it would probably be the first test. If I'm thinking about how to value this bucket of $1.6 trillion, that is the first test. If yes, then you're on the increasing-value scale and you can make it out. And it's still going to be hard—see Wix for details.

If no, then you're in the question of how to value a company with a mid-single- to high-single-digit growth rate at best. I think sales are very sticky, and you might be hit, and I think that's a separate valuation question, right? At 9 times, these things are trading now at 8 to 9 times cash flow. You might be hitting a point where just the money allows you to be a deep-value player.

The P/Es of something like Salesforce, excluding stock-based comp, are 11 or 12 times forward P/E when the market's at 20. This is unparalleled. So, if you want to make yourself a value play, money can still be made, but it's a grim way to make money.

The only way to still be a growth play is to pass the Jason test. That's my zoom-out comment here, right? Again, you said it in financial terms a few weeks ago, which is reacceleration, but today you're actually articulating the predecessor test. If you have agentic workflows, then you'll have reacceleration. Then you'll be in the Jason happy bucket.

And if not, then you'll be in the tragic value bucket, and you'll have to do hard things that would make Dario and Sam cry. If you're going to make this thing cash flow, it's going to involve SBC reduction. It's going to involve headcount reduction. It's going to involve a bunch of grim stuff. And you can probably still make money, but you're also capped.

Nothing magical will ever happen to a high-single-digit-growth tech company that's kicking off cash. At best, you build a mini version of IBM, CA, and the top 5 executives make money. It's not going to be fun. So, you're right, Jason. Would you buy ServiceNow?

Jason Calacanis

No, I wouldn't buy any of them. I wouldn't buy them because I'm looking for products that are more than a 60% solution. I don't know. It's just the world's moving too fast.

When we started this pod, I wasn't sure if I believed it or not. I was probably on the fence, but there was definitely a sense that the models might plateau, that there'd be parity, that they're all pretty good. They could all basically do a chatbot.

It's clearly not the case today. If we tie it to the beginning of this conversation, the models have radically accelerated their power since December, since Opus 4.5 and more recently, and we haven't used Mythos or whatever; it's going to be even more powerful. I'm not optimistic that anybody building slowly to last year's spec can compete.

It's too furious. It's too furious. And I also think that the problem with moats is they keep your customers in, but they don't lure any new ones in. No one's excited to cross the moat except the folks that want to breach the castle walls.

This whole moat discussion, I think, is at the edge of moronic, right? It's at the edge of, "Hooray, you have a moat, and your customer—I signed a 5-year deal." The average ServiceNow deal is between 3 and 5 years. So what? That doesn't bring in anybody new. It doesn't bring in agentic revenue. It just means I'm trapped.

Prisoners don't create growth, other than at the margin.

First of all, I love the moat analogy. That's great. What you're basically saying, because I've listened to this, is that Jason is not a buyer of any stock that's not a growth story, right?

Jason Calacanis

Right. I think it's good, right?

And I think so. I've come to the conclusion you were right on that. One of the reasons I enjoy doing this pod is when we argue, sometimes I change my mind, right? Let's play out the value track just for another few minutes.

I think the problem with the value play is, at this price, I think you make money on Salesforce, but unless they get regrowth, you're going to make single-digit returns, and the overall Ibson small cap is 11. So you're going to underperform. I actually think the other thing you're wrestling with in terms of these stocks is the following.

The weird thing right now is the public markets don't have access to growth on the growth side of software. They don't, so right now the trade is: sell SaaS, buy semis, which effectively means sell software and buy AI-making AI. What you don't have yet in the public markets is AI-native companies, starting with Anthropic and OpenAI.

Therefore, Mythos is a wonderful word, because in fact you're comparing the practical values of owning Salesforce with the mythical values of owning this company that's growing 10×, where you've never seen the actual financials. No one's seen GAAP financials, but, "Oh my God, it's amazing." Everyone's always going to want the myth. Pick your girlfriend-boyfriend analogy: it's the practical realities of the person you're with now versus the mythical example of something that could be.

My other aha is that these stocks aren't going to trade in the same fashion until 5 or 6 of these foundation models and 4 or 5 other AI-native companies are public. Then you can finally, as a public-market investor, say, "Okay, now I can choose. Do I want Anthropic, growing at probably 5×, at 30× revenues, with huge losses and big stock-based comp? Or do I want boring-ass Salesforce, growing at 10%, with 30% operating margins and, at that point, no stock-based comp losses?"

At least now you can have a choice. Then we'll find out how to evaluate those 2 things, and that will probably take 6–12 months after the IPOs. My point is, until then, what you're dealing with is the mythical desire for the as-yet-unrealized relationship—the practical reality of the person you're with now versus the mythical example of something that could be.

These stocks are going to trade this way until then. That's my big aha, because no one's going to be able to value them. The only people that get out of that mess now are the ones doing what Jason said.

If you manage to claw your way to growth as a public SaaS company, then you do actually have some kind of chance of trading up. But if you've got to trade on fundamental value, you're always going to be chasing this fear that the model companies can do everything you can. I don't think they can. I think when the model companies go public, people are going to realize, “Oh, yeah, Salesforce in its current form is probably going to be there for the next couple of decades. It's worth its cash flow.”

But Jason is still right. Unless it gets its act together and makes great agents, it's another IBM. When did you last even—I don't know the price of IBM, because why would you? I believe someone's probably made money, but it's not my problem, right? You just become a boring-ass old thing.

So, Jason, in summary, you're right. If you don't have growth, and Wix is interesting because I think they've done what you suggested, Jason. They're trying to get growth, and they're getting growth. And as yet, you're right: they did a buyback, and the stock's down 20% since then.

Jason Calacanis

They've done $1.6 billion to buy back nearly 30% of the shares. They bought it at $92, and the stock fell 23% on the week. Whenever that happens, you've got to say to yourself, “That wasn't a good week.” It's kind of like an inverse Bill Gurley: instead of selling stock and then it going up, you buy stock and then it goes down. That's like the IPO premium, but the other way. It's like, “Oh, my God, that hurts even more.” Talk about leaving money on the table.

And I think the aha is, to Jason's point, they did one thing brilliantly, which is figure out how to get 2 products out the door. Maybe they should have sat on their capital for another 6 or 12 months, kept it in reserve to maybe buy another AI product or invest behind their AI product, because I think these stocks are going to bounce around in value-trap land for a long time.

So now we may be wrong if they really nail their growth on the new product and the low-code and kind of vibe-coding product really accelerates, and you look back 6 months from now and growth's at 15% and the stock's way up. You'll go, “Oh, yeah, it was fine. It was just the next-day reaction thing.”

But the point is, fix the Jason problem, which is core growth, before you try and do financial engineering. And in Salesforce, don't fixate on financial engineering. It's useful, but it's not the solution. In the end, if all you can do is grow at 8%, you can screw with your balance sheet to your little heart's content. You're never going to matter a damn. As I said, you'll be IBM. They've screwed around with that stock forever, but nobody cares. You've got to do the Jason thing first and put all your effort on that.

It is tough that Salesforce did a big buyback too, right? I think they did $25 billion of debt to buy back stock. On a spreadsheet, this looks brilliant, right? We can support it. We wish the debt was a little bit cheaper, but this is the simplest thing we can do: retire a significant amount of our shares, drive up our EPS, and keep going with our agentic transition.

But arguably, the market at best shrugged it off. At best, it already priced it in before it happened. In the short term, there was no benefit. All this financial engineering that looks great on a spreadsheet amounted to nothing in the short term, but $25 billion of debt.

Jason Calacanis

I totally agree, H. Not only that, but I think one of the big advantages you have as a public company with positive cash flow is your financial flexibility. I wouldn't trade that for anything. Because, look, one of 2 things happens: if the AI wave rolls on without a blip, and you don't have a growth story as a public company, then your stock's going to be cheap 2 years from now. There's no hurry to buy it.

Yeah.

Jason Calacanis

Right. The second thing is, if there's a blip, then the guy in the market with a public currency and $25 billion in cold, hard cash can maybe buy 5 big things in the private market that allow you to compete. Maybe you buy not Anthropic, but a second-tier foundational model. Maybe you buy one of the big apps companies.

If you can afford it, Salesforce is one of the few companies that can make a big bet here.

Jason Calacanis

The point is this: that's why I would have kept my $25 billion in my back pocket, right? What are you worried about in the short term with your stock and the number of shares? The only people who care about that don't matter. You've got to win the war.

And there's some chance that in the next 2 years there's a blip in the market. All these AI companies are burning money. And if you're sitting there with $25 billion, you could have been saying, “Come to Daddy. I've got money. Let's talk about how I'm going to be an AI behemoth.” I think that would be a better use of your time and your money. Come to Daddy, I've got money.

On that topic, Alex Wang, founder of Scale AI, is obviously now at Facebook following the acquisition of Scale. Meta debuts Muse Spark, the first model from Meta Superintelligence Labs, which Alex obviously runs. The candid truth is, it did okay. It was decent.

My question to you, on the back of this—it was decent, not quite as good as the others, but good enough—is Facebook back in the game? Is this an encouraging sign for Facebook, where you feel more optimistic after it or less, given it didn't blow anything out of the water?

Jason Calacanis

I think it's a win. If you're not in the game and then you get back in the game, that's a win. So if you're fifth, you're in the game.

I mean, you read all the reviews. I haven't had hands-on experience with the model. I don't have the level of sophistication to evaluate it, but I did read a lot of the reviews from people who did. And you're right: the summary is good in some of the things that they were working on at Scale AI a year ago, and not so good at some of the newer things that the advanced labs have been developing for the last 12 months, which totally makes sense. You took your knowledge from a year ago and implemented it, right?

But leaving aside the question of whether it makes sense to be in this game at this level, if you decide, as Mr. Zuckerberg, that you want to be in this game, then you've achieved your mission. You spent $14 billion and you're back in the game. Now you've got to move up the league table.

I felt this was a relief because, had you done all this and then produced a Llama 4 that was disappointing, where people were saying, “It's not even credible,” then you'd have felt like a failure—and you don't. So I think it was a win.

Also worth noting is that they're talking about being much more closed source, which is significant. At some point, someone's going to need the American version of open source, and Llama was that. Now they're pivoting to being more closed source, which has implications across the ecosystem and is a bit of a bummer. But, yeah, I think it was a big exhale. I'm not sure why we're playing this game, but if we're going to play it, I'm glad we're not losing anymore.

Stepping back, though, to where we're going toward the back half of 2026, if I'm Zuck and I'm looking at the fact that Google owns its own models, which have become extremely competitive, that's one of my big direct and adjacent competitors. If I want to be in the big leagues, maybe I just have to own this. I'm not Apple. I don't want to be stuck buying tokens from Anthropic or OpenAI.

I'm Meta. I have one of the dominant consumer advertising and other platforms on the internet. I have the dominant social networks, and I sure better own this. It turns out it is not a commodity. Maybe this is way too much money down the drain, but this is core to our existential existence, just like it is for Google. And I don't want to wither.

That may not have been where this all started, right? But the goal of Facebook is not to provide API tokens like Anthropic. That's not the direct goal. It certainly isn't to encourage the open-source community to rise up and use Meta products. This is to stay in the top echelon of software, of consumer software companies, and it's worth $14 billion.

It's worth $14 billion, right? If you just don't want to become Apple and depend on everyone else's models, you just don't want to be that. And if you can just do better than the Kimi open-source stuff, et cetera, that Cursor is doing, it might be worth it just for that. Just to be in that zone between what I can rework purely from open source into what I can buy from Anthropic—if I'm close enough and this is my core—it might be worth owning.

Ruthless. This is as competitive a company as exists on planet Earth. This might be the most competitive company on planet Earth. Meta is ruthless.

Yeah, I mean, if you look at the big-play scoreboard for that, it's like: bet $1 billion on Instagram, win $100 billion-plus, maybe $400 billion; bet $70 billion on the metaverse and lose it all; bet $14 billion on this and clearly have a win. Somewhere in the middle between those 2 outcomes, you feel good. I agree.

The other thing—and again, it's not totally my expertise—is that when I look back to when we started this show, a lot of folks thought AI would kill Google Search and maim Facebook, that Facebook was dying as a platform, and that Google Search was of course dead because ChatGPT was going to destroy Google.

Fast-forward to today, and these are record-growth businesses. Google Search—we started the show talking about AI Overviews and other things—is a better business than it has ever been, as are Facebook and Instagram. So it makes sense to triple down there.

This is not a time to retreat. This is not a time to retreat for either of them. It is not a time to retreat. It is a time to get that lance out, go straight into battle, and knock those other guys off.

Jason Calacanis

Meta surpassed Google, I'm sure you both saw, as the largest ads engine in the world. I think Meta is at $243 billion.

I haven't killed it yet. I hope that stock price comes back. Speaking of ads engines, OpenAI projects $2.5 billion in ad revenue for 2026. The ads pilot was at $100 million annualized in just 6 weeks, with 600 advertisers. We touched on it before, but they're guiding to $11 billion in 2027, $25 billion in 2028, and $53 billion in 2029. Is ads the great comeback for OpenAI in H2 2026? Is this the shining light that we should be directed toward?

Jason Calacanis

It's both obvious and inevitable that a consumer product like OpenAI's ChatGPT is going to have to be ad-supported. So yes, they're making exactly the moves you'd expect. The little chatter of, “Oh my God, that's a bad idea,” or, “It's not doing enough,” that we saw a few weeks ago—it's all over. This is just going to happen.

Three people have done it at super scale already. Google did it in 2004. Meta themselves did it in 2007 and 2008, and then in 2012 in mobile. We always forget Amazon. I think Amazon got to about $100 billion-plus in ad revenue over the last 5 or 6 years, or probably 7 or 8 years at this point.

The move is clear. They're talking about getting to $100 billion in 4 years, and it all makes sense. The funny thing is, I looked at all that and thought, “Yep, that tracks.” I'm mentally giving them 100% credit for getting that OpenAI is typically not aggressive in its projections, so let's assume this projection is aggressive.

This is going to sound really awful, and I hate even saying it: $100 billion is amazing, but it's not enough relative to the market cap. In other words, the big hope for me is that you can build a $100 billion ad business in ChatGPT, which makes sense in the context of a total of $1 trillion in ads, with Meta already having $300 billion of that, Google already having $200 billion in ads, Amazon already having $100 billion of that, and TV having to eat too. That's probably a realistic high-end estimate.

What it means is that you need another $100 billion-plus from your enterprise business. That was the big aha for me: consumer alone isn't going to be enough to feed this beast, because your competitor who's all-in on enterprise is already at $30 billion. It may be that corporations want to buy more intelligence than consumers do, and that $100 billion in consumer ads won't support your burn. You need more.

The $100 billion is what, 15% of ad spend?

Jason Calacanis

10%. 10% at a trillion worldwide.

10-something, 15%. I like this as a goal for 2030 because it's clear what people should be doing. We can't just add $100 million of ads to ChatGPT. We have to build something that's essentially as big as several of our competitors.

It's well understood why it works. We're not directly in commerce. We don't have the advantages that Amazon does, but can we achieve the scale of Facebook and others? Yes. This is our job, guys. Every week, we're going to iterate on it, improve it, and make it better.

It is mathematically possible. This is not as aspirational as the enterprise stuff. This is our effing job. We're going to review it every week, and we're going to put some of our best team on it. It's doable.

If it comes up short by a year or 2, like Elon—I mean, it would suck for the IPO, but it's doable. It is achievable. Because it is achievable, I think it will be achieved. I actually think it will be achieved.

Jason Calacanis

I think you're exactly right, and it's clarified a whole bunch of things they've been lacking.

Clarity on the enterprise side. I will say one thing: there was this memo from this week that leaked from Denise Dresser, the CRO and president, saying, “Well, first of all, Anthropic is overstating their revenue. We're still ahead.” She was also saying, “Hey, we have the capacity. They're out of capacity,” and so on.

At first blush, this memo seemed like a flashback to something Mark Benioff might write, whom I love, but it seemed more appropriate for Salesforce than for OpenAI. Her last gig was CEO of Slack. At first blush, I thought it seemed out of place at an AI leader, but then I thought about it and realized this is the exact type of messaging you want to win traditional enterprise customers.

So I'm pretty bullish on OpenAI in the enterprise. All this enterprise DNA they have probably didn't help them an inch over the last 12 months, but in the future, when all the models are so powerful and big enterprises are trying to decide between a couple of top brands, I think the ability to sell this directly to the enterprise, versus coming in from the bottom, coming in through the CTO, or coming in from functional groups, is going to be very powerful.

OpenAI said they're going to double in size, and a lot of that is around selling motions to the enterprise. I think it's going to work. I think they're going to run a lot of the traditional playbook, which is going to work better and better in 2027 than it did when the last year was “ask your developer.”

That was the land, and that's why Anthropic won: “Ask your developer. I don't want Copilot; I want Opus.” Your developer picked it for your app. I'm using the old Twilio mantra because it worked for years until it didn't, and I think it's going to work for LLMs until it doesn't, until you go deep into traditional enterprises.

If OpenAI is going to be the number 1 or number 2 brand for as long as we do this show, I think they may be able to outsell it versus, “Hey, the world's going to end from Dario. Thanks for letting me in the lobby. Everyone's going to be unemployed next week.” That one may cast a chill in the CIO's office. “Thanks, guys, for having me. Most of you won't have a job next week.”

Jason Calacanis

I would have 2 comments because I'm not quite in that place. Maybe 3 comments. The first is the aha I had from this: the comment on compute is the ball game. Whatever you think about the long-term overinvestment—and I still angst about that—there's no doubt in my mind that right now everyone is compute-scarce, and there's going to be no restraint. No one's going to blink on their investments for 2026.

You actually know the next 4 quarters of NVIDIA announcements. You know the next 4 quarters of every one of the infrastructure vendors, which is every single thing we can make. If you can count the wafer starts at TSMC, you can predict NVIDIA's revenues. Everything is going to be sold out from now and for the next 12 months, because if the rate-limiting constraint is compute, everyone's just going to buy compute.

That's the first big thing: compute. They've got an interesting position relative to Anthropic in that they were more aggressive, so they have more compute.

The second consequence is that you're going to see some compute rationing. People are going to start allocating tokens to, effectively, the highest bidder. You're going to see a lot of throttling. You're going to see plans like that—that's why they shut Sora. You're going to see some of the Claude plans get throttled down. That's what money is for: to allocate scarce resources. It's actually the definition of economics, the study of the allocation of scarce resources, and they're going to start allocating those scarce resources of compute via price.

The second big trend is that I don't know if you're going to see that level of flip from, “Oh my God, it's all Anthropic,” to, “Oh my God, it's OpenAI.” I think it's a 2-way fight. Anthropic has the advantage of clarity and focus. OpenAI has the advantage of the consumer business, and they're going to have to slug it out.

I think Anthropic, just in the way they've played the last 12 months, has a slightly better lead right now, both in terms of perception and in terms of developer friendliness. But OpenAI isn't going to roll over and die. I want to come back to a point I made earlier, which I've only processed through now. Sometimes it's important to say the big thing very clearly.

If you look at Google, Google's consumer business and ads are 2/3 of the value, and the cloud is maybe 1/3, roughly. The big money has been in consumer. Bear with me when I say the obvious, but consumers actually don't want to buy AI; they want really great ChatGPT. There might be some models like friends and companion models, but fundamentally, when I go home, I want to buy Netflix. When I go to work, I want to buy intelligence.

It may well be that the zoomed-out comment from all this is that enterprise is 2/3 of the ball game in AI and consumer is 1/3 or less. That's the flip of last time, because you would have thought 2 or 3 years ago, when ChatGPT exploded, that was a really great launching point. But if enterprise is the better place, then yes, you're going to have a good consumer business, but the ball game may be compute. From a customer's perspective, the ball game may be 2/3 enterprise and 1/3 consumer, which is the mirror opposite of the internet.

And that's just one of those big-picture comments, because I realize I don't go home and want to do cognition. I go home and I want Netflix. I go to work and I want thinking, and they're selling thinking. It's an enterprise business.

By the way, you get into a really fun and interesting discussion, which is above my pay grade, but I started to see comments about it: Do the things you do to make a consumer model consumer-friendly—the happy version—mean that the same model continues to work really amazingly well for both? If the enterprise wants clarity and concision, if the consumer wants a little friendlier answers, do you really find that divide? Does the tone and the persona of the consumer model and the enterprise model start to become different? I don't know if that has implications. It's above my pay grade, but it's in a category of something to think about.

I think very much so. And I think you've seen it in consumer research studies. They've seen that, actually, younger people like OpenAI because it's much more supportive of their emotional challenges.

Jason Calacanis

And you're exactly right. In business, I don't want support. I want to be told, “These 5 things are good and these 3 things are bad.” When we're doing investing, we don't want supportive; we want a decision. It's almost the exact opposite. I want harsh critique: “This is a stupid deal, Rover. You looked at this 3 years ago. It was dumb then, it's dumb now. Stop, you idiot. Here are 5 fact-based reasons why this is wrong.” And if you give that in a consumer app, you're not going to have great lifetime value or retention.

That was interesting. Aaron Levie had a tweet this week about his latest roadshow meeting with CIOs from Box. He talked about how so many of the CIOs meeting with him now are token-maxing. What he meant was that they are creating fixed token budgets—really, dollar budgets rather than numerical token budgets—for the coming year, and they're making the departments fight it out per project.

Now, this is where I think the game is going to change again with the leaders. When most of the budget is essentially rogue—when it's developer teams picking Anthropic almost universally last year, and you're giving them the budget because the output is so intelligent—you’re finding budget for that team. Or you're using discretionary budget to bring in little agents. It's one thing.

When the CIO takes control again of how many tokens are used across a large enterprise, that's a very different calculus of which vendor I choose from. And if the CIO prefers to buy OpenAI because it's got a more traditional sales motion, it's packaged better, it's better used, there will be exceptions. There will be exceptions in departments, and they will get exceptions. But overall for the enterprise, I'm standardizing on OpenAI for 2028. It is the right choice for our Fortune 2000 company.

It's just a different world when, whatever the $60 billion in OpenAI and ChatGPT revenue, all the enterprise stuff is still in some sense rogue today. So much of it is rogue. It is out of budget. It is out of band. And as that changes, it will change which vendor we buy from.

Jason Calacanis

If you are correct—and I'm not sure you are—what it says is that the people who should go into couples counseling are Microsoft and OpenAI, because they need to get their relationship back together again. Who is the dominant path to every single enterprise in the world? It's Microsoft. They need to get some couples therapy. They need to accept that they're different, that they have differences, but they can reconcile, and they need to start, because otherwise they're going to get their clock cleaned. Frankly, I think that's an indulgence that OpenAI can no longer afford.

You're right that the other guys have stolen a march on developer love. Microsoft can go top-down. I don't care about your long-term competitive dynamics; you need to kiss and make up here, people. If we agree that enterprise is two-thirds of the game, Microsoft is the key. Figure it out, go to therapy, talk through your issues, and get this thing back together again.

And then the other comment to make is I just want to give an advert for Aaron. I read his tweet last week about his comments from the roadshow, and look, I said this before: We were lucky enough to back Aaron 16 years ago. Many years—in fact, last year—we had him back at our annual meeting just to talk. Aaron is a walking investment insight. I read his tweets and I'm literally sending them out to the group and saying, “This is the latest thinking on what you should be thinking about, about what CIOs are thinking about. Just read this, and then you'll know.”

Jason Lemkin

It's such an insightful tweet about agents. He's not a doomer on employment at all. He's like, “People are going to be rolling this thing out, and if you understand what they're doing, you will have a role here.” And he has a really good feel for what's happening. He talked about token-maxing. Beyond Silicon Valley, there are going to be meaningful budgets. There are going to be constraints. And this thing gives you a real sense of how CIOs on the front line are rolling out AI. I just think it's excellent. So I follow him all the time.

I thought it was so excellent. I actually messaged him and said, “Do you want to come on the show this week and do it?” And he said, “I would love to. Let's do it tomorrow.” And that was this morning. So I'm actually doing a show with him tomorrow about this tweet thread.

Jason Lemkin

He's in that rare combination of, frankly, being grounded enough—in terms of 15 years calling on CIOs—to really know what they think, and at the same time being, frankly, young enough and flexible enough to really understand what AI is doing. It's literally like an investment insight. One of my colleagues even said it when he spoke last; it was like having an investment memo spewed out on enterprise AI. Good for you.

Yeah. Here's the one tough thing, and I almost don't want to say it because I love—

Jason Lemkin

You can say it.

But if he can't reaccelerate Box with his incredible depth—10 out of 10—if he can't reaccelerate, what hope is there for so many other leaders, so many other unicorns and others? If he can't get Box to 20–30% growth, I'm giving up on the rest of the world.

Jason Lemkin

It's a totally fair comment. I've given up.

Absolutely.

Jason Lemkin

And I hope he can. Look, I admire them so enormously, and I really hope they can reaccelerate, right? Because I do think—

Right, because I do think he's—

Jason Lemkin

He knows everything that's happening, and he's not pretending. He's not like so many folks who are pretending. He's as engaged as he's ever been, right? As stressed as he's ever been. He can't work any harder. If he can't get this reaccelerated, good God, who the hell can? Who the hell can?

We had the financials for SpaceX leaked: a $5 billion loss on $18.5 billion in revenue. The reason it's so important for the audience is that there are so many endowment funds and managers who hold SpaceX in some way, who are awaiting the IPO later this year. The loss is driven by the xAI acquisition, not by operations—important to add. But the $2 trillion potential IPO is a big number, and the math needs to be worked out.

Okay, keep going. So, $18.5 billion in revenue at $2 trillion is 108×. Did these numbers change your perspective? Did they confirm an opinion?

Jason Lemkin

They didn't change my perspective; they confirmed it. But let's break it apart a little bit. The first is—okay, technical accounting on you now, right? When did xAI close? Because pooling accounting is gone. Pooling doesn't exist anymore, where you retroactively recast the financials as if the companies were together. So that's only the loss from when xAI was acquired, and I think it was late last year or early this year.

So we actually don't know the actual run rate, right? Do you understand me? In other words, if the deal closed on October 1, then that would only reflect 1 quarter of loss. So anyone hypothesizing on its profit, or its profit excluding that, or that it's only—quote-unquote—only a $5 billion loss, until I see the actual GAAP financials, I don't know. Let's start with—

It could be $20 billion. Right, exactly. $20 billion loss. But I think the rough trajectory of it will be something like the following: we have an amazing launch business with a near monopoly on cost effective launch and that price could come down with the next generation rocket. We have an amazing business on Starlink. I think the whole xAI, in retrospect, I think you'll look back and go, I'm not sure I would have paid $250 billion for xAI. And then the question, as you say, is how do you value that relative to 100 times revenues? We've talked about this before. I'm not going to say it's right or wrong because very few things trade at 100 times revenues for any extended period of time. Let's just say that. So it's clearly underwriting a level of growth. It's underwriting a reacceleration of growth even of the Starlink business, which is plausible based on the future things they're doing, but feels like a lot, he said gently.

It appears to be the most expensive IPO at scale of all time.

Yeah, it appears to have no one at scale that has IPOed has ever IPOed at a revenue multiple approaching this. The case will obviously be all the future things. You know what is space, and you read the bull case. As I say, I'm trying to avoid the 'I don't believe it' and express my concern rather than saying, 'Oh, I think that's crazy.' I just say you have the existing business and then a series of new initiatives around direct to cellular and, obviously, data centers in space. You can articulate a massive market, and the question, as I've said before, is: so you take these adjacent things. And if you give them a 100% probability of happening and a 100% probability of happening right now—in other words, no NPV because it takes 5 years to make it happen—then you probably get to $2 trillion. If, on the other hand, you apply a probability of it not happening and a time value of money, you get to a lower number. Maybe that's a good way to reduce it.

Jason Lemkin

The Elon believers are saying, “These are the future things, and I ascribe a 100% probability of success.” It's like, I'm going to give them credit for today, even though it's going to take 3 or 4 more years. It's basically the Elon discount rate: the Elon discount rate is zero, and the Elon probability-of-failure rate is zero to get to $2 trillion.

If you put a more conservative number in both of those, you probably end up in a different place. You still have the upside. You still have the long-term story, but are you getting paid for the risk? That's a way of framing it. It's not, “I think it's silly because 100 times revenue is just too much.” I think that's a reductionist argument.

What you're really saying is, “I'm looking at all this future time and perhaps being more sober about the probability of it happening.” I'm revising my prior from, “Oh, 100 is crazy,” which is too simplistic, Rory, to, “What are you saying about these other markets?” When you feel that it should be at 30 times revenue, you're effectively saying maybe it takes 4 years for the data centers and direct-to-cellular to happen, maybe the discount rate for that is 15%, and maybe the probability of success is 70%, not 100%. You have to get paid for the risk.

Jason, what topic do you think we should discuss that we have left?

Jason Lemkin

Some private stuff.

There's some private stuff.

Jason Lemkin

No, some private market stuff. Simple, humble venture.

The first, for what it's worth, is the topic I put, but I actually don't think it's as interesting as the larger topic: AppLovin's 898 employees. This is not a brand-new AI company. Last week, it had $4.5 million in revenue per head.

I've been thinking a lot about this. You have the Block memo and what Jack Dorsey wants to do, and every Andreessen chart of the week is showing how efficient the next generation is, right? ElevenLabs and everyone else is so efficient. The meta thing—my captain-obvious learning from all the conversations I have—is that it's a choice. Everyone wants to be small by choice, and this is what I think is going to be disruptive for the next year and a half.

As VCs, you get really excited when you see an efficient company because, all things being equal, they don't need to raise as much, I'm going to be diluted less, and it's less risky. Everyone wants to invest in the next version of Veeva: we raised $3 million and got to $30 billion. That's the venture dream, no matter what anybody says. That's good.

Jason Lemkin

We raised $3 million and we're worth $30 billion. I don't care what you do. Absolutely. Whether it's bagels or health care software, $3 million.

So we want that, and we see hints of this in the employee thing. It's not quite that simple when the gross margins are lower, but what I'm seeing everywhere is that everyone just wants to be smaller by choice. AI is an enabler because it lets my best engineers do more. AI is an enabler because I can get rid of those SDRs.

Rory O'Driscoll

Jason, I actually tweeted last night, “The core test when evaluating a team is knowing what I know now about the person, having worked with them, would I hire them again?”

Jason Lemkin

Yeah.

Rory O'Driscoll

And you said that's not the question. What did you say was the question?

Jason Lemkin

Would I replace them with an agent?

Rory O'Driscoll

It's the same thing. I'd rather have an agent than a mediocre person. Everyone thinks that. Not everyone says it out loud, but provided it wasn't a mediocre agent, as you've articulated earlier.

Jason Lemkin

Yeah, but I know how to build a good agent now.

Rory O'Driscoll

Okay.

Jason Lemkin

Everyone in 18 months will figure out how to build an AI agent. They don't know how to today, but agents will get easier and easier to train. We touched on this briefly, but you don't need prompt engineers anymore, right?

I built, for fun, yesterday on Replit, a fully functional website in about 6 minutes that has video, audio, and everything. My prompt was, “Create a whole website around my theme of the recycled mediocre in this post.” Recycled mediocre is when you keep hiring the same mediocre folks again and again.

It did the whole thing from that prompt. It created the whole site, pulled up all the context, created an incredible horror image, then created the video out of it, then created the connection, and then pulled up all the context. My point is, you don't need to be a prompt engineer. The most mediocre prompts in the world do magic now.

Right now, getting an agent to work requires an FTE, and it takes weeks or sometimes even months and lots of training. That shouldn't be true in 18 months, right? It should be as magical as prompts are today. I think we're all going to be good at agents in 18 months, and we're all going to say, to Harry's point, “Do I want to work with that person again, or would I rather replace them with an agent?”

We're going to choose to be leaner. It's not about the money; we're just going to choose to be leaner for many reasons.

Rory O'Driscoll

I think you're right on the directionality. I just want to make a business point, which is that the simplistic revenue-per-employee metric is not useful because you can't compare the efficiency of a company like Cursor, which has a very low employee count but massive gross-margin costs, with Salesforce.

Cursor does $4 million per employee. Salesforce does $700,000 per employee. Cursor must be more efficient. Well, it turns out Salesforce has 30% operating margins and Cursor is losing a lot of money. Why? Because they spend a whole ton on tokens.

You can compare companies in the same business on an efficiency metric, but the one-size-fits-all revenue-per-employee metric doesn't really cut it.

That said, 2 comments. One is that AppLovin is an amazing business because it has fairly high gross margins and a low employee count. It's one of those businesses where you just have to go away and understand how it fits at that interstitial moment in mobile ad networks.

At this point, given The Trade Desk's downturn, it's the most successful ad network business by far. We could digress onto why that is, but it's a one-of-a-kind business. It's a $4.5 million-per-employee business where, unlike Anthropic, there are no capex costs and, unlike Cursor, no token costs. It's just a money-printing machine. I'm jealous.

Jason, the second comment is that you're still exactly right: the trend everywhere is to grind down the headcount. Do you need them? What can be automated?

Maybe, reflecting in my own mind, the way to say it is that it's not fair to say to a SaaS company, “AppLovin does $4.5 million, Cursor does $500,000.” What is fair to say is, “Last year, you were at $500,000. This year, you better be at $600,000 per employee because Jason's telling you to, and next year maybe you've got to be at $800,000.”

I do agree that if you're not making progress on that metric as a software company, you're not with the program. On that basis, I think you're right.

Obviously, I'm just sapping off the knowledge of smarter people than me. Would you buy AppLovin today?

Rory O'Driscoll

I don't know now. You always worry that ad network businesses, over the medium term, get ground down, but it has been able to survive for the longest time. It's gotten rid of its gaming business and is purely focused on this.

For some reason, it's found a way to exist in the Apple ecosystem, with all the privacy issues, as the only way to do some of this targeting. I need to spend a lot more time thinking about it, but it's been an astonishing run for them. It's probably the biggest standalone beneficiary of mobile ad networks—clearly mobile ads, after maybe Meta and The Trade Desk. Amazing win.

Jason Lemkin

There are 2 things I wanted to touch on. One was Thoma Bravo shutting down the growth-equity business. Is this foreshadowing a lot of other growth-equity businesses shuttering, or is this just Thoma Bravo being independent?

Well, first, before you guys answer, can you educate me? I'm ignorant. I don't understand the whole Thoma Bravo empire and what it truly means that they're shutting down growth equity versus the other vehicles. I don't understand it.

Rory O'Driscoll

I think it's pretty straightforward. The core business, which puts 90% of the money on the table, is buying control positions in software companies, with some leverage, and running them—adding and building, adding on other companies to them, and ultimately selling them either to another private-equity buyer. That's most of what they do; it's 90% of the money.

They started doing non-control and minority positions in late-stage, high-growth companies, and it's just a different business. I think it's different enough from the control business. In a control-position business, you're trying to buy value. We may look back and say many of the prices they paid for those control positions in 2021 and 2022 weren't value, but you're trying to buy value where you have control, you're going to be EBITDA-positive, and you're trying to pay down the debt and do all those things.

You know, classic venture growth. You're still hopefully growing 50% to 100% minimum, per our discussions. You're probably still losing money. You're not in a controlled position as the PE investor. And in a period like right now where your core business is threatened, the first rule when threatened is that you retreat to the core.

PE guys regularly come into growth venture at the top of markets thinking this looks easy, and they regularly retreat from those markets when they discover it's hard, right?

Shouldn't they be going back in in 2026? Isn't this a time to be re-entering?

Rory O'Driscoll

Possibly. But, again, 2 reasons why not. One is your core business is under threat. The 1 thing you don't want to be doing with your investors is saying, “We have this thing that 90% of your money is in, but we're fussing around with this other 10%.”

Even if it's a good business, it doesn't matter, right? We gave you $10 billion to invest in control positions in software companies, and we gave you half a billion dollars to screw around doing something else. You're having problems in your core business—how about you fix that? It doesn't even rise to the level of, “Is it a good opportunity or not?” It's not the opportunity we have.

You said retreating to core, Rory. I completely agree with you. Core is business. I'm sorry, I'm never down on businesses, but challenged businesses like Coupa, like Anaplan, like Medallia. I mean, I would say, can you help me? This feels like a cluster of pain. Separate comment.

Rory O'Driscoll

Yes. I mean, I think, look, one of the things is that they're the same type of companies as were hit hard in the public markets. So the same discussion we had in the public markets applies here. In other words, these are mature, plain-vanilla SaaS companies with single-digit growth rates, right? They're just not traded every day in the public markets; they're traded in the private markets.

So the question is, what does that mean? The first thing is, these kinds of companies are trading at 2 to 4 times revenues, right? The equivalent companies in private are probably valued today at that, and many of them were bought at 10 times and have leverage. So that's a pretty tough place to be.

The negative spin is, if you apply the same math of 3 or 4 times revenues and then deduct the debt, you have little or no enterprise value, right? And that's terrifying. That means you could see big losses in some of these PE funds.

Now, the positive spin that they would give, which kind of goes back to Jason's thing—I’m not sure I fully believe it—is, if these software marks in the public markets are totally wrong and then 2 years from now they're back to 8 times, then it'll be a tree that fell in the forest: no one will know, and in 2 years' time they'll be able to go public with Anaplan again at 8 times. Maybe that happens and maybe not.

If I was articulating to Thoma Bravo why it's going to be okay, the 1st argument would be that it's way overdone and these things are really worth 8 times revenues because they're profitable. In the end, things trade at 15 times cash flow and not 9 times, and we'll all be okay. That's 1 argument.

The 2nd argument could be some version of the Jason one, which is that sometimes the advantage of private ownership is acute clarity. If you're going to make the transformation-to-AI bet, I bet these guys are going to articulate that we will make that happen because we will own these things. We will replace management if they're not capable of doing it. We will hire other people who can do it. Maybe we'll buy assets.

I'm not sure I buy that, but that's probably part of the argument, which is PE's argument has always been transformation. To date, the transformation has been about cutting costs and being more efficient. I don't know if they can pull off transformation where transformation is, as Jason says, taking Wix and adding a 60% agent. As a privately held company, you haven't passed the Jason test.

The question you have to ask these guys is, can they add a 100% agent that you can charge for? If they can and they rekindle growth to 20%, then they'll have earned their massive carry. If they can't and these things don't bounce back, then you're right: you could have a train wreck. That's their ball game right now, right?

That's why they're all looking for AI experts. It's why, on a going-forward basis, they're pitching buying new companies and AI-enabling them. But for their existing portfolio, it's all about, what do you add to Coupa or Anaplan to make it AI-forward—at least not AI-first?

On the 1 hand, I think we're going to look back on this and see it's all a shame. If you have 10,000 happy customers—50,000, 100,000, 150,000—who are reasonably happy, not thrilled—

Yeah.

Rory O'Driscoll

—but reasonably happy, and you've had 12 to 18 months to build them an agentic product, you've had access to the LLMs, and you've been able to carve out 50 of your best engineers to work on it, you didn't take advantage of your installed base for real. Not in the moat way, not prisoners, but I mean, if you didn't take advantage of the fact that 90% of your customers are not at the bleeding edge of AI and sell them an agent, this is such a missed opportunity for the leaders. It's tragic.

It's tragic because most folks have not made their decisions on agents. We're going to look back and see these teams were so mediocre and so paralyzed. I’ve got to tell you, when I talk with folks who are so out of ideas, people should not be asking me what they should do with their agents. They should be showing me their agents and asking me for constructive criticism on them.

People are paralyzed with fear. They don't want to work twice as hard as they used to, and they don't know what to build. It's a tragedy because even today, selling to the installed base is much easier than finding a new customer.

It's not too late to sell to your installed base.

Rory O'Driscoll

Yeah, just call them up. They will take the meeting. This is the great tragedy. I think a lot of private equity firms are probably pretending their playbook's going to work: “I'm going to hire this AI expert from Stebbings, O'Driscoll, and Lemkin. It's $2 million a year. They're coming in with their ties and their checkered shirts, and they're going to teach us how to do AI and get us to a 60% solution by the end of the year.”

It slips a bit, and it's a tragedy. I'll tell you why. It's a triple tragedy.

This is something I didn't know. Granted, in my brief tenure at Adobe as a VP—and that was not a high point for Adobe—it was during the transition to the cloud. But I will tell you what I learned at Adobe that folks don't realize.

They said the same.

Rory O'Driscoll

I never underestimate competitors or big companies. I’ve got to be careful, right? But I will tell you what I learned at Adobe that folks don't realize: there were 100 super-amazing engineers, either by design or accident, who were working on projects that weren't quite going to get there, or they were available.

Sometimes they were on the back half of their careers. Sometimes they weren't quite as cracked as the top engineers at Replit, Lovable, or Cursor, but they were great. My CTO, the toughest critic, would actually say, “Let's go steal these 5 guys. They're actually great.”

They exist.

Jason Lemkin

So it is a crying shame that you can't take a team of 6 at Coupa, at whatever, build the world's best product, and ship it to your 10,000, 20,000, 50,000, or 100,000 customers. We're watching tragedies in the making, and it's sad.

It's sad because deep down, they're still running the dated playbook of a big release every 4 to 5 years and a quarterly release that changes a few pixels and adds some workflow. Deep down, all the companies I've talked to are still running that playbook, and it's a tragedy because they have the opportunity.

But Stebbings, O'Driscoll, and Lemkin AI consulting is not going to get them there. That's who PE firms want to bring in. These guys aren't going to make it work. Everyone is right: they have the base, and they have the opportunity.

They're going to end up in these Medallia death spirals where they're defaulting on debt and defaulting on billions of dollars, and they can't afford it.

To make the math work on the LBO, they don't need to attract a whole bunch of new customers. Once you do the PE deal, you've already accepted that you're not a growth story anymore.

But to Jason's point, Jason's exactly right. If you can upsell 20%, 30%, or 40% more by delivering this 100% agent, it might be the next most amazing company, but you'll be cash-flow positive. You'll pay off your debt. You'll create enterprise value, and 5,000 or 10,000 customers aren't going to have to do a migration in 2 years when you file for bankruptcy.

Jason Lemkin

What?

I'm going to ask you 2 questions.

Jason Lemkin

Oh, God. You've got the binaries on them.

So, who's going to go out first: OpenAI or Anthropic?

Jason Lemkin

Anthropic.

SpaceX and Anthropic, OpenAI in that order.

Jason Lemkin

It appears that SpaceX has already filed and is on track, so we already know the answer there. The fact that Anthropic just added the NVIDIA CEO to the board is a sign they're getting ready to IPO as soon as they can.

I mean, I'm sure he's going to add value in health care, but that means nothing. We're trying to IPO very soon, right? That and finding who the hell will chair the audit committee are clear signs you're going to IPO as soon as possible. So, given that they have that and OpenAI is sending out war memos, I'm just voting that they go out first.

Jason Lemkin

Yeah, not even a difficult question.

Will Sarah Fry, the CFO of OpenAI, be there when they go out? Yes or no?

Jason Lemkin

Look, I do know one thing. CEOs and CFOs have to (a) be wildly aligned and (b) the CFO should probably report to the CEO, right? And right now, I believe the CFO in this case reports to the president, VJ, which seems an anomalous arrangement, right?

Rather than saying, “Yeah, so-and-so is in or out,” what I would say is this: if that IPO is going to happen, and if this team is going to make it happen, then they need to be in absolute sync, and they probably need to have a more traditional reporting structure so that people don't have one more thing to think about, or wonder why this company is weird.

I always tell my CEOs up and down, dude, on the things where you're unique and different, where you're changing the world, do it as differently as you like. But all the boring stuff, just give the market what it wants. It wants the CEO or the CFO reporting to them. They want to be in sync.

It makes everyone's head hurt if the CEO and CFO are saying different things about something as fundamental as when we're going to go public. Stop the leaking, stay in sync. If that's just not a thing, no one wants to hear that.

No one wants to hear that because those are going to be the 2 people on the roadshow. They should be able to finish each other's sentences. You should be able to put them in separate rooms, like a police interrogation room, and each of them should say exactly the same thing. They should stick to their story.

The idea that you have separate stories from the 2 of them is just palpably absurd. So rather than saying who's in and who's out, that's what you have to do. Again, the same therapist that they're using for their relationship with Microsoft could actually do some internal relationships, too.

Interesting. But I will say 2 things. One, on the one hand, in isolation, if I'm running something like anything at scale, but especially at OpenAI, I want no daylight between me and my top lieutenants. No daylight.

Now, you could argue a CFO's job is to create a little bit of distance, to be that objective person in the room, and there is some truth to that, right? But there can't be daylight, or it's not going to work. So, if there is that daylight and it were that simple, you make a change, and you make a change before the IPO so it's least disruptive.

Now, having said that, if you are running a company at scale and there is already a ton of transition on the senior team—which there has been, a ton of turnover—I've seen a lot of times roles like CFO and others where you're like, “Listen, I just don't want to change one more thing. Yeah, there are some issues here, but Sarah is so experienced that things work. Workday finally works. We finally got all these things to work. This is not—I got 99 problems—this is not one that I want to tackle.”

I have often seen something like this where you shouldn't have the daylight, but it's not poltergeists streaming through, and it's just not enough of a problem. If your management team is super stable, you have time to work on these things.

But sometimes, islands of stability in your management team, even if they're not perfect, are just not worth another turnover on the senior team. Every time you replace someone, it takes its toll on the team, especially if they're popular, especially if they're liked and respected. It especially takes its toll if everyone thinks they're terrible: you move them out, and there's cake and a party on Friday. But I wouldn't be surprised if she's pretty popular in her own way, and it takes its toll.

The raw ingredients of success are there to have OpenAI be an amazing mega-IPO. I say, listen, get your therapist to make up with Microsoft, get aligned with your CFO, focus on the 2 big things, which are getting the ads product out and getting the enterprise cranking, and stay the course. You have the compute. Get it done.

If you have an executive who's truly arguing with the CEO in public, in the media, or someone like Dario back in the day at OpenAI going directly to the board with craziness, they've got to go. It doesn't matter. They've got to go.

You cannot be out in the media arguing with the CEO. And you cannot be Dario, no matter how smart you are, going to the board and saying, “I will only stay at OpenAI if I directly report to the board.” It doesn't matter how good you are. This is brutal. Sometimes you have to let some of your best people go because it's too dysfunctional. They've got to go in those situations.

Whenever you get a phone call as a VP from a board member, you're like, “Okay, there's a problem here.” Maybe the CEO goes, maybe the VP goes, maybe there's a problem you can solve, but I'm not going to be as absolute as “you've got to go.” Your antennae go up an entire notch when you get that call.

And you're right. As for briefing, most companies aren't interesting enough to have a media briefing. It's almost like these companies have become political-level drama. I just saw a fun tweet from Martin Cassada that was basically saying, “Enjoy all the drama. Enjoy all the petty backstabbing and all that.”

It's because this is such an exciting moment that the media is focused on. Because they're focused on it, you get all this. This is just what happens when you're in the center of the universe in terms of tech, so roll with it.

But you are right, Jason. You want to be the tightest ship in this sloppy, sloppy world. Whatever you do, if any VPs get this far on the pod, do not reach out to your VCs to say there are problems with your CEO. You are losing your job.

Not only are you losing your job, it doesn't matter if you're right. At some level, you're probably right. If you're a passionate VP and you see issues in the company, you reach out to Rory or Harry on the board, the odds that you're 100% wrong are 0%. But you are not going to—it's not worth it.

What are they going to do, fire the CEO over you? 0.0%. Unless there's fraud, 0.0%. You're gone. You may be gone in 3 months, or it may be that afternoon. You're gone. Just don't do it, VP. Just resign. Just resign with grace.

Jason Lemkin

Yeah, it's a much longer discussion, but yes.

What's the Shakespeare quote, Rory?

Rory O'Driscoll

“All the world's a stage, and we all play a part.”

Oh, he had to finish on a Shakespeare quote. Incredibly cultured. Thank you so much, guys.