Anthropic 融资及其对潜在 IPO 意味着什么？Mag7：Google、Amazon 上涨，Meta、Microsoft 下跌

The most aggressive quarter in capitalism. This is leaning in like you’ve never seen leaning in before. This is the top of the distribution pulling away.

This week, it was the Super Bowl of earnings: Mag 7 earnings. So, what happened? Meta lost; Microsoft and Amazon, thumbs up; and Google, a home run. Next, is the SaaS apocalypse over? Atlassian, 29% up; Twilio, 20% up; SaaS, 23% up. And then, in private markets, Sierra raising $950 million at a $15 billion valuation. And then finally, Sam Altman versus Elon Musk: week 1 of the trial begins.

Jason Lemkin

When the music stops, who has a chair with $1 trillion on it? Without the AI initiative, Microsoft is flat revenue. Anyone on LinkedIn who talks about their team, fire them. “My team this.” They’re also precious about their team. Lead from the effing front with AI. A CMO today should be able to run their own campaigns.

Ready to go? Boys, there’s a lot for us to report on this week. We had a big week of earnings; Mag 7 Super Bowl was the title that I had down. $540 billion in combined revenue, $700 billion in AI CapEx. I thought we’d start with the clear winner, which seemingly was Alphabet. Cloud backlog nearly doubled to $462 billion. Now, that’s Alphabet’s entire 2025 revenue. Do you agree Alphabet was the runaway winner from this mega earnings season?

Jason Lemkin

It’s jaw-dropping at that scale, right? I think the meta-theme of this episode—and we can tie it into Twilio and Atlassian and Palantir—is just this jaw-dropping acceleration. Even since we’ve been doing this show, it’s obvious the CapEx boom has been happening. Rory’s been great on this, right?

But to see Google accelerate at this scale—60-some-odd percent, I think—it once again makes you wonder why you invest in anything else. Why you invest in the amount, the sheer force, of spend.

And as a side note, just as a personal note, everything’s clicking at Google. When we started the show, folks were wondering: would search die? Because everything would go to the LLMs. Clearly, it hasn’t happened economically. It hasn’t happened in advertising.

Honestly, I was checking even our own little SEO at SaaStr. It’s up 60% this year—the highest ever. The highest ever.

A lot of folks are struggling to protect their cash cow versus investment in the future, right? Should Uber invest in autonomous driving, which it needs to, or should it invest in Uber Eats, which is on fire? There are so many trade-offs here, but Google has no trade-offs.

Everything—the only trade-off it has is, where do I put my chips? Because I need all the TPUs for myself and for my customers and my partners, and I need them for Replit, who hosts every website on it. They just have to figure out who’s getting this massive backlog.

So, first of all, I agree on the framing. Actually, I’m going to quote a blogger—a Substacker I read, Evan Armstrong—who described this quarter, and I thought it was a great description. He writes something called, I think, Napkin Math or something.

He described this quarter as the most aggressive quarter in capitalism. I’d actually amend it to say the most aggressive quarter in American capitalism, because it is a uniquely American thing. This was an astonishingly aggressive quarter.

And I love the wording, because there are 2 things going on. One, the largest 5 or 6 companies on the planet are accelerating at scale and doing plus or minus 20% growth overall, 30% to 40% growth in some of the subsectors. So, that’s wildly aggressive growth, and then even more aggressive CapEx.

These same companies are doubling down on CapEx, letting it grow 50% to 60%, such that CapEx is now eating most of their free cash flow. This looks like leaning in like you’ve never seen leaning in before.

And you wrote a really nice piece that said, normally, it’s the new guys—the open commerce being aggressive—and the incumbents moving a bit slowly and defending their turf. These are 5 of the 7 largest market-cap companies on the planet saying, “Hell no, we’re not going to get pushed around.”

In fact, 6 if you include Nvidia, which didn’t report this week. We’re just going to make the bet too. That’s the first zoom-out comment. This is the top of the distribution pulling away, which is kind of a sobering thing if you start thinking about all sorts of inequality and all sorts of those kinds of issues. But just from a wild perspective, these are amazingly great companies doubling down. That’s kind of the first big-picture framing here. You just can’t take it for granted.

Second comment would probably be: yeah, you’re right. Of all the companies, Google did, quote-unquote, the best. Its existing business—Jason, you framed it right—the existing business where you could have made a disruption story around search hasn’t happened. And the cloud business has accelerated.

But I’m going to make a controversial take: for all of these companies, I won’t say it’s disappointing, but this is only the start of it. It’s not the main event.

Across all of the companies that have been successful out of here—and it’s obviously Google, it’s not Meta, which we’ll talk about in a second—it is Microsoft; it is Amazon. If you analyze what they’re boasting about, and if you go 1 level deeper, it’s 2 things.

One, we sold a lot of compute to the LLM companies to make their tokens. And then the second thing they’re boasting about is, “Oh, and by the way, we bought some of those tokens and sold them to our customers because we have distribution, too.”

Both of those statements are true. Both of those statements made the revenue line go up, but if you zoom out a million miles, what you say is, “Oh, let me get this straight. You, the 5 largest market-cap companies on the planet, are effectively working for these 2 privately held companies and doing their distribution and doing their CapEx investment, while ultimately they own the IP.” Hmm. Just an interesting phenomenon.

And that’s kind of the second big-picture comment here, which is: this is a bunch of the largest companies growing quickly by servicing these other companies. It’s super interesting.

And then the last comment, going back to Google: I thought everything was really good, but it was interesting that—yeah, let’s talk about Gemini. You see all these evals of Gemini, and you know, it’s nearly as good—nearly as good for coding, right?

And then they cited a number, and I wrote it down. I apologize; I couldn’t find it in front of me. The month-on-month token growth in Q1 was pretty good. It was like 60 or 70 billion tokens per second, or some vanity metric like that.

But the truth is, Anthropic probably did 15× the tokens in Q1. The interesting thing is, I go back to thinking every other business is ancillary to the business of making tokens as an LLM.

The only guys who are even in the game there are Google, because they actually have their own model, and their token growth—their Gemini growth for coding and related things—significantly underperformed the other 2 guys.

So, where I’m going with this, the last thing I would say is: the most aggressive quarter in American capitalism is an underperforming quarter relative to the privates. Jason disagrees.

Jason Lemkin

No, no, no. I actually just want to make the point without connecting it to another section quickly. The point is super interesting, of course. They’re at the mercy of these 2 privately held companies, right? They’re the hyperscalers.

On the other hand, Palantir blew out the quarter, growing—I should have it at hand—80-something percent in revenue. And I watched—I never do this, because I just read it—but I watched Alex Karp’s analyst discussion. It’s great. He’s so entertaining.

He has a T-shirt just like Harry’s today. And he’s with his team. Really interesting. He made the point—and obviously he’s talking his book, to use Rory’s point—but he was really insistent that there is no value at the LLM level to Palantir because, at Palantir’s scale, this is not a simple B2B app. This is high-end AI.

He’s like, “There’s no difference between the LLMs.” So, it is interesting that where the value is across the 3 layers—from the application to the LLM to the infrastructure or hyperscaler—it is, if nothing else, a challenge that it’s fluid.

It’s just going to be fluid. It’s hard to predict. I’m not—I know you’re not challenging me. I just, when Alex said that, I don’t buy it today, but I don’t not buy it, right? It’s so fluid.

Totally. By the way, before I say what I agree with, I think he is living on a different plane. Look, I think—and let’s play it back to the hyperscalers. Let’s compare it to Microsoft, which went down after the results.

Palantir had an exciting enough enterprise product line that, despite the whole “LLMs are going to eat everything” story, they can make that statement and make it credible. And I believe it’s credible for them, and we can talk about why later, right?

I don’t think Microsoft can make that statement: “Hey, our Copilot is so cool that we don’t care which LLM we use, and enterprises are just ripping it off the shelves,” right?

So, I agree. I think that you’re correct for Palantir. There’s that same level of insatiable enterprise demand for Palantir. The only insatiable enterprise demand that Amazon, Google, or Microsoft are experiencing is the insatiable enterprise demand for sell-through of the LLM products that they now host and can sell using their large distribution channel.

And it's a great business, don't get me wrong. They get revenue—I mean, right now they probably get more revenue from the LLMs than the LLMs are getting, because they get the revenue going in on the hosting. In Google's case, they get the revenue going in on the chips, they get the revenue going in on the hosting, and they get the revenue on the distribution side from selling.

Only a small sliver accrues to the LLM, but my instinct is that, over time, that's the attractive sliver vis-à-vis those guys, not vis-à-vis Palantir, to your point, Jason. I think Palantir—because this is all about it—there's clearly a wall of money out here, and everyone's trying to figure out who gets to keep it when the music stops, who has a chair with a trillion dollars on it, right? And that's what we're all trying to figure out here.

Jason Lemkin

The only one thing I might add is that, in that Substack you quoted and others, there remains a lot of skepticism that this investment is consuming all of their free cash flow. Are they really capturing value? Microsoft owns some of OpenAI's IP, but are they really capturing any value? We went from free-cash-flow engines to no free cash flow, right?

The only thing I would say—I know this is Captain Obvious, but it's certainly clear to me. When I worked at a Fortune 500 tech company, it was clear then and it's clear today: if there's one thing these companies are good at, it's financial engineering, day in and day out. They have worked through 27 sensitivity analyses, and maybe they're making the wrong bet. Maybe they will regret having squandered their free cash flow on a bunch of chips sitting in depreciating servers, but they know exactly what they're doing.

They know exactly where they can offload risk to CoreWeave, they know what NVIDIA should take, and they know what to do. They may be making the wrong bets, but it probably doesn't matter because they'll scale them back, right? But I do believe that it is very thoughtful financial engineering. When OpenAI blew up and they fired Sam, Satya kept saying that this was not such a huge bet for us back then. Remember, he kept saying things like, “This is important, but this is not going to ruin Microsoft.” So my only rambling point is, I think they may be wrong, but I think they know exactly what they're doing to the last decimal point.

I don't know. Maybe a better statement is this. I saw a dialogue on Twitter about 6 months back where someone said, effectively, “Who are you to second-guess these folks? These are the smartest people on the planet making bets.” The response that really struck me was, “Look, no one at the height of the capex boom thinks, ‘I'm just doing something dumb here,’ because they wouldn't do it, right?” Everybody, when they're spending a trillion dollars, thinks this is a great idea, it's going to come back, and then sometimes you're wrong, right?

I think they are making informed bets, and right now, one of the things they're wrestling with is—for example, Microsoft was a little conservative a year ago, and now they're wrestling with that. It turns out that being aggressive is the winning strategy, and one of the definitions of a bull—of a crazy bull market—is when the most aggressive person makes the most money, right? So we're in that stage now, right? But you are right about one thing: if they overinvest, the great thing is—and this is where it is different from '99—if they are overinvesting and they throttle back, they still have their existing businesses. That's your point.

Jason Lemkin

I think it's low risk—lower risk than the Substacks actually claim it is, right?

It is, I agree. That's why I've come up with this distinction between overinvestment—which may be happening; so far, it's not even overinvestment, we can't prove overinvestment—and a bubble. I don't use the word “bubble” because the word “bubble” is, to your point, Jason, making a financial and valuation statement, right? I don't think you have to make that statement here. You're simply saying, right now, the computers are all finding a good home. Will they over the next 5 years? Who knows.

It's interesting. I want to go back to your comment on Satya that it doesn't matter that much to us. One of the most interesting statistics about Microsoft is that, in this quarter, excluding—taking out their “AI businesses”—and to some extent they do this reallocation; all the good stuff is reallocated, right? But if you take out Copilot growth and Azure growth, the rest of the business is flat to slightly down.

In other words, Satya, if you didn't have an AI business, you'd be another SaaS company trading at 3 times revenues. Welcome to our world, right? So it does matter. I mean, 3 years ago that might have been true, but right now, if the AI bet is wrong, all these valuations are wrong, despite the fact that they're not outrageously valued on a P/E basis. All the growth is coming from these initiatives. It was stunning to me. Let me repeat: without the AI initiative, Microsoft the corporation is flat on revenue.

The allocations have been great, and the more you allocate, the smarter you look. Six to 9 months ago, there was this, “Oh, I don't want to buy from OpenAI because they might not have the money.” Now, Cohere is killing it, OpenAI looks like a good credit, and the more compute you have, the smarter you look. I don't want to sound like Debbie Downer here talking about capital misallocation when right now it's working. But, yeah.

Jason Lemkin

Yeah, so a good bet?

The AI ARR is $37 billion. That capex spending will be $190 billion. Does that concern you to hear?

Jason Lemkin

The most aggressive quarter in American capitalism, baby. That's right. I mean, yeah, I know. Look, if they're wrong, they just throttle back the future, live off the cash flows, and have a great big digestion period where the stock looks overvalued and goes down. So it's not as terrifying as—let's just say you had the same bet and you financed it with $150 billion of debt. That would be beyond terrifying. But no, it is an aggressive investment well in advance of revenues, so by definition, it's an aggressive bet.

I have a slightly different take, for what it's worth. We know this, but you especially see it when you work on the other side, at an extremely profitable public company with massive margins. Your cash is so trapped. You cannot spend it; your EPS goes down, your dividends might be impacted. But there are moments in time when the public market lets you spend it and doesn't take a hit for it, right? Some of this is financial engineering, where it sits on the balance sheet, et cetera. That's important, right?

One of the reasons Salesforce Ventures and Google—Google owns so much of SpaceX—and everybody is doing this is that you could do something with the cash. You're allowed to make investments. It hurt Shopify this quarter. But when you have these windows and you don't have to be like Jeff Bezos, day in and day out convincing Wall Street to let you spend, when Wall Street lets you spend and doesn't punish your stock, you should spend every dollar on the balance sheet that it lets you spend, because you'll build it back up.

But it's trapped when growth slows. When growth slows, you're literally at the mercy of getting another half cent into your EPS. It's a terrible place to live. Being at the mercy of not being able to spend is awful. “Oh, we have $6 billion on the balance sheet, but if you can't spend it, who cares?” Literally, who cares? You can't grow if you can't spend it.

Jason Lemkin

I understand what you're saying, and you're not incorrect about the wider constraints and how Wall Street and perception impact your ability to invest aggressively. But the result you articulated is that it raises the risk of bad capital allocation, because if you're grabbing the moment not because the ROI is there but because the permission is there, what you're basically saying is that your decision to invest is in some part predicated on the street's willingness to give you that permission.

Now they're giving you permission, and I think you're correct: everyone's allowed to lean in right now a little. But it does heighten the risk of groupthink and the fact that you end up allocating badly. Having said that, I've just got to say it again: right now, the allocations have been great, and the more you allocate, the smarter you look. Six to 9 months ago, there was this, “Oh, I don't want to buy from OpenAI because they might not have the money.” Now, Cohere is killing it, OpenAI looks like a good credit, and the more compute you have, the smarter you look. I don't want to sound like Debbie Downer here talking about capital misallocation when right now it's working. But, yeah.

Yeah, but let me just give you one example. Sorry, I was talking with the CEO of a $20 billion-plus public software company where growth is modest today, but revenues are very impressive. We were talking about AI and agents, which is what people want to talk to me about, and he was talking about how they were saving roughly 100 basis points on their LLM costs. I'm like, “You're dead.” You're dead because you don't have permission from Wall Street to do the opposite. You don't have permission to spend 10% to decrease your gross margins 5% or 10% so you have the best agent in your category.

You're trapped in a death spiral. The margins of this company are very impressive, right? But he's so focused on driving them up to keep Wall Street happy and to keep the Carl Icahns happy. The room for maneuver is so narrow at a $20 billion market cap. When Satya and Google can spend, if I were running one of these companies, I would force you to force-rank it, of course, but spend it all before we're allowed to—

Jason Lemkin

Spend it all, boys, because the day will come when we can't spend anything.

David Friedberg

Again, I always recoil from the vehemence of the direction, but I think you're right in the sense that if your number one priority right now is to optimize your LLM spend dollars, you're missing the point. Over the medium term, you'd probably want to optimize, but if there's positive ROI in LLM spend in terms of agent functionality, yes, you should be doing it.

As usual with you, Jason, I agree with the broad thrust. You're exactly right. If you think you're going to make money by saving money right now, that's not the case. You do want to be a little more circumspect than some of the investment you're seeing, but err on the side of aggression.

So maybe it's a good time to talk. By the way, I did find that number for Gemini token production. They boasted about the fact that their Gemini token production went from 10 billion per minute in Q4 to 16 billion per minute in Q1. All I'll say is, Anthropic 10x'd in that period of time, and tokens probably went up by more than that.

What's your takeaway from that, then?

Rui Ma

My takeaway from that is, every time you look at the evaluations, it says, "You've got the big 2, obviously Anthropic and OpenAI, and then, close behind, you've got Gemini and Grok. Then you've got the open-source guys 6–12 months behind." When you look at what's actually going on, the big 2 guys are getting all the money.

Gemini, for some reason, doesn't appear to be able to do a great job of developing code or Lovable and getting that kind of traction, because coding is where it's all happening. Grok is obviously nowhere until it gets Cursor done. Even with the open-source models, you do see people using them, and you do see the Palantir perspective, which is that you can get something done with open source.

This really matters because there's one view of the world—the Palantir view of the world—that 5 years from now these are all API calls; they're all commodities. If you look at the benchmarks, you could convince yourself of that story. But if you look at what's actually happening, there are millions of developers who have that same choice every day, and they are choosing over and over again, because of the model or because of the harness, to go with the big 2.

I think Google has outperformed everyone in terms of being AI-relevant. They are by far the best of the 5 companies that reported this week. But in terms of mindshare in coding—taking that as the mother lode and the epicenter of the revolution—they're nowhere compared with the other 2 guys.

Help me out. We have Meta crushing earnings: revenue of $56 billion and EPS of $10.44 versus $6.67. It's an insane beat. They got crushed because CapEx was raised again, from $125 billion to $145 billion. Why did Meta get crushed on CapEx increases when Google got applauded?

Rui Ma

Two companies are spending $100 billion to $200 billion, and one of them has revenue coming in that's clear and attributable, and that's Google, while one of them doesn't. That's Meta, right? That's the big picture.

Let's dive down one level. Why is Meta doing this? You can imagine 3 scenarios, and I think there's only 2 of them. One scenario is that they're going to be yet another third-party hyperscaler provider. It doesn't look like that's happening. They're not competing with Google to provide compute to Anthropic—not that they shouldn't, right?

The 2 reasons they use it: one is the reason they started using a little bit over the last couple of calls. "Oh my gosh, we're optimizing our ad performance a whole ton using these models, and it's making the numbers better." Let's dive into that a little. There's no doubt the numbers are amazing. If you do the math, they're saying they're getting a 10–15% lift in the last couple of quarters; the performance is better.

But it's super hard, unless you see an A/B test, to know whether you're really getting that lift. That's always the case with lift analysis. You need to see half of the ads optimized using, let's call it, Meta Llama 5, whatever the new model is called—I can't remember—and half not optimized. Are you really getting lift? You probably are getting some lift. Is that worth $10 billion a year? Maybe $15 billion? But they're spending $150 billion.

They've been using this justification for the LLM spend: it'll optimize our existing business. I've always felt that's because if it takes $150 billion in CapEx to give a 10% lift on a $200 billion business, it's probably a mistake, especially when some of that technology might be available from third parties.

What was noticeable in this call was a little more of the "we're just going to build next-generation experiences" qualitative commentary. Simply put, if people go from talking to other humans and looking at news to talking to chatbots, Facebook wants to be there when that happens. I think that's the justification.

The market is looking at it and going, "I get it. This isn't a business. This is, 'I want to show up if something happens.'" It's a $150 billion bet on the future that's not quite articulated. You just put a slightly higher discount on that. That's all that's happening here.

I don't know why they're spending, and he doesn't have to ask anyone for permission. Maybe he'll be right, like he was on Instagram or WhatsApp. Or maybe he'll be wrong, like he was on the metaverse. But it's not like Google, where you can say, "Oh, I get it. They're spending $150 billion, but they're getting $60 billion back, it's growing 80% year on year, and in 2 years it'll be cash-flow positive." It's just a different thing.

Jason Lemkin

I do think that Wall Street is also, to Rui's point, building all these spreadsheets: GPU depreciation, CapEx, where it's going. The Meta model doesn't support that.

Rui Ma

Agreed.

Jason Lemkin

It's indirect, right? It just doesn't support it.

Rui Ma

Well put.

Jason Lemkin

Nor does the math tie, but it's too confusing. They're all running spreadsheets and arguing over the inputs and outputs of these capital investments, and they just debate it.

Rui Ma

I like that, Jason, because effectively what you didn't say, but it's true, is that Mark isn't running those spreadsheets and doesn't give a shit about those spreadsheets. "Thank you very much, guys. Knock yourselves out. I'm going to spend to be relevant."

Their little heads are hurting when they run the spreadsheets because they can't touch it. They can do it for the other guys, and they can't do it here. I think you're exactly right. He didn't make $200 billion by running a spreadsheet at Harvard. He just built a product and got the traction, so he isn't going to change now.

Of the 4—Amazon, Meta, Microsoft, and Google—you can buy 1 and sell 1. What do you buy? What do you sell?

Rui Ma

You buy Amazon and you sell Microsoft, but these are weakly held.

Why do you buy Amazon? It's the one we didn't touch on. For everyone listening, Amazon has $181 billion in revenue, AWS at $37 billion, and its fastest growth in 15 quarters.

Rui Ma

It's weakly held in the sense that there are 2 net sellers and 2 net buyers. Amazon and Google are doing well; Meta and Microsoft are less clear. You obviously pick 1 on each side.

I picked Microsoft because that sentence—that they're flat excluding AI—made me really pause. With Meta, even though what they're doing is crazy, they can always stop, and they'll still be the best ad network business on the planet. That's the sell.

On the buy, honestly, it's too boring to buy Google at this point. It was great a year ago when we were saying it was great. Now it's up; it's slightly the highest valuation on a P/E basis of those 4.

To some extent, it was an on-the-fly contrarian bet that Amazon now has access to Anthropic's models and is going to get that lift. They have the AWS distribution, and they now have access to Anthropic's models, so they're more aligned there.

One thing that we talk constantly about is how AI impacts the top line and, to a lesser extent, the bottom line. There's a secondary, derivative effect that benefits everyone and doesn't benefit Meta at all, which is that we're not just using AI to make applications better; we are building many, many more applications.

There is an application boom. This is the greatest boom in application building in the history of our lives. The amount has exploded. Right now, you make even more money selling tokens than you do selling basic AWS or GCP services, but there's a double boom going on here.

I have to put Meta last because they're not benefiting from the application boom either. They're not benefiting from it, and there may even be a crossover where AI gets a little mature, which I do not remotely see happening, okay? But even if it did, the application boom has just begun.

The thousand flowers of applications are blooming. Maybe this was the quarter SaaS bounced back from the SaaS apocalypse, but I think in a couple of years we'll look back and laugh at the SaaS apocalypse and say, "My God, the explosion of B2B applications was like nothing we've seen before." They were just old guys that learned a new dance and didn't die. But who cares? We'll look back in a couple of years and we won't care because of this app explosion. Everyone's building everything.

Jason Lemkin

It’s so wonderful. That could lead to a renaissance for Amazon beyond what we’ve seen, right? Because of its traditional strength. It benefits Microsoft in the enterprise, but not below that, right? And GCP is in Gemini, or there’s this weird thing where it’s very cost-effective and developer-friendly. I can’t predict, but I know Meta loses in the app explosion.

Before we move to the SaaS apocalypse, I just want to mention that Palantir—we added it at the last minute—was a home run in terms of performance. RPO is up 134% at $4.45 billion. Rule of 40, they’re at 145%. Borrowing from your tweet here, Karp’s letter was blunt: this number has only been matched by AI infrastructure companies Nvidia, Micron, and SK Hynix. I don’t know what to say other than, holy cow, these numbers are really good.

Rui Ma

He also said—and it didn’t quite tie to the forward guidance—but he also said that they were doubling. So they’re accelerating from there. He was clear. Now, I couldn’t quite get that to tie to the Wall Street guidance. That happens when you shoot from the hip, but he was very clear that they’re doubling. So they are accelerating.

I just think he was shooting from the hip in terms of the timing of when the RPO would land. He was clear, and, Harry, he was very clear that Europe has just started. It’s so far behind—not at the startup level, but at the buying level—that that’s just started on the commercial side.

Jason Lemkin

The way I was thinking about it is, they’re just in a very enviable position, right? Because if you’re a large corporation, I always think that whenever you’re selling application software, you want to be one of the top 2 initiatives for the most senior person you’re selling to. That’s how you make money.

Right now, they’re selling to the CEO, and the most senior initiative—the top 2 initiatives—for every CEO in corporate America is to do something in AI. That’s what your board is telling you. So what are you going to do? If you think about it, you don’t have a large number of choices.

You can sign up for something with Anthropic and OpenAI. You can buy Copilot, or you can buy Cursor, or you can buy Claude for every one of your employees. That gets individuals using AI, but it doesn’t move the needle, right? What you really want to do is some corporate-wide initiative that’s big, right?

All the AI app companies that we all fund tend to be fairly point solutions, because they’ve been founded in the last 2 or 3 years, right? They’re really good at saying, “Let’s take Sierra. We’ll talk about it in a second. We’ll deal with your customer support.” That’s probably the most call-high company in AI land, in new AI land. In other words, Bret Taylor can call higher than anyone else; most of these 25-year-old founders can’t. So that’s as high as they can call, right?

You can spend $2 million with Sierra. You can spend $200,000 with Harvey, or whatever. And that’s an issue. If you’re running corporate America, if your number 1 task is to do AI, you don’t spend $200,000, because that doesn’t solve the problem, right?

What you really want to do is some corporate-wide initiative that’s big, right? And then there’s this 1 company that’s been selling to the US government for 20 years that can move in $20 million and $100 million chunks. They can literally say, “You want to redo your entire go-to-market infrastructure? You want to redo your entire business intelligence with AI? We can do that. We did it for the US government. We did it for the DHS. We’ve done it for JPMorgan.”

They credibly move in $10 million chunks. So now this corporate CEO can say, “I have 2 initiatives this year: launch that new product and make this company AI-first. I just signed Palantir for $10 million. They’re going to bring in people. They’re going to make it happen. Initiative done. Tick it off. Report to the board. It’s June 30th. I’m on track for the year,” right?

It sounds stupid, but big companies have to spend big money to do big things. There are very few software companies that are situated to do this. It’s why IBM has existed for 30 years longer than it should, right? Because they can do these kinds of big deals and big initiatives. It’s why EDS made a lot of money as an outsourcer.

Palantir right now is the only business that can credibly say, “Hey, Mr. Corporate America Fortune 500, you want to move the needle with an AI initiative that has measurable, demonstrable results and is about enterprise-wide transformation? We can deliver you that,” right? The alternative to them is Accenture with a bunch of cloud licenses. What do you do, right? I’d go with the—

Guys who have won in the war.

Rory O’Driscoll

Palantir is like General Catalyst for AI transformation, for the AI people. You need to move $100 million; you probably won’t get fired. Cool.

Absolutely. And it has this more clearly demonstrated value. I know that sounds really simplistic, but just think of it from the perspective of the CEO, right? You can’t have your top 3 initiative be a $200,000 spend. With any of these AI companies, it’s just like, “Whoa, that’s a bit of a weenie bash,” right?

You come back in and say, “We hired the guys that have been doing this for all of corporate America and for the government. We spent $5 million. Here’s a 3-year plan.” You’re done.

Jason Lemkin

To add on to this, maybe 2 things. One, the numbers have obviously gone up, because I’m dating myself. When I was a VP at Adobe, there were only 2 or 3 big initiatives at a time, to Ro’s point. That’s it. But they were all $20 million a year and up. Maybe they’re $30 million today. That’s just what it cost at Adobe’s scale in the Fortune 500 to have something that was critically important for five figures of employees. That’s what was budgeted.

I believe back in the day Salesforce was $26 million a year, and Workday was $24 million. Those were the big projects. They took 5 years to deploy. Now Palantir can deploy it in less than a year, which is magical.

The other thing I would add—and it was funny, having been doing B2B for a while—is that Alex Karp, on the earnings call, first talked about the defense business. He talked about how important it is to support our country, and that the only thing more important than our customers is supporting our country. Clearly, they’re deeply embedded in defense in the Western world.

But then he went to the commercial side of the business. What he said to me was super interesting, even though it was obviously different from my entire year—the last year. He said, “My entire year, on the commercial side, I get brought in by a stakeholder—marketing, revenue, someone. Then I’ve got to sell to another stakeholder and another stakeholder, and it takes a couple of years.”

He said, “In the last year, every stakeholder shows up to the meeting. Everyone is there.” So not only are they a top 2 or 3 driver—to Ro’s point—to drive corporate change, but the CEO and the CFO, importantly, are saying, “Now, everyone come to the table now.” And if Palantir is the answer, or at least the bet, they’re not going to evaluate for 2 years. They have to do this now.

Alex was so aggressive, in his somewhat charismatic, weird, goofy way, on the call. But he was taken aback by the compression of the buying cycle. It felt to me almost like the COVID buying cycle today, because we just never saw folks on the commercial side do deals this big like Adobe, but instead, everyone’s there. Everyone shows up to the damn meeting.

Rory O’Driscoll

Look, AI sparks the imagination in a way that digital transformation, databases, client-server, or SaaS just doesn’t. It’s got that spark of, “Oh my God, is it live? Is it sentient?” Corporate America now believes it is the way to transform their company. And you’re right, Jason: every board is telling every company to get on top of this.

Who’s going to be the C-level executive who says, “No, I don’t want to join the meeting on the most important corporate initiative. I don’t think it’s going to work,” right? By the way, what it means now is that those lily-livered naysayer voices have gone to zero.

So you do think there’s going to be some pretty interesting misallocation. As I say, if you’re selling $5 solutions right now for the biggest corporate imperative, you’re in a golden place. Good for them.

At a $349 billion market cap, though, is it priced to perfection, or is there upside?

Jason Lemkin

Of course it’s priced to perfection. It’s priced to more than perfection. I will admit, I tailed when I saw this quarter, because when you pushed me again a few weeks ago to name stocks, I’m proud that I named Atlassian, and we’ll talk about that in a second.

But on the expensive side, I said Palantir is the best situated. And yes, it’s wildly expensive. You really have to grow for 2 or 3 years to get into this valuation. But if the boom has legs, they’re the most likely to grow into it.

It’s still terrifying. I woke up the night after I did that podcast, and I’m like, “Ooh, that’s a risky one,” right? Atlassian is a value play. It’s a no-brainer. But this one—wow. You’re leaning in, but this quarter kind of justified that. You do the math, and 2 years of doubling makes it look cheap.

But Alex predicted basically a year of doubling. So there’s an argument—if you think that’s cheap, I haven’t fully thought it through—if you think it’s cheap, it’s not the craziest place to get to, right? Because 100—100, we’ve already—the CEO’s already said he probably fudged a quarter or 2, but he said it’s coming.

Jason Lemkin

Agreed. Yeah.

Rory O’Driscoll

The thing, just to—I know it may be Captain Obvious, but going to Rory’s narrative about how corporations work in Palantir, it’s just a reminder. Every conversation I have is a reminder that no one has this expertise in-house. No one.

It’s the worst gap between in-house and external expertise in our lifetimes. And so, for years, this is going to benefit Palantir. It’s going to mean that even if we’re head-scratching about why Anthropic and OpenAI are setting up these consulting entities, which seem goofy, right, they’re not, because the dollars are going to go up and the initiative is going to go up.

But the inability to have anyone in-house who can execute—there’ll be folks down the line. Some HubSpot agencies will figure this out. Some Shopify dev shops will figure this out. Everyone in these ecosystems thinks the majority of HubSpot and Shopify agencies who actually deploy—most of them are not direct to these SMBs. You need an agency to deploy HubSpot. Most of them are gone.

They have no AI play. But the ones that do are going to have infinite demand, because whether it’s enterprise or SMB, no one has this expertise. No one. It’s almost embarrassing how few people have this expertise, right? Even Coinbase today: “I’m laying off 15%.” I don’t need managers managing managers. I need folks who will actually do the work in AI, right? If Brian doesn’t have the people, what hope is there for the rest of the world?

Agreed.

Jason Lemkin

And I think, when you say it, in one sense it sounds almost derogatory: no one has the expertise. But the real truth is this. I’m donning my microeconomics hat here, right? Logically, what you want is for someone to take the time—6 months—to develop the expertise and then sell it to me, and I only have to pay them for a week of their time. It totally makes sense.

I even think about how I learn AI. A good slug of it is doing it myself. But when I get stuck, I just ring our chief data scientist and say, “Look, I could spend a whole day or two slogging through this point, but just tell me the freaking answer so I can keep moving here, right?” That, in a microcosm, is what you need in every organization.

David Friedberg

You’re a line manager in marketing. You’ve been told you need to roll this out. You just have to hire the expert who knows. Which is why, to your point, Jason, this is a Darwin test for every consultant and, frankly, anyone looking for a job, too. That kind of gets to the undergraduate unemployment discussion.

If you don’t develop the skills you need, you deserve your fate. I truly believe if you develop the skills that people want and focus like a laser on the things they actually want, there’s real demand for that skill in all these companies.

Jason Lemkin

I call it the $250,000 SDR. There is now a market, instead of SDRs being worth $60,000 in the US or $80,000, for a small number of $250,000-a-year SDRs. But there’s almost no need for a $60,000 SDR with 6 months of junior-college research who can’t spell ElevenLabs. You just don’t need that person anymore. You needed them 3 years ago; you don’t need them today.

But you can spend $250,000 per year for someone who is as productive as 20 human SDRs. Those are the skills you have to have.

Before we move on to the SaaS apocalypse next, it’s so funny. We are about to do something incredibly funny. We’re about to skip the one line item in the agenda just to point out that Apple pulled—Apple beat across the board. Tim punches out on a high. Stunning results. Great quarter. No real AI story yet.

Thank you, everyone else, for getting caught in hysteria. Meanwhile, we’re just building one of the 2 largest companies on the planet and doing really well here. Not wasting our money on CapEx. Continuing to do buybacks. Just working for the stockholders. I just felt the need to call that out in passing.

Jason Lemkin

I thought the memory-chip supply constraints were interesting.

That is interesting, and it’s not just for Apple. One interesting statistic I saw is that everyone’s talking about how the CapEx budgets have all been raised this quarter by 10%, 20%, 30%. A significant slug of that CapEx increase is for the same physical amount of CapEx, just at a higher price, because the memory portion of anything you’re doing has exploded in cost.

I agree, and I think you’re going to see some impact from that right down to the price of your iPhone next year. It doesn’t appear to have impacted demand, but you probably will see that filter through. That would be the iPhone you can’t afford, transported to the US on a plane that can’t have jet fuel from the Far East. But, yeah, there you go. Much, much pain to come.

Memory prices and, obviously, memory stocks are a factor not just for Apple, but across the whole CapEx story.

David Friedberg

Well, I mean, Apple basically increased the price of the Mac mini this week from $599 to $799 because of memory.

Oh, wow.

David Friedberg

The $599 model sold out because of these damn OpenClaws. I believe it’s maybe the 16 GB versus 8 GB. I forget what the difference is, but it’s not $200 even at current prices.

All this stealth inflation—it’s off topic, but it will lead to a lot of socioeconomic stress when folks can’t afford either a $799 or $599 Mac mini and the average individual is hurt. It’s like the rental market in San Francisco. It’s almost unimaginably competitive versus even when we started the show. It was cheap to live in San Francisco. Now you can’t rent anything. They don’t even exist. No one’s going to leave.

Jason Lemkin

I see bid prices like $3 million higher than the $5 million listing.

David Friedberg

Yeah, I just made $30 million on my Anthropic vesting. My partner wants a house. What am I going to do, argue that it’s not worth $5 million? I mean, it’s been 6 months. I’m just going to buy it. I don’t really care, right?

Last week, we were in a bit of doldrums and we wanted more positive news. The SaaS apocalypse could be over. Atlassian up 29%, Twilio up 20%, Five9 up 23%. Boys, is the sunshine coming out? Is the SaaS apocalypse over, or is this a case of 3 good-performing companies pulling away?

Jason Lemkin

Well, look, Rory categorized these companies the last few times we did this. Let me just break it up. First of all, Five9, I’m not interested in. Reaccelerating at 9%, I don’t care. I’m not saying that they’re not benefiting, but I don’t think that deserves to be with the friends.

It is heartening to see Atlassian and Twilio reaccelerate. These are older companies, right? Atlassian is still founder-led. Mike was on the show, right? He’s great. But everyone was worried about Atlassian, right? Twilio pushed Jeff out, who was on the show, who we love. Somehow, moldy oldy products were able to reaccelerate. So, 2 great stories.

If nothing else, they support the idea that for folks who are benefiting from AI, we’re past the bottom. We’re going to see some more quarters come out that maybe aren’t as good before the next show. But this is my bias and my perspective: Wall Street missed a subtlety that I think you guys will agree with.

Atlassian got really good at monetizing its AI product. It’s Rovo AI. It sold the F out of it last quarter, and people were happy to pay for it, okay? Its net new customer count is still slowing. Twilio did the opposite. This company was dead when we started this podcast. Now it’s reaccelerated to 20%, but accelerated. Their disclosure is less clear, but their net new customer count may have grown 40% in the last year because of AI and other startups and other companies.

The ACV per customer hasn’t gone up quite as much, right? But there is an explosion. ElevenLabs uses Twilio. All these AI folks use Twilio. So Twilio won on 2 points: folks use it for AI, and they had net new customer growth. Atlassian monetized its base with AI. Thumbs up. Passed the test.

But it’s not clear that it’s attracting new customers. So you might be deferring bad news ultimately if your AI story gets more revenue from your base but does not expand it. To be proven.

David Friedberg

You’re right on the facts. Going back to—you said, “Is the SaaS apocalypse over?” No, I would say it’s not over in the sense that I didn’t think—look, 3 months ago we said, and I think Teams is a buy, Atlassian’s a buy, right?

If you just avoid going from guardrail to guardrail, the big picture is that when people price these stocks like it’s all going to zero, you’ve got this potential for a 20% or 30% bounce, and you saw that right now. Taking Atlassian, for example, I don’t think it’s going to 10x from here. It’s never going to be an AI-first, 5x–10x growth company.

What it is going to be is a really well-run company that, as you say, is—if you can’t add new customers, maybe your steady-state growth is 20% or 30%. If you can add new customers on top, maybe you can go higher. But these are going to be cash-flow-positive companies growing 30%. They’re going to deal with the SBC issue, and then instead of being worth 3 times, they’re going to be worth 6 times.

You're going to move in that bounded range. My point is this: you're going to move in that bounded range, and when people's heads start exploding, saying it's all going to zero, and they trade at 3 times, but the company's fundamentals are decent, you can buy and get a 2x. And when people start believing it all and they're still pretty lofty, like ServiceNow was before the last announcement, when you're trading at 6 or 7 times revenues, you're very vulnerable to a correction, right?

I think you've got a bounded range, and you're right: the low end is Five9—you know, it grew at 9%. The high end—I mean, Atlassian's growth rate was pretty impressive: 32% in GAAP revenue, right? And I'm not surprised at the lack of new customers, because I always felt these stocks would slow down independent of AI. These markets are fairly well served.

Atlassian has done a great job over 15 years of meeting the need for this product. Now, additional customer growth either is new company formation or taking away from someone else. So it's not surprising. It's a bounded, well-executed company.

I'm delighted for them. I'm delighted—I said buy it at $62 two months ago when we were asked. As I often say here, things are proving out to be exactly what they should be, which is these are solid, high-growth, cash-flow-generating companies probably worth closer to 6 times if they're going north of 30%, maybe even a little higher, right? That's what they traded at for a decade and a half before COVID. And there you go.

Jason Lemkin

It goes back to your early framing: Are you getting value from new products? Do your existing customers adopt them, and are you getting new customers? When you do the latter, yeah, it's good. Why would anyone in AI—why would Level Up or Replit bother rebuilding this stack? Especially if you can buy it on an API basis, just call it a day.

David Friedberg

Well, the risk would be if, at the infrastructure layer, someone had built a better Twilio—something that was better—you might switch, right? But Sierra's $15 million deal runs on Twilio, right?

There's a lot of—I'm not a total expert, but obviously there's reliability, right? There's infrastructure under the infrastructure. And so my learning is, for a lot of folks, nobody built Twilio. Nobody beat Twilio. It's not just software. They could—it's possible, right?

But now they're a beneficiary because this infrastructure was good enough, like Apple. I mean, it's a little attenuated, but there's something. It's good enough to benefit from all the trend happening. It's good enough.

Who are the other traditional SaaS companies in SaaS jail that should be released in the same way that, hopefully, Atlassian and Twilio are being released?

David Friedberg

Well, here's where Aaron and I diverge. That's why I'm worried Atlassian isn't a 2-pronged AI beneficiary. To be a 2-pronged AI beneficiary, you have to be able to monetize your AI and attract new customers. There are 2 prongs, and the ones that have done it so far are close to infra: Cloudflare, Twilio, Datadog, and even DigitalOcean, which proves anybody can do it, right.

Jason Lemkin

If you're the 11th cloud provider and you can grow 352% in stock price, anyone can do it. And the one I'll tell you—the one I'm waiting to see is HubSpot.

This week, HubSpot announced that it will put agents on parity with humans in its coming release. Its platform will be completely open to agents, and it'll make sure that the agentic version of HubSpot is at least at parity with the human version. That's the right vision.

Now, are they going to overcharge for it? I mean, I'm not even sure anymore it matters outside of SMBs, but I want to see if that works. It's a little late, but it's not too late. In a way, it's Marc Benioff's headless vision that he talked about, right?

But let's see if HubSpot can become the hub for agents in all of its categories—for SMBs, for GTM. It should reaccelerate dramatically. Let's see if it works. If it doesn't work there, I think we can write all the rest off—all of its peers, I mean.

It should work, but HubSpot has such a broad customer base. It's relatively more tech-centric than Monday, right? So I think this headless thing—I'm not saying it should lead HubSpot to double, right? It's not mathematically possible—I believe over the next 12 months, this strategy, if it's real, should lead to genuine reacceleration at HubSpot. Otherwise, there's no hope for this class of categories, because they're going to make it completely open to all agents.

David George

I think, to your point, Jason, far more of them are going to fall into the not-reaccelerating category. Reacceleration will be the exception, not the rule, right? And then, for the others, it's a question of whether it's a slowly evaporating ice cube or a quickly evaporating ice cube.

The underlying issues in this apocalypse haven't changed, right? For the drama to reduce around it, we need a few more folks to be in that category, I think. And I'm not sure Atlassian gets both prongs, but I'm here for it because it lifts all the boats, right?

If we get 3 or 4 of these, then we can kind of—the underlying issues are there, but we can move on and talk about other things.

David George

Just to push a little on Atlassian again, because I actually think you're changing the goal. I thought your other goal was not new customers—which is obviously the best of all—but I don't think it's realistic any more than I think Zoom will ever get a new customer again. Everyone who needs a Zoom account has one.

But the other test you applied, which I thought was a good one, was: Can they sell new products to their customers? Atlassian's daily active users and their AI revenues did take a jump, so I do think there's some lift there that you can get.

It would be great if 10 or 15 of the companies started doing that; then you'd have a sense of what good looks like for SaaS companies. I think that's what people are struggling with, because there's no universe in which any of these companies become an LLM, or a Sierra, or a Harvey, or anything like that. That's just not going to happen. You can't get there from here.

The question is, can you return to 30% growth with free-cash-flow positive, stock-based comp under control, and gross and net retention such that no one's terrified that you have a zero terminal value? If you can do those things and demonstrate relevance, then the advantages you bring to the table in terms of scale, in terms of a couple of billion dollars in revenue, can all come to the fore, right?

And you're right, a couple of these guys have done it. If more of them do it, then we'll know what winning looks like. And to your point, Jason, then it'll become painfully clear what not winning looks like.

And let's be frank: what we saw in Medallia was an investor walking away saying, "This thing isn't winning. I just can't get there from here," right? Even though it had positive EBITDA, that was 5 or 6 times coverage on the downside. There's just nothing here, right?

So, in many respects, seeing what Twilio and Atlassian have done will be a positive for the people who are on that journey and will be the nail in the coffin for the people who aren't, because it'll be like, "Oh, that's what it takes to win, and you're not doing it, Mr. Fill-in-the-Blank."

Okay. On to the next. Onwards to private-company land: Anthropic at $44 billion. Jason's question here, which he put in, was exactly the right one: Are there enough developers for this level of revenue growth to continue?

And you can immediately say yes when you look at the service value of developers. But when you look at this, it's $100 million per day. How did you guys react to this?

Jason Lemkin

I think that's the right question. If I was to pick 1 number that I'd like to know, which would give me an informed opinion on this, we're trying to figure it out. We're doing some work in the portfolio. What is the steady-state token spend as a percentage of salary dollars per engineer in a fully mature, AI-first organization?

Because I was just looking at this, doing the bottoms-up knowledge-work analysis. And I think in anything other than pure AGI, which is too arm-wavy for me, most of the other jobs have task-automation potential. In my view, that's sub-10%. You can do some of marketing, some of sales, some of accounting, but not all of it.

Whereas for coding, you can do a ton today. And we don't need to argue the former yet, Harry. The point is coding: it's pretty clear that it can be done, that there's a huge amount of automation that can be done.

So now, in a way, it's not as clear yet for the other areas. Let's just go with that for now, right? Therefore, coding is the tip of the spear. Coding is—pick your cliché—the canary in the coal mine. Therefore, if you know what the long-term, steady-state automation of tokens as a percentage of salary is, you know how big this can be.

At 20% or 30%, Anthropic can grow into that multihundred-billion-dollar revenue category. Maybe even half a trillion dollars. At 5%, it gets a lot harder. So, to me, that's the question.

When you look at Andrej Karpathy's thing, right? He used to use it for 20%, and now he helps with the final 20%.

David George

But how much would you have to pay Andrej Karpathy to code for you? I can tell you it ain't $250,000 a year.

Jason Lemkin

Probably about a billion a month.

David George

That's my price. I'm $20,000 an hour just for me. That's what I quoted this morning.

I love it.

David George

I got one of those requests to help them look at this vibe code at the $20,000-an-hour rate. I got a yes. I don't know if it'll really happen, though. I just do it for fun, right? That's my price.

It's interesting that David Sacks, Marc Andreessen, and others were pointing out that the number of reqs out there—job specs for developers and engineers—is up, right? Are there just enough developer dollars and everything? It's interesting that 20% doesn't sound high to the scale analysis, but it's not going to come from a net headcount cut in developers and engineers who are AI-pilled. It's not coming from a loss of humans, right?

Jason Lemkin

I totally agree. To be very clear, this is what I don't know why people struggle with. I think—words I never thought I'd say—David Sacks and people are entirely right on this thing, right? Aaron says it really well. If 20% spend on tokens triples the effectiveness of your developer, then the ROI on developers goes up, so the number of developers will go up.

Yeah, that's the point people are missing. If the ROI is higher, then if you can attract them, the number will go up. You'll hire more.

Jason Lemkin

Yeah. Now you have all sorts of second-order effects. It's what you're pointing out, Jason: there'll be lots more competition and lots more bundling. I think software companies will have to cover a wider surface area because there's just going to be more competition.

David George

Products are easy to make, but there's no doubt that if 20% token spend—hypothetically, if you have a $200,000 engineer, and $40,000 on tokens doubles that person's productivity—and you had 10 engineers, my guess is you'll have 15. You're like, "These engineers are really good." Now, you might spend less in sales and marketing because your product will be better, so you might have more gross margin. You might have more in cost of goods because your token spend will go up, but your efficiency should go up.

Can I share one slight bull case on it that I didn't even realize until this week? People are definitely missing the point. If you can find engineers who are AI-pilled and AI-capable, you'll hire more and more of them because they're infinitely more productive. Of course you will if you're growing. If you're shrinking, the calculation is different. If you're hyper-growing, you will hire unlimited engineers if you can find people above the line with AI.

A funny thing is, we built this AI VP of marketing and AI VP of customer success at SaaStr. They're pretty good now. It took a while to get them pretty good, and I have to tell you, I'm too busy. I never looked to see what they cost each month in tokens and everything. I never looked until literally this week for the first time. I think it was Sunday.

What did you hear?

What do you think it costs per month to run a full-time, semiautonomous AI VP of marketing and customer success in tokens and everything? These are autonomous now. They're doing everything. They replaced many people.

Jason Lemkin

$2,800 a month.

Jason did not want a single efficiency maximizer there. So I'm going to give you four or five. I'm going to give you $2,000.

Jason Lemkin

So, $6,000 total for 2 full-time agents.

David George

You're right. There's no attempt to—

You're following me. $2,800. Oh, I'll give you $3,000.

It's like The Price Is Right, isn't it?

David George

Absolutely. I did think that. I mean, is Jason going to have you? Yes, there's a bit of gamesmanship in that answer.

Two of them full-time—2 full-time human equivalents—cost $254 a month.

Oh, wow. Interesting.

David Friedberg

$250. I sent this to Amelia on our team, and she's like, "Per day?" in Slack. I said, "No, that's the whole month."

It may tie to Notion saying that open-source LLMs are fine. It may be that not only is engineering the first and best use of LLMs, for all the reasons we've discussed, but the next wave are great too. They're just as good. We don't need as many tokens as we thought to replace Jason on the marketing team. We only need $254 a month for both of them running full-time, right?

It's going to be a little higher this month because it's a SaaS renewal, but not much. That's deflationary and weird at the same time. Amelia literally thought that was per day: $254 a month to run 2 highly valuable autonomous AI agents for marketing and customer success. Pretty crazy.

Chamath Palihapitiya

I would have guessed you'd be sloppier than that.

David Friedberg

I am sloppy. You get it. I don't have time. I don't care.

Chamath Palihapitiya

I don't know what to make of that information. It's odd because it would have been my prior, right? When we did some of the survey results on engineering spend, it's 2%—we found one at 15%. So it's not anywhere close to 20%, right? That implies that someone's spending hundreds of dollars a month, even in engineering, not thousands, right? So I'm trying to disaggregate that from the engineering—

David Friedberg

The caveat, obviously, is you don't have to refactor a massive code base 8 times a day to run a lot of agentic workflows, even very, very high-quality ones, right?

To be clear, it's kind of crazy. This AI VP of marketing we built now comes up with 3 great ideas every morning, which are better than any human is going to come up with. It's exhausting. It's so many ideas, right? It can do all of that for $94.27 in tokens last month.

It beats any human in ideas, not in execution. This beats any ideas that humans have for $94 a month. That's just an argument that, at some point, as we get into these other categories, they're just not as attractive as engineering for token consumption, right?

Chamath Palihapitiya

And that's why I had said I think it's 20% and 5%, but what you're saying is even that might be high, which is interesting. I'm frustrated at myself now because that would have been my gut. I will admit, because I've done the data and looked—for example, some of our companies are selling a software product that's LLM-enabled, and I said, "What's the token intensity there?" It's sub-10%, and that's on the COGS line, not the employee line, right?

I talked to our internal guy about what we're doing, and I'm like, "That feels relatively small." So my prior would have been around there. But if that's the case, I really struggle to figure out how they're doing $44 billion, right?

David Friedberg

That was my question. It's clearly true. It's just that I need someone smarter than me to take it and tie all the math together, right?

Chamath Palihapitiya

What it means is that there must be small numbers of engineers literally doing half the salary to get to that kind of number, right? The second-order question is, are they at the tip of the spear and this is the way the world is going, and therefore the 20% overall number will happen? Or are they token-maxing for performative reasons, and is some of this overdone? Could you get the same result with a more efficient use of tokens?

In which case, obviously, that $44 billion is not as much of a one-way street as I thought, right? That's why I go back to, "I don't know." The answer is evidenced by the fact that my guess was wrong and influenced by Twitter, but I do know enough to know it's the right question.

If you know the relationship between human spend and token spend for coding as the model load grows, you probably have a handle on what's going on, right? That's why we're surveying our companies as we speak.

David Friedberg

25% seems like a good framework for today. We'll see where the rest of the year takes us. There's so much. You could literally—some token-maxing may be performative, but imagine you could run a code review every single day, in real time. You're constantly finding bugs.

The agents already find bugs themselves in real time, but what if you could do meta-reviews constantly every day across massive code bases and iterate on them? It's easy to see this going up an order of magnitude.

Chamath Palihapitiya

Agreed, because that's the positive spin. The one thing you always have to remind yourself is that tokens get cheaper by the year. Every turn of the crank on GPUs gives you a 3×. Every turn of the crank on LLM optimization gives you another 3×.

You're looking at 10× prices falling every 18 months. So, in a weird kind of way, it might pay you to be inefficient now to get good at doing this, and you'll just become more efficient over time. So that's why I go back to this question: how much intelligence can you stuff into this economy, and at what price?

At what price? Roughly a $900 billion valuation, according to the latest. I've had 7 LPs ping me, saying, "Harry, you know everyone. How do we get in? How do we get in?"

Chamath Palihapitiya

Do we get shares? How do we get in?

How do we get in? Anthropic eyes a $50 billion round at a $900 billion valuation.

Chamath Palihapitiya

I'm going to admit, here's one where I was wrong 2 weeks ago. I thought they wouldn't have to do it; they should go straight to the IPO. But that was stupid old-world thinking, because the truth is, you shouldn't do it if, A, you have to spend a whole bunch of time doing a raise, and B, those raises give you rights around an IPO block or anything like that.

But we live in a world where they can just send out a freaking email, people respond in 48 hours, there's no drama, and they take the terms they get. So yes, they should grab the $50 billion. I didn't, and good on them, because—

Yeah, you had 48 hours to decide. Nothing is better than any IPO on planet Earth.

Chamath Palihapitiya

You're right, Harry. You know, we talk about public versus private. As long as you can raise capital like that in the private markets without any statutory liabilities or disclosure liabilities, why would you go public?

That was going to be my question. Does it do anything to the timing or the price of this supposed Q4 IPO?

Chamath Palihapitiya

It might do nothing to either of them, but what it does do, if you're the Anthropic CFO, is allow you to exhale. You always hate to have to do something that you have to do, right? There's no doubt that, had they not done this, they would really want to get public in the back half of this year. You could see a scenario where that's not possible due to circumstances beyond your control.

I think what it does is—there's a mild anchoring effect on pricing. I'm sure there are no blocks on IPOs, so, to a rounding error, we should assume these shares are powerless and illiquid and have no votes and no knowledge. There might be a mild anchoring effect to the high end, but the real point is it just gives you that degree of freedom. You don't have to do it.

One bit of math I did, and tweeted because you guys helped me think through it, is really obvious: if you have a company, every dollar of revenue that Anthropic generates means someone—either Anthropic or its partners—has to invest 3 or 4× in capex, because it's just expensive in terms of compute to service AI.

And then, on top of that, if you're going 10× and you have to forecast 1 year out, you're now guessing not your capex or your revenue today, but your capex a year from now. So, when you're doing $10 billion in revenue at a run rate, you're actually making capex predictions that might be 10 times that amount, times 3 times $3 per dollar of revenue. You're committing $30 billion in capex for every $1 billion in revenue you have. It's amazing.

A lot of that risk is laid off to the hyperscalers, but when you zoom out, my big aha—and that's why I was wrong 2 weeks ago—is that there is no such thing as too much cash on your balance sheet. There is no such thing. Dario is entirely right. This is the riskiest game of financial guesswork I've ever seen.

You're betting 10 times your—I mean, you're somewhere between 5 and 10 times your revenue at any point in time to meet the capex demand 1 year out. It's huge. There's never been a bet like this before, and the only thing you can do is de-risk the bet: raise capital.

But I do think it somewhat decreases the odds of an IPO this year.

Chamath Palihapitiya

Yes.

I think there are 2 factors. Just mathematically, if you can raise $50 billion literally in 48 hours with no rights or anything—and again, I've become Team Sam and OpenAI recently. With my agents, as my agents have taken control of my life, I've changed my allegiances to some extent.

But if OpenAI said today they were thinking about spinning out the hardware business they just bought for $6 billion, and with the slight drama between Sam and Sarah Friar, if OpenAI pushes out its IPO timeline and Anthropic is well-funded, they may be less of a rush to deal with the headaches of being public. So, if they really feel like OpenAI is a second-half 2027 IPO, this may be just for the sport of it. I think the combination of the two decreases the odds of the IPO this year. We'd have to check Polymarket.

Do we think that, after we've all said all along that both will go out this year, this is actually the first sign of true slippage and that both will actually go out in 2027?

Chamath Palihapitiya

I think the truth is now they don't have to. I mean, OpenAI, remember how quickly we forget, was $120 billion 6 weeks ago. Anthropic was $30 billion and now has another $50 billion. Neither of them have to go public this year.

I'm with you, Harry. I think you described it well. You still will, with a favorable wind, if things are organized and you feel you're predictable. You won't if you're not. And now you don't have to. So, in the right circumstances, they'd be crazy not to go, but you can't control the circumstances.

David George

First of all, you have another IPO pricing in advance of you that has way more risk and story risk in it in terms of SpaceX. So, you can imagine the world getting a little disrupted because of that. There was a war on, as a reminder. Lots can go wrong.

If you're the CFO of Anthropic, you go home after you raised $50 billion after 2 days' work and say to your spouse, "Good week at the office, hon. We got it done." Probably in Excel.

I logistically don't know how they do it with the amount of 10s, 20s, and 30s, and just the logistical challenge of collecting $50 billion from every family office and institution under the sun.

David George

The key is having a Brex and a Ramp account. You have to split it up between the 2, right? That's the insider trick.

You get the credit card points. I like—genuinely, that's why I think you end up with these minimum check sizes that are huge. That's why you end up with people doing bundling and SPVs so it looks like a single check size. You just end up with those structures to make it happen, right?

I remember speaking to one of the leading GPs at one of the leading firms. I said, "How much does it cost to take a meeting with you to get one as an LP?"

David George

Yeah.

$250 million.

David George

Yeah.

That was the entry price. I was like, "Wow, I'm really thrilled that you're on the show. I'm thrilled that you're on the show. That's awesome."

Speaking of OpenAI, OpenAI's chairman, Bret Taylor, is out in the market raising $950 million at a $15.8 billion valuation for Sierra. They're at $150 million in ARR, which is a 105× revenue multiple. On the negative side, it's a $400 billion customer service market; on the positive side, how do we read this raise?

David George

I'm starting to get worried. What I mean is, on the legal side, we have clearly, if nothing else, proven that the TAM is a little larger than we thought in agent tech. I'm not 100% convinced the CX market is as big as $400 billion, whatever you want to call it. I'm not convinced it's grown 10× because of AI. I believe it's grown a bit. I believe it might have grown 50%.

Everyone from Bret Taylor to Eoghan at Intercom sees these agents emerging, and they're going to do sales and CX and enterprise. I'm not saying it's not true, or that you're not replacing all these humans that we saw earlier. I'm not saying I'm even right, but I'm worried that the overall TAM is being flattered by the desire to reduce headcount. It seems to flatter the TAM expansion.

I just think it's TBD at this valuation whether there's a $100 billion company here or not. I don't know.

Bret, do you agree?

David George

My guess is the customer support software market that exists today is probably $20 billion or $30 billion, and the customer support labor spend is $400 billion. So, you're exactly right.

If you're selling a story that says we're going to replace the old customer support software with new—you're going to replace Service Cloud, which is the Salesforce product, which Bret Taylor obviously knows really well, with Sierra—that's not a great business, because you're going to be grinding out replacement for the next couple of decades, right? You have to buy into some level of this TAM expansion from labor replacement.

No, no, no, no, no—that's not right.

David George

Not any TAM expansion from labor replacement, but actually expansion into sales and upsell massively to—

Rory O'Driscoll

I agree. There’s a lot of—I agree because, in the abstract, you have TAM expansion from labor replacement, but once you have 3 or 4 companies competing for the same thing, then your competition is not labor. Your competition is 3 other companies, all of whom are using LLMs for the same thing.

So, I think, to state the obvious, there’s a fair amount of leaning in here at 100 times revenue in a world where you can buy Anthropic for, you know, 30 times revenue or 44, yeah, 20 times revenue, right? Just to state the obvious.

Now, I think the fun thing about it is that, when you have this whole dialogue—software is dead, right?—there are 2 sub-dialogues of “software is dead.” There is the SaaS apocalypse: all software is dead because there’s a SaaS apocalypse. And then there’s the more terrifying version: all software is dead because the LLMs are going to eat everything, right?

What I like about this is that this is the guy who’s chairman of the still-biggest LLM company, OpenAI, and they clearly believe that there’s value to be added on top. It’s the same story Alex Karp is telling: there is value to be added on top of LLM software to build a large, independent company. When people are giving me the “is software dead, are the LLMs going to eat everything?” story, people are voting with their dollars to say it’s not, right? With companies like Sierra. And I believe them to be correct.

One rule of thumb, going back to my failed math earlier but now trying to get it right, is that one of the things I look at in these companies is token intensity, which is how much they spend on tokens on a cost-of-goods-sold basis. In other words, how much does it cost to deliver a next-generation Sierra customer agent? My guess is their LLM spend is sub-10% of revenues. In other words, LLMs are not the dominant portion of the value they deliver. It’s the LLM plus the software plus the whole thing plus all the domain-specific knowledge.

So, I think the fact that they can raise this kind of money is healthy, and it speaks to the belief in the next-generation software companies. The multiple—you can definitely look, anytime you’re paying 100 times ARR for any software company, no matter how fast it’s growing, you really are leaning into a very aggressive future. I hope they’re right.

Rory makes such an important point here. The SaaS apocalypse assumed that no one was going to buy software. So, there is a counter-narrative to that, if nothing else. We want to buy SaaS, as is Palantir from this week, right? So, there’s meta-level good news, but maybe not for a lot of our portfolio. At a meta level, it’s great news.

Jason Lemkin

Are we really running out of ideas that much, GV and Tiger? My friend Tom Hulme runs GV, so I love him and I’m taking a dig at him here lovingly, but, seriously, if it’s a $100 billion company with no more dilution, you’re doing a 5.5x.

David George

Yeah, but we want Tiger to be back. That’s good for everybody, Harry. So, let’s distinguish between GV and Tiger, okay? We want Tiger to be deploying lots of capital into our portfolio companies. So, let’s cheer them on, okay? We need Tiger to be strong again.

Rory O'Driscoll

No, it doesn’t. I think, Harry, we’re all Pavlovian. Investors are the most Pavlovian things out there. The things that we do, the things that feel good, we do more and more of until it feels bad, right? And the truth is this: buying marquee assets at absurd prices has been by far and away the best strategy for the last 3 years.

True.

Rory O'Driscoll

So, you’re just going to do more, and you’re just going to keep doing it until you overshoot. Is this the moment they’ve overshot? I don’t know. I would have guessed $380 billion for Anthropic. Ooh, that feels a bit lofty, right? The point is this—

Jason Lemkin

The point is also, on the flip side, OpenAI would buy them today for $40–$50 billion to get Bret Taylor as CEO.

Rory O'Driscoll

Actually, I don’t think they need to. The truth is this: we’ve got to say it—something has happened at OpenAI. We suggested this a while back, and when I suggested it, I was also saying it shouldn’t be necessary, because really all we need is for someone to tell the team at OpenAI, “Come on, just stick to your knitting. Do 2 or 3 things, do them well, and get rid of the external noise.”

To be fair to them, you’ve seen some progress in that direction, right? So, you don’t need to have Bret Taylor run this company. I think whatever’s working, Mr. Altman has decided to focus a little bit and reduce the extraneous noise, and they seem to be getting better performance. I’m kind of, to Jason’s point, being a little bit back on Team OpenAI—not from a bandwagon perspective, but from an all-you-have-to-do-is-just-do-the-ordinary-things-well perspective, and you’ll do great.

Which has more upside if you were to put a dollar to work: Sierra or Anthropic?

Rory O'Driscoll

Anthropic. Not even a question for a second. Not even a question for a second.

Wow.

Rory O'Driscoll

I mean, yeah.

You’re saying it’s more likely that Anthropic is worth $6 trillion than Sierra’s worth $100 billion?

Rory O'Driscoll

Yes. I think the likelihood of both of those happening is low, but yes.

That’s interesting.

Rory O'Driscoll

You can play it. My sense is—I’m not good at the VC gossip, Harry. You’re much better at this than me—but my sense is they’re just different bets, because Anthropic is a bet that there’s boundless upside. Byron Deeter was on CNBC or whatever this week saying, “By far, this is the best round to invest in,” right? A little bit, he’s talking his book, but I definitely—we all know Byron. I believe he believes that this is the best round, okay?

I believe Sierra—I don’t know. I met Tom once through you when he was at SaaStr London. I really—but I believe that they all believe there’s also downside protection in this deal. And it’s just a different type of deal. People were kind of snippy when they did the round at $10 billion. Again, I don’t know gossip, but I was at a big event, and people were like, “Well, they did that round because they wanted access to Bret.” I don’t believe that, right?

But the sense that there’s massive downside protection, when folks are still struggling to deploy capital in venture, is important. It’s not easy. These bets are risky, and they’re very expensive. The fact—I don’t know how big the round was—$950 million? I can put half a mil—half a billion—into Sierra, and at least my downside’s protected.

You don’t want every deal to be like that in your portfolio, but there’s some comfort in having downside protection, even if it’s pretend. But it’s part of venture investing. Downside protection is real; we just overstate it, right? And the number of potential acquirers is real. We actually overstate that, too.

Jason Calacanis

Well, worst case, Vlad told me this the other day. I was talking about my portfolio, and he’s like, “Well, worst case, this one will exit for $1.5–$2 billion to one of these 3 folks.” I’m like, “Thanks, Vlad. I feel better.”

Let me explain. Yes, that’s because the training data doesn’t include the decades that I remember. Well, let me tell you, there are worse—there are—

Worst case, Jason.

Jason Calacanis

Worst case, yeah. Worst case.

Okay. In other news, Musk versus Altman, trial week 1. Roelof, going back to this that you mentioned earlier, Musk admits xAI distilled OpenAI models—partly. Importantly, he said partly. Greg Brockman also claims that his stake is now worth $30 billion. Another revelation that came out. Boys, what’s the analysis on week 1 of Musk versus Altman?

Roelof Botha

First of all, thank God for this gift: the tech version of TMZ. It’s going to be the gift that keeps on giving. It’s a little bit like rubbernecking at a car accident because it’s going to be impossible to tear your eyes away from it, right? And yeah, Evan’s going to come out, and it’s just going to—yeah, people aren’t going to look great.

Which is different from—actually, we should talk in a second about the actual legal issues here, which are much more distinct. But yes, in no particular order, the distillation comment wasn’t a good look for Elon. Being asked under oath to rank the models, and having to rank OpenAI and Anthropic above him, probably hurt deep in his soul, right?

It just shows that whenever you get to this kind of lawsuit, it’s always embarrassing for both sides. I feel for Greg Brockman, and you have this private diary where you write your inner thoughts. Suddenly, you get a document-retention request, and now your private personal diary, because it doesn’t have attorney-client privilege, is out there forever, and they just narrate you. It’s kind of—right?

On the $30 billion thing, people have been like, “Well, of course. The Twitterati are all, ‘Elon’s $30 billion. He didn’t put any money in.’” Well, yeah. That’s how equity works when you’re a founder, and he’s a founder. He is one of the top 2 or 3 executives in a company worth $800 billion. It would be surprising if he was worth less than $10 or $20 billion.

The most surprising thing is Sam Altman’s worth zero on this, right? Which I still think is weird and a mistake. It would be easy to troll both sides, but it actually doesn’t matter to the legal issues. The legal issues, I think, are going to be—you know, the one that no one talks about that’s super interesting is statute of limitations. Did Elon—

The earlier you see threats from Elon about, “Hey, this might be something I want,” the earlier it says you should have known then and you should have sued then. He may lose on the statute of limitations. The judge may just decide, “Dude, you have to bring a case within 3 years. 5 years have passed, so I’m not going to rule on the merits. It’s been fun listening to you guys for 2 weeks, but I’m ruling it out on that,” right?

Then another one that’s super obscure, but the minute you hear it, you go, “Oh, I get it.” A lot of the money came through his donor-advised fund, and anyone uses a DAF. It’s super efficient. You distribute stock into it, and then you advise the fund how to spend the money. But it turns out that the DAF is a separate legal entity, and once you give the money to the DAF, it’s not your money anymore. So the DAF is the person who’s been harmed when Elon’s DAF gave that money to OpenAI. Elon may not have standing in the case.

These are the legal issues that are going on underneath the surface. What’s going to happen here is that sometimes, especially in a pure jury trial, if someone looks like a jerk on the stand, it can impact them. Both sides are going to look like jerks on the stand, possibly, because at times they are jerks. But it doesn’t matter, because the jury is advisory in this case. In other words, they have the right to give advice on some of the issues, but ultimately, it’s kind of a weird thing: the judge decides.

So, in the end, this is going to boil down to 1 judge, probably a series of legal issues, and then, at the margin, the merits of the case. When you look at it, on the first day Elon looks maybe slightly better. Regardless of that, no one looks great—maybe that’s the truth. But in terms of the merits of the case, Elon probably went backward just because of these technical issues.

It’s hilarious. But I will admit, I say all that and I give this virtue speech about how we shouldn’t be looking at the car crash, and then I’m looking over at the car crash just like everyone else. It’s like, “Oh my God, you did that.” It’s just going to continue.

Jason, we can choose 1 final topic of the week. You get to choose, because Roelof likes to hand this over to us.

Jason Calacanis

Okay. Do I have choices?

Yeah. You have choices, dude. You can choose any of the remainders. There’s a lot in the VanEck Private Markets, Founders Fund’s new $6 billion fund, Coinbase cutting 14%, Vanta’s 63% growth to $300 million ARR, and Rogo raising at $2 billion.

Jason Calacanis

I’m going to give a quick shout-out to Vanta. We don’t have to spend the time on it. Another story of reacceleration of growth. It’s not just Palantir. We had Rippling last week, 70% of a billion. Vanta, 63%—I think every week we’ve got to have a shout-out. So, shout-out to Vanta, our good-news story of the week.

But I don’t think we need to talk about SOC 2 compliance this week. So why don’t we just leave it as a cheerlead?

Even though I hit it at the beginning, I do think, at least as we record this, that the Coinbase thing, if people really think about it, is pretty important and transformational. Brian is saying, “I just don’t need anybody at Coinbase who isn’t also an individual contributor anymore. I just don’t want anybody here,” is what he said this morning when we recorded this.

So, with a day’s delay, we do not need managers and managers, and we do not need managers. If you can’t ship and manage, if you can’t deliver a campaign and be the head of marketing, right? If you can’t manage sales agents and be a sales manager, I don’t want you at Coinbase.

That’s how I think all founders have felt. Then we give up at a certain point. We give up—it’s always true of the first 50, right? And then we start to give up. Around 500, I’ve learned, in the old days, you would just capitulate. Then you have managers and managers, and you start not meeting people as CEO before they leave, right?

I remember I was at Aaron Levie’s office when they crossed 500, and he said, “This was a learning moment for me. Now people leave Box before I’ve ever met them.” He didn’t mean managers and managers, but we capitulated to this.

Brian is saying, “No more with AI. We will even have teams of 1 that are self-managers,” right? I’m living that today. Every founder wants this world to exist where there is no one anymore working at my effing company who isn’t shipping, who isn’t committing, who isn’t building.

I think if this really hits, he’s putting into words what many of us have struggled to say: We don’t want managers anymore. You’ve got to build or you’ve got to go. Interesting. Build or go. Everyone wants this. Every founder wants this. It’s not the layoffs. Build or go. Build or leave.

What percentage of managers can also build?

Jason Calacanis

5%. They’ve all got to go. They build Kanban cards and talk about their team. Anyone on LinkedIn who talks about their team, fire them. “My team did this.” They’re all so precious about their team. “My team. My team did this.” That means they did nothing.

Lead from the effing front with AI. No, but it also says AI lets you lead from the front, and this is who we want to work with. This is who we want to invest in. This is who we want on our teams. We want folks who lead from the front.

Managers of managers lead from the rear. They lead from HQ, from their comfy office and their mug, and we’re done with it. I’m done with it. Anyone who talks about how great their team is, get your 4 months of severance and 2 weeks for each year you’ve been at the company. Good luck to you.

Roelof Botha

As usual, there’s a part of me that wants to recoil against this, but I’m coming around to your perspective. I’m going to start with Coinbase and Brian Armstrong, because a while back we talked about this, and I did a tweet on it that got a lot of pickup on all the performative, lying reasons why people are blaming AI for terminations: either because they’ve overhired, because growth has slowed, because they just spent all the money on capex, or, as you pointed out, because they need different people—all of which are different from the pure “I just can do more with less,” right?

So my assumption is, when people use AI as a justification for layoffs, I’m now mentally guilty until proven innocent: you’re lying. But—and this is the big but—I have to give Brian Armstrong credit. He’s demonstrated clarity of thought in how he thinks about hiring, firing, and company culture.

Go back to when we were locked in what turned out to be a fairly destructive period of conflating politics with how you run your company and bringing your political self to work, which proved divisive. He was one of the ones who early on said, “We’re not going to do that.” He didn’t say, “I believe in this and you believe in that.” He just said, “We’re not bringing that to work anymore.”

You remember that manifesto he did, and basically said, “I’ll pay you. I’ll give you severance if you want to go.”

To a lot of backlash at the time.

Roelof Botha

A lot of backlash, and in retrospect, an excellent call. It was an excellent call not only because it wasn’t a political call—it wasn’t taking the other side; it was taking no side—which has turned out to be entirely the correct position.

So my mental model is, when he says something like this, I’m going to assume there’s going to be less cant and hypocrisy in it than in the average CEO blaming AI for the thing, right? That’s the first comment.

And then, to your second point, you want to believe that managers should manage and, at some point, Alfred Sloan—the guy who built GM—was managing the modern corporation. But there’s a part of me that agrees with you, Jason, that if you’re not hands-on, you just don’t have a feel for it.

The best CEO at every level—yeah, you can’t be an individual contributor all day, because Jamie Dimon’s not out there making loans, right? But you have to do enough to know what’s happening on the front line with the new stuff. You can’t be so dissociated from it that you don’t understand, because then you’re in the grip of the experts, right? Today, that thing is AI.

If you take a different example, if you look back to 2008 and 2009, the financial crisis, and really examine what happened, those companies—the financial companies that went bust—it was like, “CDO, CLO, my 25-year-old kids are doing this, my quants are doing this, and I don’t understand, so I’m just going to accept—I’m going to be, as Jason says, a manager of managers, and I’m going to accept that they did the work correctly, and it’ll be fine.” They all went bankrupt.

I think Jason’s right today. If you’re not using the technology yourself at least 10% of your time, then you’re in the business of listening to other people tell you things that you don’t know are true or not. I always run my own models for our companies. I just want to see the numbers. Where does the cash go? It makes me a little old-school, but I think there’s an element of hands-on authenticity that Brian Armstrong is speaking to, and I think you’re speaking to it, Jason. I actually think it’s a thing, because without it, you just end up disconnected.

Jason Lemkin

I think it’s more. I think that was true last year. I think Brian wants more.

Jason Calacanis

Yeah.

Jason Lemkin

So let me give you an example. Last night, we had our AI VP of customer success reach out to 120 sponsors for SaaStr Annual, asking them exactly what their issues were, telling me everything they had to do, and getting it done.

You don't want a chief customer officer who has to tell 3 people to do that. Today, you want a chief customer officer who actually understands: Why is that possible? How did that happen? Why is that better than anything a human has done on my team? They can then talk to Claude Code or Replit or whatever and make it better tomorrow.

That's who you want on your team. It's not just that they dabble. It's not that they want to, because pre-AI, a chief customer officer could not reach out to 150 customers at 12:32 last night, as we did. Now with AI, our chief AI officer reached out to 200 and something customers at 12-something in the morning.

You want that person, or move them out. Promote the director who knows how to do that. It's a waste of time having that overfed person talk to the VP, who tells the director, who tells someone else, and it takes 3 weeks to do this, when we did it at 12:32 last night with 100-some-odd sponsors.

Point taken. The $10 million in revenue?

Jason Lemkin

They wouldn't ask for feedback; they'd ask for a meeting to get the feedback, which would take 2 weeks to get the meeting to get the feedback. Then they'd do it all over again.

And for sure, that's what's always been frustrating. But now we know that the best executives—let me make an even simpler version. A CMO today, and I know this will be really triggering to 90% of CMOs, but you know it's true, should be able to run their own campaigns.

And this is why. I'm not saying you have to spin up Marketo or HubSpot. I'm saying you should be able to tell your agent. Our agent, our AI VP of marketing, started running its own campaigns over the last 2 weeks, and it's better.

You don't even need to know exactly how that works. But the CMO should be able to interact with the agent and say, “Let's talk about the 3 best campaigns we should run to support scale, whatever,” and run them themselves now, because you don't need a team to do it.

You and the agent should do it, and you should want to do that. You should be passionate about it, and you should get rid of the executives in your organization who are resisting that. Get rid of them.

Yeah, I love it, but what percentage of CMOs can do that? 2%? 1%?

Jason Lemkin

Maybe you just need a director or VP who cares, and you don't need a CMO.

But the interesting question on that—let's say it's 10% or less—is: Do they become the most successful CMOs?

Jason Lemkin

Yeah, they're going to crush it.

If Jason is right and there's a correlation with success, then the one thing I do know is that economics is Darwinian. Over time, that 10% will become 20%, will become 40%, because the people who can't do it will be forced out. Right?

I think what you're saying, Jason, is that what it takes to compete is changing. If the people who have it succeed, capitalism works. It excludes the people who can't do it.

Jason Lemkin

Yeah.

And we might be better off without them.

Jason Calacanis

And that's why all the CX executives show up to the Palantir meeting: They know Jason's on the board, and he's going to put them against the wall if they don't know how to do this.

It's a genuine comment. It all ties back: If there's that level of imperative to do it, then anyone with an ounce of the survival gene in them will make sure that they're in the room when those decisions are happening and will want to be part of it.

I want to ask 1 question in a new round called “Rage bait but real.” I often tweet things and people think they're rage bait, and it's not. It's just genuinely how I think, and you can tell me if I'm an idiot or not.

Working from home on Fridays is BS, and it's an excuse for a 3-day weekend. Real or rage bait?

Jason Calacanis

Well, but you always rage-bait. They're going to be work-from-office days on Monday and Friday. They always have been. My logic was that we had our 2 partner meetings every Monday and every Friday in the old world, pre-COVID, and why should we change just because that happened?

Once in a while, we allow people—if you're on the road or if you want to make a longer trip—you can use 5 shifts a year and work from home on a Friday. But 90% of the people are in the office every Monday and every Friday, because those are the best 2 days to get people together.

My question.

Jason Calacanis

Real.

Yeah.

Jason Calacanis

Real.

The question is, is there a problem with it? Is that the question?

My point is, everyone said to me, “You're full of it. That's not true.” When you have 4 days a week in the office and Fridays working from home, yeah, it's an excuse for a 3-day weekend.

Jason Calacanis

Right. And the 90% of folks who don't want to work hard, and the 5% of the remaining 10% who do want to work hard but find it so hard that they think you're toxic—and Brian, well, at Coinbase, you've gotten an email to your personal email this morning: “Sorry, your Coinbase account no longer works. We had to do that to protect our customers,” it said in the email.

Your email no longer works. That's what's going to happen to those people criticizing you. They're not wrong in terms of quality of life and the way they want to live. They are right. You're both right.

They're all right that what you're describing is not the way they want to live, and you are right that they should not invest in those companies. Not only should we not invest as investors, you should not invest your time.

If you want your equity to be worth something, if you want to contribute to this world rather than just collect a paycheck, we are going to more and more bifurcate into folks who just want to collect a paycheck so the agent doesn't displace them and folks who want to change the world.

And that's fine. Let's not conflate the 2 like we did in 2021.

I love it. I like that. “Rage bait but real” is a new round.

Jason Calacanis

“Rage bait but real.” Exactly.

“Rage bait but real.” I've got more for you next week. Boys, thank you so much for doing this. Exciting week, as always.