Anthropic以9000亿美元估值融资300亿美元 | SpaceX提交S-1：将如何交易 | Cerebras首日大涨

As it’s becoming painfully clear now, no one in America other than us here in California likes the AI trend. We have people who are brilliant scientists who, politically, are utter morons. And the people who are utter morons at AI but brilliant at politics are going to have us for lunch.

Starting off, Andrej Karpathy joins Anthropic, and Anthropic eyes a $900 billion valuation for its latest fundraise. Then we dig into the public markets: Datadog up 31%, Figma up 12%. What happens from here? Next, Cerebras’ IPO smashes expectations and breaks the $300 mark. And then, finally, SpaceX: they set June 12 for the largest IPO in history, with a $1.75 trillion market cap and $75 billion being raised.

At least when NetApp was busy destroying the world, they were smart enough to pretend it was all about bringing friends together and not destroying democracy. We will regret that lack of transparency. We’re going to have to reflate and hire thousands and thousands of people per tech leader to avoid social unrest. We see no signs that there’s a short-term crash coming.

Ready to go, boys? It is so good to be back. We’re going to kick off with our “This Week in Anthropic.” Starting, as always, we have two. We have Anthropic in talks for $30 billion at above a $900 billion price, nearly tripling from $380 billion in February. Greenoaks, Sequoia, Altimeter, and Dragoneer. And then yesterday, we had Andrej Karpathy announcing that he was joining Anthropic.

So, boys, over to you. How did we read this news?

Jason Calacanis

Well, divide up the two. The financing—yes, we talked about it last week. There’s nothing more to say. They can pick their price, they can pick their investors, they can tell the amount, and they can tell Evan to jump, and Evan will say, “How high, sir?” So, it’s all happening. They’re going to raise $30 billion.

We discussed, obviously, the question of how those folks pencil out the return. So, Rory, I don’t mean to be old-school here, but I’m feeling old-school. Do you think ARR multiples still matter? I mean, if $900 billion is 18 times June revenue, it still feels like a better deal than any of the ones I did last year.

Rory O’Driscoll

There’s no doubt that statement is correct. Agreed. To put it really simply for listeners, you’re writing checks in the private market for companies with $10 million in revenue. You might be paying 20, 30, 40, or 50 times ARR, right? Maybe it’s 3–5x, but it’s 5 years away from an IPO.

And here are the last 3 rounds. For the last 3 rounds, really from the $150 billion round at Anthropic onward, they’ve been so close to an IPO that you can assume there can be an IPO. So, there’s no IPO risk. There’s no “it will go away” risk. Typically, when those risks don’t exist anymore, the only risk you’re taking is valuation risk.

The truth is, every time, the multiple on this has been significantly lower than the multiple on your median Series A, Series B, or Series C for a higher growth rate. So, it’s been the best trade out there.

David Friedberg

The meta question is this: At some point, everything’s DCF, right? At some point, it has to be. I guess I’m not even sure I believe that anymore, but certainly that’s the public-markets equation. Everything is ultimately the discounted present value of your future cash flows.

I haven't been a part of any of these Anthropic rounds, unfortunately. But every growth round I’ve been a part of has still been about ARR multiples at the end of the day. No one’s really doing discounts for lower gross margins like we did until, say, 2021 or 2022. So, I guess the meta question is: Is this a fair metric for Anthropic? Its margins are improving, right?

If 18x really is fair, geez Louise, I think everyone desperate to get it is right, because it’s the best deal going if an ARR multiple is still fair.

Jason Calacanis

You ask—sorry, it took me a while. The coffee’s got to kick in first. By the way, we’re doing this early, at 8:00 this morning, so I’ve been up since 5:00, but the coffee is only kicking in.

No, that’s a totally separate question. If ARR multiples are the right metric, then you should buy the one that’s at 18x ARR, growing 10x a year, and is already so freaking large that it will clearly and visibly go public. If ARR multiples are the proxy for value, then this is the best value in the venture universe.

Which is why very smart capital allocators whose mandate isn’t sector-specific but can range anywhere should stick their money in. People like Greenoaks and Altimeter are doing this deal because they’re thinking, “Yesterday I could do a $20 million ARR deal; today I can do a $50 billion ARR deal, and the multiple’s better when it’s so obviously a good deal,” as we mentioned, given the trade for investors.

Why would Dario and Anthropic do it at that price if it is so obviously a good deal? Because you’re giving away $30 billion over $900 billion, which is 3% of your business, to de-risk it for another year of monstrous burn, where you’re committing to, I don’t know, 6.5 gigawatts this year.

The mental rule of thumb is that the total cost of a gigawatt of high-end compute is $40 or $50 billion. So now you’re not spending $40 or $50 billion; you’re persuading hyperscalers to spend on your behalf. But you have to have—I mean, this is a big-ass balance-sheet war.

To me, it’s a no-brainer to do it. They’re going to raise, and they’re going to raise again, and then they’re going to raise again.

Do you think they’ll raise again before they go public?

Jason Calacanis

I doubt it. It just depends on the trajectory now, because they’re saying they’re going public in November. By the way, the other comment is that when you stop having to raise, that’s a disaster, because it means that you’re growing. Why are you raising? You’re raising for capex, and you’re raising for growth.

Once that hypergrowth stops and you don’t have a capex need, you also don’t have a growth story, and that’ll be a very different place to be. Right now, this is the highest ROI on equity dollars. So, that’s why they’re getting it. They should just keep raising well in advance of the need, because the needs are so great.

I know it sounds silly, and maybe it’s my actual read, but Sam has been pushing the valuation to the absolute max for as long as he could, on the thesis that he needed infinite capital. He’s always been clear on that, right?

The last OpenAI round was more expensive than almost the contemporaneous Anthropic round, even though Anthropic appeared to be out-accelerating them, because Sam just pushed it to the max, which you can if you’re a great salesman and have demand of one more dollar than supply.

It seemed to me Dario is the opposite. He actually does own shares in his company, rather than indirectly through a VC fund, but it’s pretty diluted. He’s giving away 90% to charity. So, he just wants to get a deal done in a week that is fair.

He does this deal at $380 billion. It seems fair. And all of a sudden, he turns around and Sam’s done a deal at twice the price. So, he does this at $900 billion. I think when Anthropic’s worth $2 trillion to $3 trillion, he’ll do a round at around $1.6 trillion in 48 hours. He’ll just do it. He won’t push it to the max like Sam does and create stress. But I actually think these rounds, ironically, the last two intentionally traded at a discount in order to not rip people off and get it done low drama in days. I actually think that was the trade-off. We’ve all worked with founders like this who enjoy maximizing every penny from the round, and others who want a 70% deal and truly want it done in a week—not just in the email, they truly want it done in 72 hours.

Rory O’Driscoll

Yeah, it’s a super point, Jason, because I’m remembering the details now. Compare the last 2 rounds—it’s so revealing. The Anthropic round is, “We’re going to raise $30 billion. It’s going to be cash. You’re going to send me an email confirming you’re in, and then we’ll collect the money. End of conversation.”

The OpenAI round is, “Well, Amazon, you’re going to give us $50 billion, but $20 billion is going to be upfront. The other $30 billion is contingent on us going public and our AGI. And Masayoshi, you’re going to give us $40 billion from SoftBank, but you’ve got to borrow $30 billion to get that $40 billion. So, we’re going to give you a little time to pay that money.”

So, we’re closing on $110 billion, of which $20 billion is clearing now and $30 billion in 6 months’ time, depending on the lending market. It’s like, “Jesus, give me a break.”

I think you’re right, Jason. Philosophically, the Anthropic team seems to operate under, “If I want to raise $30 billion, I should probably get a check for $30 billion and call it a day.”

Speaking of getting a check for $30 billion, Jason, I really wanted your thoughts on this one. Marc Benioff was on All-In, and he said that Salesforce spent $300 million on Anthropic tokens this year, almost entirely coding.

The question I have for you is: When you look at your usage and how you use it today, is that about right and what you would have thought? Is that way more? Is that way less? And how do you think that will change for Salesforce over time?

Jason Calacanis

It’s actually not that much per engineer. This is great. Marc’s one of the great classic marketers, probably the greatest marketer in classic B2B of all time. He’s just a force of nature on all vectors: company, physical size, gravitas.

Rory O’Driscoll

I think that works out to about $15,000 to $20,000 per engineer per year.

Jason Calacanis

Yes.

Rory O’Driscoll

It’s not that much. That’s, I think, just table stakes today. What’s his fully burdened cost for a developer at Salesforce? Probably $500,000 for an engineer, all-in, with their share of the building and snacks.

So $20,000 a year is 4% additional. Cheap, man.

Jason Calacanis

Good. First of all, I did the math on this one because I actually think this is the most important question about Anthropic. I skipped the other 2. This is the one. And actually, I'm going to give you an A, Harry, for math on the fly, which is very hard to do. I couldn't have done it on the fly; I did it this morning.

Right. This is—and it turns out $300 million is astonishing, first of all. So, the numbers are—

Salesforce spends $5.8 billion a year on engineers, right? So it's roughly 4% of the spend. If you want it per head, they have 20,000 developers out of their total 83,000 employees, so it's $15,000 per head per year, which is $1,200 to $2,000 per head per month—$1,200 per month, right?

We actually did a survey. I'm not sure if I'm front-running it; one of my colleagues is going to publish it. We surveyed 40 portfolio companies and external companies: What are you spending per year, per month, per developer? The average was $1,200 to $1,300, and the median was lower, right? So obviously, you have some token maxing and then a wider dispersion.

So, first of all, you're exactly right. It's in the strike zone of normal. It's not—I mean, it's only the big number because they obviously have so many developers. So your rough estimate—the math—is exactly right.

Then the question is, what does all this mean? Where is it going? And I want to say, first, before that, it is still astonishing. One of the reasons that Anthropic and OpenAI are just amazing businesses is that 90% of the stuff you sell to Salesforce as a vendor—if you were selling them an ERP system, obviously they have the CRM system—my guess is, if they bought SAP, God forbid, would they spend $5 million or $10 million?

This is the only product that, for a company like this, is probably the only vendor line item other than maybe rent that comes anywhere close to this amount. From nothing 2 years ago, this is the largest single external spend that every software company is making. That's the first big aha, right?

Jason Lemkin

And—

It's both nothing, and it explains Anthropic's mediocre rise. It explains everything. It's both at the same time, right?

And then you have to say to yourself—because remember, I kind of said it when it comes to valuation on Anthropic, is it good or not?—the question, when you try to figure out how much money these companies can make, is that coming up with some kind of heuristic relative to R&D spend is the key. In other words, how much of every knowledge worker's wage and how much of every coding wage is going to get translated into tokens?

We did a rough-and-tough estimate, and we're refining it more. If you start thinking about a trillion dollars' worth of token revenue across Anthropic and OpenAI, which is what the 4-year projections are saying—and they better get it, because otherwise that capex is going to look pretty sick—if they're going to get a trillion dollars, my rough math says it's something roughly like 5% to 7% of every knowledge worker's salary and 20% of every engineering salary.

So, if that math is correct, that's kind of what it takes to get to a trillion. In other words, Salesforce might only be a quarter of the way there. One of 2 things is going to happen: either they stay at $300 million, in which case the TAM for some of these token businesses like Anthropic will have been overestimated and there'll be a real correction, or they 4× their token spend from here. In 2 years' time, Benioff is on saying, “We spend $1 billion on tokens,” and we'll talk about the people consequences for that in a second. But one of those 2 things has to happen.

Even on the macro level, worldwide software business across everyone is about $1.2 trillion. R&D spend is roughly 20%: $240 billion. The interesting thing is, if you get 20% of that, you only get $50 billion, right? So, to justify these Anthropic valuations and these OpenAI valuations, you're really going to have to eat a ton of what is otherwise opex. You've got to replace 20% of R&D salary.

So either those valuations are wrong, or Benioff is only a quarter of the way on the journey, and he's probably ahead of most. The numbers for OpenAI and Anthropic are so large that you really have to start thinking about what percentage of the total wage bill for engineering in the software development market you get. If you're not tracking to 20% across most R&D spend, then the 3- and 4-year projections for some of these companies will be a bit lofty.

Jason Lemkin

At a meta level, you have to be a bull, right? Because the trend has just begun. Marc's $300 million is just the start of what he's going to spend.

On the other hand, I do think—I hate to use myself as an N=1 case—but at SaaStr itself, I mean, we now have 21 agents, of which 3 are autonomous. The direct token cost that we spend—the direct AI cost altogether for both of us—is about $2,000 a month. That's going to go up, but the base case is the models will get better, they will get more efficient, and we will get more efficient. The base case is $1,000 for each of us.

Listen, this doesn't include third-party apps. It doesn't include tokens we buy inside of Salesforce, so it's higher. But if you think about it, there is a base case there that everybody is using, that every knowledge worker has this attached. The numbers Marc is throwing out are about right for folks not at the bleeding edge of token maxing. This is the base case. It's not today, when SAP and Uber CIOs are saying, “We're out of tokens for the year,” but I do think we're ahead of most. At our little team, we're only spending $2,000 a month in direct token costs.

On the next Anthropic round, I think I'll give you another example. We had SaaStr Annual last week. Rory was a celebrity, Harry. We could talk about it. He was literally mobbed. You saw the pictures.

Our very last speaker—it was kind of him to come because it was the last one and people were tired—was Andrew Bialecki, the co-founder and CEO of Klaviyo. It's very interesting because he's a true engineer and a turned B2B founder. He requires every single employee at Klaviyo, if they're anywhere close to product, to be committing code. Anyone in product, anyone in design, everyone there—and every single person has to be running AI or agents to do their job.

I couldn't believe it. It was 100%. It is. And they built their own custom framework to require it. He went through it all. It was very cool. My point is, he knows his stuff.

We built this AI VP of marketing, AI VP of customer success—everyone thinks it costs $8,000 to $10,000 to run these autonomous agents. I go backstage with Andrew, and we're talking about it on my phone. He's like, “How much do you think it costs to run an AI VP of marketing?” Because he's done it. He's like, “Maybe $250 for both of them.” The answer is $257 just to run the agents.

His point at the end was this: a lot of this stuff is not as expensive as we think. We do not need to worry about token maxing at Klaviyo, and everything we're doing is agentic. We have our own agentic framework. The most junior product person—every single person has to be doing this. We have to manage it, and they have a harness that manages the model and gets it to be thoughtful.

He's like, “If you do this right, it's not as expensive as—” And the fact that he guessed it within a couple of dollars—I mean, he was the only person who got it right because he's doing it. The only one who got this number right because he's doing it.

So, I'm not bearish, but this is the base case: we need half, a third, or a quarter as many tokens as we think we do, outside of the folks running massive workflows 24/7, right? At 1% of R&D spend, it's lost in the noise. At 5%, it's real. That's a layoff, right? At 20%—which, let me repeat, is what it takes for these overall TAM analyses to work—that's huge. It's 1/5 of your payroll costs in engineering.

Jason, every public company CEO wants your advice on agents and AI.

Jason Lemkin

Yeah.

Are you more bullish on Klaviyo after seeing the inner workings of Andrew? And is he a top 1% public company CEO on AI?

Jason Lemkin

I am. It's a good question. The one thing I've been thinking about a lot recently is that, when we kind of bounced off the lowest point of the SaaS apocalypse, you think about Atlassian, Figma, and a few others that have seen, at the end of the day compared to their highs, very, very modest bounces off the hard bottom. But, more importantly, growth is reaccelerating there. That's the most important thing.

When you see Figma reaccelerate to almost 50% growth, when you see Atlassian—seemingly struggling when Mike was on the show—reaccelerate north of 30%, the meta question is: with Atlassian, it's definitely from Rovo, their agent. With Figma, it's a mix, right? But when you see Twilio come back from the dead to 20% growth, you have to ask yourself: is there a little more time than we thought? A little more time, right?

The whole world is not in San Francisco. Think of all the buyers and all the users. So this is the question for Andrew, for Marc Benioff, and for others: these founder-led companies that are iconic, with great CEOs, at the end of the day, maybe they have enough time—maybe another year—if they're just getting going on their agentic journey.

Salesforce is further along than I am. I'm not sure that means their stock will reaccelerate until it is proven. I think that's what we learned from last quarter. Monday.com did sort of beat expectations and bounced. HubSpot said Q2 was going to be tougher, and it got hit hard.

So, you’ve got to show me the growth, right? That’s the mantra. But I am somewhat more bullish than 90 days ago that there’s just time. Klaviyo is arguably the single most beaten-down public company software stock because of the delta from Shopify. It’s trading at 3-something times revenue, and Shopify is at, what, 12 or 14? I’ve got to look it up.

If I were a long-short team, I might propose that one on Monday. But you’ve got to show the growth, man. You’ve got to show it, and they’re not—even though he’s ahead of the internal agent, they’re not way ahead of the game for the external one. And that’s the bear case, right? If your competitors are there, why aren’t you there today? Why don’t you have this?

So, yes, I’m optimistic, but my flip side is you’ve had 18 months, and you’ve had since December, since Claude 4, to destroy your space. Why do you let these dumb little startups out-hustle you? You’ve got 2,000 engineers spending $300 million a year. That’s the bull case: you’ve had time. But I’m getting more optimistic that if the leaders build the best agents in the space, 2027 could be good for them. I’m getting more optimistic, and I was pretty bearish a couple of months ago.

I am in the same place. I made some comment about that, and positive reinforcement from Datadog and Figma, and everyone’s like, “But they’ll never get back to 2021 prices again.” I commented on that because, of course, they won’t, right? As I said in my comments, they’ll never get back to 2021 prices, and I’ll never be 21 again.

Yeah, I like that.

Rory O'Driscoll

There’s nothing you can do. Every tech cycle has a set of industries that, once in their life, get valued on prospects and futures. It’s like being 21. It literally is like being 21, and people will believe everything about you. 5 years ago, that was SaaS. Today, it’s AI.

Once you lose that veneer, you’re going to be valued for the rest of your life on some variety of revenue growth and cash flow, right? And what that means is you’re highly—it’s almost impossible to ever get back to 50 times ARR again, right? All these companies—and everyone goes, like, it’s the Figma thing. It’s so annoying for them. I feel so bad for them: “Oh, but your stock’s down 80%.” Yes, from the idiot price that idiot people priced that at, right?

In terms of objective performance, which is how you’ve got to measure these companies, you’re right, Jason. Somebody’s like, “Figma’s reaccelerating. Datadog’s doing really well.” You’re going to be in a band. The outer edge is probably Datadog at 17 or 18 times sales; your good performance is Figma at 6 to 10 times sales, and your crap is 3 times.

So, to me, it’s, yeah, you have time to become a good, normal company. I mean, the big comment is the SaaS businesses are amazing, but the AI businesses are an order of magnitude, or maybe 2 orders of magnitude, larger and are earlier in their growth life cycle. So you’re never going to get the attention back on you again. That’s the deal. That’s the deal for SaaS, but you can still be worth $10 billion and, as a company, do $1 billion in revenue and grow nicely.

Just to provide context, Datadog had their first billion-dollar revenue quarter, 32% up, an all-time-high ARR, and crossed $4 billion. This was a great quarter. Ollie and the team crushed it. So, to the point, it’s now more realistically priced from its exuberant pricing. Is that the summary?

Rory O'Driscoll

Yeah, it’s the summary. Exactly. And in that context, it’s worth trying. I mean, the difference between being in the shadow and not getting out, and being in the penalty box and then getting out, is quite significant.

I mean, you’ve got to like where you are. As Jason said, if your team’s Atlassian, you’ve got to like where you are a lot. If you’re Datadog or Figma, you’ve got to be pretty depressed about where you are. If you’re Wix, given you did the buyback and it hasn’t worked, there’s plenty—

We’re going to discuss Wix. I do just want to cover Figma first, though. Jason, you’re always rather opinionated on it.

Jason Lemkin

But accelerating for the second straight quarter, NDR 139%, a 2-year high. This was a great quarter of results. I sold all of mine at the end of last year.

Well, you’re still ahead. That was the right time to sell, to Rory’s point. I’ll tell you what I got wrong on Figma for sure. Like, dumb-dumb Lemkin. I’ve got to come up with something asinine with the L, right? A Limited Lemkin or something like that.

Rory O'Driscoll

Loser Lemkin.

Jason Lemkin

Loser Lemkin. I like Limited. Can we go with Limited? Can we be a Limited Lemkin?

Rory O'Driscoll

Limited Lemkin.

Jason Lemkin

I was completely right that Make is the worst vibe-coding product I’ve used since I’ve been on this journey. And I’m right that it was not important to senior management, and I’m right that they left $500 million or more on the table by not building a Replit-level competitor. I’m 100% right, and I think it’s the tragedy of folks being slow.

However, Limited Lemkin—I missed the Captain Obvious point. Figma is building software. So everyone that is in the business of really helping—there is a software explosion, but part of the explosion is a software explosion, and you can see it in companies that Harry and I have invested in, like WorkOS and RevenueCat, that have exploded because there’s a software explosion going on.

Even though Figma lost the vibe-coding race so far, it is a beneficiary of the software explosion. One, right? Two, it has internal AI tools, like Andrew from Klaviyo was talking about. One, it’s selling credits to make your product a little better. That almost sounds cynical, right? But what it’s really rolling out now is the ability to internally vibe improvements to your Figma designs.

The agent can look at your Figma design. It actually officially rolled out, I think, today. It’s been in beta for a while, but it rolled out while we recorded this. Instead of just designing something, it can say, “Hey, let’s update the workflow in here. Let’s update the journey.” This is not incredibly difficult, but it is difficult at Figma scale.

So, even if they’re not going to help you vibe products, the fact that they are making the building of software more efficient for their audience—you know, it’s like Atlassian not adding massive new customers because of AI, but adding massively more value for their base. So, they’re a little slow to that. It’s in beta today. I mean, it’s getting to be summer, but it looks like it’s pretty good.

And ultimately, Limited Lemkin insists that anyone like Figma should be modestly accelerating today because we’re building more crap. Their new ability to agentically improve design likely will be a big deal for their customers. They likely can get another 50% or more of revenue out of their base.

The question is, to the extent that folks who would have been using Figma are now doing mockups for software products using Lovable—which you hear a lot about, using it as a way to describe your product rather than doing a Figma—that’s actually a workstream that you would want to own if you’re Figma, because what you don’t want to have is people going around your product flow, right?

So, I do think you’re right, Jason. I think it’s not just the extra $500 million. I’m willing to bet they probably have an imperative to make sure that their customer doesn’t leak out to a design flow where you do your actual design in something like Lovable, where you have a working prototype, not just a Figma design.

Jason Lemkin

So, that one, for the moment, appeared to be more of a Twitter mania that overstated the design capabilities in Lovable, which is a little bit ahead of Replit, but they’re pretty limited for anything that’s professional-grade, like for a Figma person.

What Figma missed is: I create this design. It’s beautiful. I loop in the product team to approve it, and then I click a button that says, “Push into full production prototype,” and it just works. The irony is Replit and Lovable both have that as a native insertion point for a reason. If you use these products today, they have, right in the prompt, “Upload a Figma design,” because they know this is their number-one ICP: someone who wants to take that static design and put it into production.

Why Figma doesn’t have that natively is a loss of $500 million and going up. It totally is.

On the flip side of these 2 great quarters and kind of exciting, happy news, Wix is down 45% since the stock repurchase. Rory, you mentioned it and touched on it. They’re now a $2.2 billion market cap. Base44 announced last night, actually, that they hit $150 million of ARR—

Rory O'Driscoll

Which is because the business isn’t growing. The pre-AI business is no longer growing, right? Maybe it’s not. I think the pre-AI business is terminal. That’s what the public markets are saying, right? And they don’t think that the AI play is enough to rescue it from terminality, right?

The $150 million from Base44 is very impressive, right? But it’s substitution revenue at a high level, isn’t it? Because it hasn’t materially grown the revenue. It’s substitution. If you were to assume that it’s a terminal business in terms of the core business that’s existing, forgetting Base44, surely you would then ascribe the same to Squarespace, which—

Jason Lemkin

I think it’s terminal as well. I think Wix and Squarespace—one thing people forget is that, if you look, it’s true: not only are they being terminated by 2 vectors, one is obvious, one is less obvious. The obvious one, which is more true than people think, is that it’s already better in many cases to vibe-code your own website because you get what you want. They already have templates and integrations—not for folks that are truly tech-fearful. They should still use Squarespace and Wix.

Rory O'Driscoll

They're great products, but they're so limited, and you can build something in a vibe-coded platform so nicely in 10 minutes. Anyone who's not tech-phobic should use these products. The other thing that we kind of forget is that Wix and Squarespace, for the last 5 years, were low-end Shopify competitors. That's where their growth was: merchant services, payments, and e-commerce.

There was a whole world 4 or 5 years ago where you had WooCommerce from WordPress, Wix, Squarespace, and BigCommerce, all of which were viable competitors to Shopify. The other thing that happened is that Shopify destroyed them all. There's no reason to use them. This is another category where the low end was destroyed by Shopify, which amazingly went upmarket and downmarket successfully at the same time.

It's not worth it. You can't save enough money by not using Shopify for your store. The folks who are trying to save $14 a month have just evaporated from the low end of the market. They don't need these sub-Shopifys. And BigCommerce, even though it's not low-end, got destroyed too, right? The most visceral competitor was just dead. It's dead in the water. So they got hit both ways. It's too bad. It's two currents hitting them.

Jamin Ball

I broadly agree. I think BigCommerce is right in Shopify's strike zone. I don't know the Wix mix between information-only sites and e-commerce sites, but you're right, Jason, it really matters.

And it was all the growth. Got it.

Jason Lemkin

It was all the growth. Interesting, because the core—I would say this for the core—a lot of those businesses have to be customers. It's super-low-end SMB, especially for the non-e-commerce sites. If your core acquisition engine works, it is still possible that you can convert most of your customers to the new product. Over a couple of years, you can maybe make the math work, right? If you're Wix, at least they have a new product. They did do the acquisition, right?

You get this weird compounding because when the new product is $100 million and it's doubling, and the existing product is a couple of billion and it's flat for the first year or so, it's really hard to move the overall aggregate GAAP revenue growth rate. But if you can compound quickly enough on the new product, it does move it up. That's on the revenue side. I want to separate the stock buyback from that.

That sucks because you do the classic investor-banker thing of, “If you buy back the stock at 4 times, it's cheaper than it's ever been, and that'll be good for the stock.” It sounds like it makes sense, but when shit's going wrong, things can go lower than you ever think. Sometimes it pays to keep the money in your back pocket because the stock will go even lower, right?

I think that's just a strategy that didn't work. You took on a bunch of capital, you bought the stock back, and that strategy didn't work. By definition, if the stock is down 45% since that moment, the buyback strategy didn't work. Frankly, it's better to have bought it back cheap than bought it back at the crazy-high prices at which some people were doing stock buybacks in the past.

But when your business is in trouble, things that are bad can get worse. The actual ability to have another $1 billion on your balance sheet to maybe make an even bigger acquisition, in my view, is worth more than trying to juice the stock for the short term. So I think they look back on that stock buyback and say, “That wasn't the best move to make. We optimized for the short-term value of the stock, not the long-term destiny of the business.”

The flip side is that Marc Benioff said they did a buyback. What did he borrow, $20 billion or something like that, to do it? Not even at the lowest interest rates, but he said, “I did it to offset Slack and Tableau. I did it to offset the dilution. I got it back.” Right. My theory.

Rory O'Driscoll

What does that mean? I actually challenge that. I hear him saying, “We're doing it to offset stock-based compensation,” and I want to beat my head off the wall. The only reason you buy stock back is because you think it's way cheaper than it should be, right?

If the stock was trading at $1 trillion, would you buy it back at $1 trillion to, quote, “offset Slack dilution”? I think it's a bogus answer for people who are just trying to manage some second-order metrics when it just doesn't make sense. I've been on boards where bankers pitch doing a buyback to offset stock-based comp, and I literally want to bludgeon them to death.

Right. Well, buying at high prices is stupid. Buying stock at low prices is clever. End of analysis. I think a lot of these deals, and maybe even Salesforce, are really just to hold off shareholder activists because that's their first play.

When you're down in the dumps, you've been on boards like this, Rory; you've watched it. You want to, to the extent you can, placate the shareholder activists without giving them what they want. The simplest thing you can do is use all your cash to buy back shares because that's their play. If you take them out of the game, you've given them what they've asked for, and they don't really have a play if they buy you—especially if you're also doing layoffs.

So if you're doing layoffs and buybacks, maybe you keep Starboard and friends away because their playbook has already been used up by the management team.

Rory O'Driscoll

Agreed. And look, if there's nowhere better you can put the money and the stock is cheap, then it pays to buy it back. But I do get the concern that you're always afraid of management with big amounts of cash, that they'll waste it on a bad acquisition.

On the other hand, in these times of change—hyper-change—you probably have 2 choices as a SaaS company. You want to decide, “I am the old thing, and the old thing is good enough. I'm going to optimize for 30% operating margins and 10% growth, and I'm an economic machine. It is what it is. If I have excess capital, I should buy the stock back, give it back, do whatever.”

Or you say to yourself, “I might have to do something more than that. I don't know what it is, but at least for the next 12 months, I'd like the option value of knowing that if I want to do a Base44 or whatever acquisition it is, I have the money.” But I agree with Jason's cynical comment. You're right: people do it because it's one of the least disruptive activist moves that you can make, and normally you get brownie points for it.

The odd thing is that if you do a buyback of the stock to appease the beast and then the stock goes down 40%, they won't remember that they wanted you to do a buyback. They'll just say you're an idiot for doing it. In the end, you're paid to be right. When you buy and then it goes down 45%, unfortunately, you can't say to yourself, “You were right.” Today, they're trading at 1× revenue. 1×, literally.

Yeah, that's the terminal—simplifying the terminal state. They also announced 2 years ago that they were shipping a Glean competitor, but I'm not even sure it exists. In a year's time, will they be higher than they are today or lower than they are today?

Rory O'Driscoll

I'd say higher, just because, again, I just think at 1× revenue, unless their revenue is absolutely evaporating. I would like to look at their churn numbers; I haven't spent enough time. You can get away from 1× revenue, especially if you had the wit and intelligence—which they did—to buy a kind of Replit competitor.

Most mature software companies that aren't growing can easily operate at 20% operating margins. So you're now down to saying you've got your cash, and do you think you can last 4 or 5 years before you go to 0? You can calculate the terminal value here, right? At 1× revenue, you are nearing terminal value for a company with a product that has high gross margins and is relatively sticky.

So yes, I think you can create more value. You will deeply regret buying a whole bunch of shares at 3× revenues or 2.5× revenues, thinking it couldn't get any worse. But yeah, that's fatal. Fatal next sentence: “Maybe now it can't get any worse.” If it's 0.5× a year from now, or the revenues have gone down by 50%, then you can call me an idiot, Harry, which I know you're dying to do.

No, it can't get any worse. I'll go and buy a load of Wix and then remind you of it every week.

Rory O'Driscoll

Remember, “can't get worse” is not the same thing as “it will get a lot better.” The question—a totally separate one—I think Jason's right about the upside in these stories. Again, it goes back to the same thing: your range of outcomes has compressed markedly.

It's not clear to me if someone said, “What would you do to get a company—let's not pick on Square—to 30% growth and make it worth 5× revenue?” I'm not sure I have a single idea. I think Jason's right. You're dealing with managing decline. As they used to say about the British Empire, Harry, you're managing decline. It's okay. It's history. You wouldn't understand.

I do think one of the areas that we think is truly downbeat—public software companies—will become upbeat in the next year.

What I mean is, as slow as they've been to react to AI changes, they have the installed base. Someone will get their moment, and it won't be anybody that isn't founder-led. One of these founder guys will put his best 50 people in a room and say, “Listen, I don't even need you guys to innovate. I just need you to build a better version of a prosumer AI app. I just need you to build a better version. Stay out of it and ship it, and we're going to sell the hell out of it to their base.”

Arguably, that's what Canva is trying to do with Canva 2.0. It's complicated. The startups are moving faster, and the models will move. But someone's going to pull this off with their best 50 people because they have 300,000 customers.

For every time I'm pitching these AI SDRs and I'm looking at HubSpot with 300,000 customers, I'm like, “Hurry up, Breeze, because you've got 300,000 HubSpot customers just waiting to buy an AI SDR from you.” If HubSpot could actually make this as good as a startup, they're going to sell 150,000 Breeze AI agents. It just isn't in the market today.

I just think there are so many challenges, it's hard to predict, but someone's going to—you'll turn around and you'll be like, “Holy cow, Drew pulled it off. Drew went from 0% growth. He got his last 50 guys, his last 50 soldiers holding back the castle, put them together, and holy cow, he built this thing.” But I don't know that we can predict who it is.

On the flip side of the challenge, we have Nebius growing 684%, accelerating faster than ever. My question is: is it justified? Is this absolutely the sign of just compute starvation and a buy, or is this, bluntly, further evidence of a bubble and overt market exuberance?

Jason Lemkin

You know, it's a mini CoreWeave, and those have been great businesses because right now everyone's compute-starved. Let me make the captain-obvious answer, as Jason would say: if compute continues to be starved, then these will continue to be good businesses. If it doesn't—if compute becomes plentiful—they will be commodity businesses, and the guys who are over it will go bust. So it's just that simple.

I actually think Gavin Baker had a very articulate comment. Ironically, the slowness of permitting and the inability to bring on data centers at near the speed that people want to bring them on might save us all from ourselves. If all the data centers people want to build could be built, and at the same time we do the Anthropic and OpenAI math that we just discussed, and that $1 trillion of token revenue turns into $500 billion, then you've got $1 trillion of capex and $500 billion of revenue.

But if, on the other hand, the $500 billion of revenue stays while the inertia of building data centers means you only get half of them built, then you're saved by the bureaucratic inertia of the great American state, right? You only build $500 billion. Compute remains relatively scarce, and people like Nebius and CoreWeave that have that compute do really well.

Does Nebius' spend grow faster or slower than data center capacity? That's it in a nutshell. With OpenAI and Anthropic in the middle, collecting the money from the first and giving it to the second, that is the bet. I don't have a brilliant opinion on that, but that's the action on the table, as they say at the craps game.

Jason Lemkin

You know, I was driving back. We had a little Napa retreat after a disastrous year, and the Tesla took us the long way through the South Bay, which I haven't done in a while, over the bridge. I'm passing Marvell Semiconductor, SanDisk—all these folks, all these superstars of the '90s. They're on fire today.

I should have mentioned Cisco, coming up South First or Zanker. Every technology company except traditional software is on fire. The old South Bay, where Harry's probably never been—he's probably never been south of Mountain View or Palo Alto. He's like, “Why would you go? There's trillions down there, man. Look at the skyscrapers.”

I was thinking that the only thing that isn't inflating today is old-school software. Everything else is on fire. So my point on Nebius is—and Rory made this point—listen, there's an argument to short it. There's an argument that there's just surplus, that there's not enough capacity in the market. But how far in the future can you short these things?

You can short SanDisk in memory, and we short Micron on the drive. You can make fun of Micron, right? Micron's had more booms and busts than a California prospector from the 19th century. But you can't short three years into the future effectively. Certainly, I don't have the skills to do so. We could take potshots at Nebius and CoreWeave and friends, but what's the point? We see no signs that there's a short-term crash coming.

Anything but traditional software is just on fire, right? Every single technology company—Cisco's back. I should have mentioned Cisco. Every software and technology company except traditional software is on fire. Going back to the big, simplistic picture here, all these companies are on fire because the hyperscalers and the model companies have decided to spend roughly $750 billion to $1 trillion a year building. Fifty percent of that goes to Nvidia, 10% goes to power, and 10% goes to networking. Everyone just gets pulled along in the bubble.

Everyone, right?

Jason Lemkin

One of 2 things has to happen. Either corporate America has to digest $1 trillion worth of tokens without any intermediate software layer—and I think it can do some, but I don't think it can do all—or software has to start working, because only if software starts working does corporate America get to spend that kind of money.

If Sierra doesn't grow, let me put it bluntly, then at some point OpenAI and Anthropic will stop growing, because their customers today are primarily software companies using them for coding. Selling through software companies to corporate America to use tokens for business purposes means that, in a weird kind of way, at some point this all has to level out.

I just think, when we started this show, CoreWeave had just IPOed or was about to IPO, and it was easy to mock CoreWeave. It was like, “Okay, well, this is just round-trip revenue to create a little supplemental capacity that we won't need in a year,” right? Fast-forward a year, and we need everything possible.

I do think at some point SanDisk, Nebius, and CoreWeave all have to crash, and Marvell and even Broadcom have to crash at some level, because they always do. I think AI can grow infinitely—we will all approach the singularity—but eventually capacity will catch up and things will catch up. I just don't know if it's going to be near enough to the present that it matters.

Jason Lemkin

I don't know if you follow Leopold Aschenbrenner, who's famous—

Yeah. He doesn't have to worry about 10 years out, does he? He'll just trade in and out of it.

Jason Lemkin

But if you saw his latest releases and his latest filings, he put puts across everything. The guy, for the first time, added very little and actually showed his first real sign.

He's smarter than me. He's making the opposite point I'm making: that this future is coming much sooner than I think it is.

The tough one is, would you invest in a CoreWeave or Nebius at the seed level today? That's the tougher bet. This is still 20VC, right? As a public investor, you can say, “Hey, SanDisk looks pretty good for the rest of the year,” right? As a startup investor, would you do one of these deals?

Rory O'Driscoll

I'll answer that. I saw a really excellent one with a superb team and a very good seed investor. I'm not going to name it. I really considered it long and hard. It was still a seed round. We tend to be A investors. It was a really talented team.

I didn't do the seed, but I always try to give a good answer to people when I turn them down, especially when I think they're A-class teams, because I say, “Hey, look, here's my thinking.” Hopefully they'll remember my thinking, and if they prove me wrong, they come back for the A.

I will admit, when I wrote out my thinking on why I'm not doing this and he responded, I thought he won the argument. So I'm actually sitting here going, “Rory, was I an idiot?” He had a compelling, at-the-margin story around capacity that I thought was interesting.

I literally ran—I did, to answer your question, Jason, logically, I chickened out. I didn't want to be that model-capacity investor 3 years into the deal, 3 years into the capex boom. But there's a little part of me thinking, Rory, that might be a dumb decision. That was a clever story with a clever team.

You are relying on the capital markets being there for the next 3 years and being able to access them. The sobering number is $500 million to $1 billion to build the capacity, right? But maybe it could have worked. I get the temptation. I literally had this happen last week.

Rory, do you always give detailed explanations to founders? I remember Jason once saying to me that founders will always argue back, and it's easier to be like, “Hey, keep it much more vanilla.”

Rory O'Driscoll

Yeah, I think it depends, honestly, because you can't do it to everyone, right? To some extent, it depends on how much time you spent with them. It depends on 2 things: how much time you spent with them.

If you've taken 1 meeting and you're a no, just give a clear no. I mean, maybe you give minor feedback. You can't write a long email, right? If you've taken 2 or 3 meetings and you felt you were almost there, sometimes I like to do it because I think it's helpful for yourself.

Some CEOs don't respond well to detailed feedback and they argue. Some of them are really professional about it, and I like them. I'm like, “Yeah, I see what you're saying. If I'm right and you're wrong, I'll be back in 12 months and I'll say to you, ‘I told you so.’” And you'll pay 3 times as much. And I'm like, “And I'll be glad to.”

I think, in the end, even if you don't share it all the time, for any deal you spend a lot of time with—and this is a separate comment—it's actually very helpful to write out your conclusions and keep them internally. I do that, because then you can test your thinking. You look back 2 years later and think, “Oh my God, I turned down that deal for that reason. I was an idiot,” or, “I nearly did that deal and I was totally wrong on the market.”

Remember, in a model that only gets trained in 8- or 10-year increments, based on the 2 deals you do a year, you get a lot of additional feedback from the 20 or 30 deals you nearly did that you just don't want to lose. So I do try and write it out for myself, and then sometimes I share it with the team, especially if I think it might be of interest.

Jason Lemkin

Yeah, I think that is a good point. I did say that to Harry years ago. I stand by it. If you've had 0 or 1 meeting with a founder, there's no upside in providing feedback. It's just an endless, “Wait, Jason, that's wrong. You misunderstand.” I just—if you've gone deep on a deal, you should share the real reasons why. It's generally appreciated.

If nothing else, if they haven't already closed a round, it's helpful to them to see the other side after 2 or 3 meetings. But it's got to be at least 2 meetings. You have to have gone deep enough to actually be able to write that email. It can't be, “I just don't see it. I just don't like it.” Then it's got to be 1 line, right?

I think I know the deal that Rory's talking about. But we shall move on. I want to talk about Cerebras, the biggest US tech IPO since Snowflake, priced at $185, which was a big expansion from where it started. I believe it was $110–$120 in the early days, which went to $150 and then $185. It popped 68% on day 1. It was a fantastic IPO, an amazing story.

Does this open the window for many more companies of that size? Not the $2 trillion companies, but does this open the window for many more companies that go out and IPO?

Jason Lemkin

I think it's good for SpaceX. I think it's good for anyone above their level. I think it's great. It just shows anything that is that—or better, in air quotes—the demand is infinite. I'm not sure if this is really going to help folks that are below that level. We don't know, right? I doubt it.

Even when you're looking at Figma, you're like, “I mean, I don't think anything sub-Figma can IPO and have a decent IPO.” This is the new grade: better than Figma, BTF. Now, Cerebras arguably is a different category, but if you look at the backlog of $24 billion—and you're Pollyanna about it—that's more backlog. What's Figma's backlog? I don't think it's $24 billion, right? I'm oversimplifying it, but I think it's got to be better than Figma. If you're better than that, you know, this is a good time to IPO.

Rory O'Driscoll

This is very much an N-of-1 bet. It's an extraordinarily complex technological product that they've brought to fruition just at the time when demand for that product has exploded. Unlike when they pulled an IPO 2 years ago, they were able to line up arguably the marquee customer for that product, OpenAI.

Semiconductors are hot; they're a semiconductor company. Inference is hot; they're an inference company. OpenAI is hot; they're selling to OpenAI. It's an N-of-1 positioning. And, for a market that's starved of ways to bet on OpenAI and Anthropic, they really only have things like CoreWeave, and obviously Nvidia. This was a chance to play.

So, yeah, I'm not surprised it went. If you remember last week, you asked, “Is it going to go really well?” We recorded it before the IPO and appeared after the IPO, and I was like, “Of course it's going to go really well. They've already raised the range. They're not idiots.” And it went exactly like that.

They went beyond even the range. They went to the maximum they could do without refiling it. It was an obvious winner category. It's worth pointing out how fickle the world is: 2 years ago, they couldn't get the deal done. So I think you're right, Jason. At the margin, it's a positive tell for SpaceX. People are very much risk-on for the kind of things that look like they have that kind of upside.

Rory, would you add it to your public book at $300?

Rory O'Driscoll

Probably not. I go back to the base rate. The base-rate return on an IPO is not from the day of the IPO—not from the day of pricing, but from the first day's closing price, which is typically way above it. The base-rate return on that is negative at 1 month, 6 months, 12 months, 1 year, 2 years, right? In other words, across 1,000 or so of them, if you buy on the pop, you tend to be a happy camper, right?

Is it a company at the right price that you believe can be a long-term, enduring company? Absolutely, yes. It's got technological differentiation like no one else, right? So, just blindly buying the day every other retail idiot on the planet is buying is probably not the best way to make money, just statistically. And base rates matter.

Totally agree. We said it's good for SpaceX, and SpaceX sets June 12th for the largest IPO in history: a suspected $1.75 trillion valuation and a $75 billion raise. My word, this would be epic.

Jason Lemkin

It will. That's 1 word for it. And, again, how does this go? I don't know. I mean, it's funny: we're recording this the day they're due to file their S-1, but I haven't seen it. I checked before I came on.

The interesting thing about the S-1 is, in 1 sense, you really want to read it. In the other weird sense, there's actually going to be very little in it that actually matters at the margin. What do I mean by that? The S-1 will tell us—and I'm really curious to read it, because it got through the SEC really quickly—everything about SpaceX as it existed in December of this year, which was without xAI, without the Cursor deal, and without the Anthropic deal.

If you think about it, the most recent year-end—I presume their year-end is December—will show last year, which is SpaceX and Starlink standalone. The leaked figures are $15–$18 billion in revenue, a 20–30% growth rate, and EBITDA-positive. I would like to see the capex before I comment, but it's pretty much a bounded, understood company, right?

In February of this year, they closed on xAI, which brought them a pitiful amount of revenue and a burn as big as Croesus, you know what I mean? That's going to be prorated into the S-1 for maybe 1 quarter, so you literally have half a quarter's information on something that's taken you from a profitable company to a loss-making company.

Then the other 2 big deals—the Anthropic deal won't even be in the financials because it's a signed deal, and the Cursor acquisition won't even be in the financials because it's not closed yet. So you're literally going to be reading this S-1 going, “Here's the company we used to own on December 31 of last year. Pretty nice company it was too, dude.” However, we since got xAI, and now it's totally different. By the way, we can't tell you much about that. You'll have to talk to our bankers.

It's going to be the funniest S-1 ever in 1 respect. I mean, 30–40% of the revenue isn't in the S-1; all the loss isn't in the S-1 except for half a quarter. Some of the bankers are going to have to tell the story via the roadshow.

So, in 1 sense, I'm looking forward to reading the S-1. In the other sense, there's just a lot to come. That's the first comment: the storytelling around these acquisitions isn't going to be in the S-1, and it's just going to be interesting how they get that across.

There are other markets and other times where people will look back and go, “You must have been mad to buy the stock on that little information.” So I think it'll—again, the probability is that it gets done extraordinarily well, and the excitement is amazing because the market is in the mood for excitement. We're selling the most exciting company on the planet at a time when the market wants excitement.

If the market ever wakes up and says it wants cash flow, it's going to be a totally different story. Right now, we're into excitement.

June 12th, this goes out. It's Elon. It's Elon, the pump machine. Does this have the mother of all pops, with retail getting behind Elon in a way that we haven't seen before? It's got to do better than GameStop.

Rory O'Driscoll

No.

Jason Lemkin

Yeah. I think there's a reason it's 30% retail. I think some of it is Robinhood and democratization, right? He's selling 30% to retail. But I just think the Robinhooders and the GameStoppers have got to be more excited about rockets than plushy toys at GameStop.

To me, it's more exciting.

Jason Lemkin

I agree. I think everyone will be wild. Everyone will want to own some of this, which means retail will buy a lot of it.

I'm going to put $2,000 on my iPhone into this thing.

Rory O'Driscoll

The reason I said no is because, remember GameStop—I go back to numbers. GameStop, I think, pops 10–30—yeah, 20x from low to high, just based on retail. When you start at $1.7 trillion and the largest market-cap company on the planet is $5.5 trillion, Nvidia, it's going to be hard to 10x from here, right? That's what I meant. But you're right: the excitement on retail will make this a super-interesting story.

You know, literally, Rory, I'm completely ignorant. Obviously, it's going to be a huge float, right? Mathematically, could the GameStoppers and the Robinhooders—I'm admitting my ignorance—make it go 10x? They're not doing any DCF analysis; they're trading. Is it possible for them to trade enough shares to create a 10x pop? Is it mathematically possible to influence the float that way? I ask because sometimes it's a thin float when you're able to manipulate it, right?

Rory O'Driscoll

The float isn't that small. It's $75 billion. I think the interesting thing is, if it's—it's already healthily priced at 100x revenues. It will be interesting to see what happens.

Harry, you can go no comment at this point. When institutions take shares in an IPO, they have a price target. If that price target gets achieved on the first day, you tend to see additional trading.

If BlackRock spends $10 billion buying in the IPO, as has been rumored—but again, I have no clue about the facts—they run an internal analysis and say, “We think we should buy $10 billion because we think over the next 12 months we can make 40% on our money.” If, at the end of the first day, it's up 40%, it'll be sorely tempting to have another $10 billion come back to market because you'll be looking at your price target and going, “I have a price target. I've achieved it. Time to go,” right?

You see that phenomenon when an IPO pops. Institutions—I used to think, “Oh, my God, they're disloyal. They'll leave.” But in fact, it's just that we bought because we wanted to be a holder, but we had a price target. You've achieved it, so it will be interesting to see that price action if it does, in fact, do a GameStop-type of price.

You could also have the Facebook effect. I'm just going to say it, not to be negative: the IPO was frankly a dismal failure. Early on, it hung around its price for a day or two, and then—look it up—in 2012, it dropped at one point 40–50% below its IPO price. It was a horrible IPO, right? Obviously, an amazing company, which can also happen. Price matters.

I think it's going to be wild. I genuinely hope it succeeds because I think the dampening effect of it not succeeding and not trading well would be pretty profound.

Jason Lemkin

I think it'll trade up to $5 trillion. I'll take this bet. But I think there will be enough Bitcoin Miami day traders and Yahoos who love the brand. Folks hate the brand; that's why he lost the trial in Oakland. I think they hate the brand too, but enough folks love the brand that it can float up 3–5x based on partially influencing the float.

Rory O'Driscoll

There's not enough. There isn't enough demand. This could be Limited Lumpkin speaking, but I'm going to take this bet that it's going to trade up 3x in 2026, just based on GameStop driving it up.

I think it's going to go to $3 trillion.

Rory O'Driscoll

I don't discount the fact that it goes down.

I don't discount the fact. I didn't say it's likely. I'm just pointing out here: you're paying $100. Where's my Robinhood account?

Jason Lemkin

You boomer. You boomer.

I am the boomer.

Jason Lemkin

$5 trillion.

Rory O'Driscoll

I think it's going to go down.

I didn't say—hang on. I didn't say it will go down. Harry, you've got to be able to live in probability.

Jason Lemkin

Did I tell you when I got into Bitcoin?

Yeah.

Jason Lemkin

I've made billions on my Bitcoin. This is going higher than Bitcoin.

Okay, good for you.

Jason Calacanis

You do realize that space is bigger than Bitcoin, right? The entire universe—there are a trillion stars just in our own galaxy, and there are a trillion galaxies. That is larger than all the Bitcoin out there. You have this completely wrong. Did I tell you when I got into Bitcoin? Did I tell you guys when I got into Bitcoin?

Okay, thank you. I'm just trying to be serious.

Rory O'Driscoll

That's an instant pass on a founder in a pitch, telling you when they got into Bitcoin. That's my flip side of too much feedback after zero. If, in the first 20 seconds, all they tell you is when they got into Bitcoin, tell them to stay in Bitcoin.

The reason I made that comment is—and I know it sounds like, again, I go back to Facebook. I remember the IPO. Facebook was the defining company of its generation. It was far more attractively priced than here. It was profitable, right?

But they pushed the limit on price, and they just hit that point where the initial trades went the other way. There was a little worry about mobile, right? They hadn't managed a mobile transition, and it really traded down over the next 6 months. So I'm simply saying, again, it's back to my point: when everyone thinks something is guaranteed, that's just when it blows up in your face.

Do I think it's the likely outcome? No. It's probably two-thirds positive, and of that, at least 30% of it is an adjacent outcome, which is that it pops amazingly because of retail. But again, I go back to the fundamentals: it's trading at 100x revenues. They've effectively decided to be a Nebius, not an Anthropic, by virtue of selling their compute to Anthropic. So that whole AI story isn't there. It's Starlink, and Starlink was the growth engine, with a CoreWeave attached.

And it's been a while since we had a bet. Rory, this is great. Jason's at $5 trillion, I'm at $3 trillion, and you're negative.

Rory O'Driscoll

No, I—okay. I mean, there's probably some kind of spread betting I can take on that. Let me think about it. I'm comfortable saying it'll be below $3 trillion. I mean, I'll totally take that bet.

The Price Is Right bet.

Rory O'Driscoll

I mean, basically, guys, you said $3 trillion, Jason said $5 trillion. I do not think, at the end of a month, this company will be valued at $3 trillion or more.

Oh, no. Neither do I.

Rory O'Driscoll

No, this is a meme and this is a casino. We're seeing the casinoization of public markets. This will go faster to $3 trillion, and it'll come fast back down.

I don't do casino betting, but duly noted.

Rory O'Driscoll

What are you doing in venture and AI, then?

Yes. Okay, keep rolling. Keep rolling.

News last night might be a big moment with Y Combinator, and I do think it's actually important. Sam Altman just offered $2 million in OpenAI tokens to every YC startup in the current batch in exchange for equity. It reminded me of Yuri Milner and DST doing the exact same with the very early YC batches. What do we think about this? And does that impact the valuations that we ultimately get them at if they can get $2 million in OpenAI tokens from Sam?

Rory O'Driscoll

First of all, it's all smart. I think it's back to you: you have let Anthropic steal a march on you—more than a march, many marches—on people, on mindshare, on respect, and you have to do all you can to earn it back. That's just one more thing: develop our hearts and minds. So smart.

Second, I'm assuming it's not transferable, because if it's transferable, then it's money. Let's be real: compute is money, right? But I'm sure they've thought of that, so it's not transferable. So, does it impact pricing? Maybe at the margin, but if you're compute-intensive, it does. But probably, if you're compute-intensive, you're raising $200 million anyway, right?

For the most part, if I think of the last YC batch, most of them are building software on top of AI, but where agentic spend would be—token intensity would be 10% of revenue—it's valuable, but it's not going to replace the need for humans. You're still going to need 4 or 5 humans to build the code, to build the agent for call centers or whatever. So they're still going to need resources at the margin.

It takes a little bit of the edge off, but it's not like any of them can build a next-generation whatever with it. It's nice at the margin.

Jason Calacanis

I think it'll increase the valuations for sure.

A little bit. Yeah.

Jason Calacanis

Because if nothing else, even if you don't view it as inflationary, they have $2 million of tokens now. That's a real investment. Let's take this seriously: we all can use the tokens. So that is $2 million of de-risking of that investment—$2 million more that they can use to add value to the deal.

There are plenty of YC companies that don't raise $2 million at Demo Day, right? So now they've raised another—now they've radically de-risked these investments. It would make sense that a typical post might go up to 60. There's some correlation I'm not smart enough to do.

It may even GameStop higher, since they're investing at 100. It's hard to predict. You can come in, Harry. If you do the deal a month before Demo Day, it's 20. If you do it the week before, it's 40.

Rory O'Driscoll

If you do it after, you're at the OpenAI price. It's $100 million. You can come in at the OpenAI price; you're welcome to. It's not even a premium. We'll do it because we love the pod. We'll let you, Harry, and Jason all in at no premium—just at the $100 million price.

You're right, Jason. If you navigate on optics, it probably causes an anchoring effect: “I got $2 million at $100 million. Why would I take another $4 million at $50 million?”

Jason Calacanis

It may shrink the size of the rounds, too. It may make it even harder for VCs to invest in YC rounds, because the average ownership for VC rounds has already been sliced to 5% or 6% at YC. This could slice it to 2% or 3%, just because you don't need as much capital. That might even be the bigger impact, potentially.

Rory O'Driscoll

In fact, depending on how much of your burn—remember, at scale, if you're successful, any software company can use up $2 million in tokens without blinking—the interesting question is how much leverage there is in that. I'd love to know, in the first 12 months of the typical YC company's life, how much of its spend is tokens to serve customers, tokens to build the product, and engineering spend that can now be replaced by tokens. It all goes back to that 20% number. Can you replace—because out of the gate, you're probably not selling so much that you're reselling those tokens. Most of what you're probably doing with those tokens is building your product. So if you can—

Maybe if you're building a Lovable or Replit, you could burn through all those tokens in 12 months.

Jason Calacanis

Yes, serving customers, but if you are—can I make a comment? If you are given that token intensity for a Jasper or Legora is probably 20% in terms of revenue, that probably means to burn through $2 million, you're going to be at $10 million in ARR. If you're at $10 million in ARR in today's world, you're going to raise at $500 million anyway.

You're exactly right. I was mentally putting it into 2 categories: token spend for engineering and token spend for full-price customers. You're exactly right—the minute you sell it, everyone will have destructive free-token programs because you want people to try your product. There'll be a bunch of premium products. You're exactly right; that's how it manifests. If Sam increases it to $4 million, $8 million, or $10 million as OpenAI grows, think about how much that could change the game. If, as a startup, you get $10 million of tokens to build another Lovable or Replit in your first year, then you're beat to the wall because you don't have to worry about anything except shipping the best 5.7 Codex product you can, because there's no issue. The first year is all about marketing. For so many of the startups we invest in now, tokens are marketing—

Rory O'Driscoll

Their tokens are marketing, because poor Michael Cannon-Brookes has built an iconic company, but he can't afford to spend the number of tokens a startup can spend per customer.

Jason Calacanis

Give me $2 million; that could just be the start. I think this is already disruptive to 200 startups, and why couldn't it go up? Well, it can if you have spare capacity. It's also very telling. At the margin, if you're tapped out on capacity, that's $2 million times 150 for Anthropic. If they really are capacity-constrained, that's 150 startups; that's $300 million you're giving up every 3 months. If you say 4 YC batches, that's $1.2 billion a year. At an 18× valuation multiple, that's about a $20–$30 billion hit to valuation. That's real money, right? If, on the other hand, you have spare compute, then it relieves some of your costs.

Yeah, but if you believe in the YC model, worst case you're going to hold it at 1×. Once cash is less of an issue, you're going to at least be able to hold these investments at 1×. It's not going to cost you anything, because if the average batch allegedly does 3× to 4× but you're paying $100 million, then at least you don't have to mark it down. Part of it—

Jason Calacanis

Respectfully, up until then, you were more right than me on that. You're wrong because if OpenAI's investments are valued at 1× and no one will give them any credit, but if, to make that investment, they gave up revenue from selling those tokens to Bank of America—

Rory O'Driscoll

Of course you're right on that. It only sort of works if the tokens are surplus or leftover or something like that.

Jason Calacanis

And therefore my conclusion is OpenAI has surplus tokens and Anthropic does not.

I think that's a good conclusion. I think it's also a bet, but it's also a thoughtful bet, right?

Jason Calacanis

It's a smart way to use them. It's a smart way to use them. Okay, what else are we—

I just want to hear, before we do a rage bait for real: Jason, Rory is a celebrity at SaaStr.

Jason Lemkin

Oh, God.

Rory O'Driscoll

They loved him.

Don't give me—

Jason Lemkin

Loved. He was 8–10 rows standing-room-only deep to hear from this guy. They loved him.

They told me—did he get selfies?

Jason Lemkin

Okay, he did the one on stage with me. He wasn't that into the selfies, but, yeah, he was meh.

Okay. I hate this. Keep going.

Rory O'Driscoll

Oh, yeah. No, it's okay. Do we want to do—

Jason Lemkin

In all seriousness, I do think it's a reminder that there's a large thread of Rory super-fans, right? They like the thoughtful deep dives. Rory's got a few skills. Harry and I are self-aware; we're not claiming we're something we're not. Rory has a set of insights and skills that I'm very self-aware of.

Rory O'Driscoll

I would say to the LPs listening that they at least justify a 6× fund. I would say, at a minimum—would you agree?

Jason Lemkin

At least, I'll give him premium carry at a bare minimum. Right?

At least a $199 fund. I'm cutting this out. We're going to move on to Rage Bait in a second.

Jason Lemkin

Celebrity.

I just want to point out in passing, because you mentioned it, that we did call the OpenAI–Musk lawsuit correctly: not just dismissed, but dismissed on a technicality—the exact thing I said last week. The jury said, “Yeah, they just said I could in two.”

Jason Lemkin

My wife used to be a public defender, and one of the real tells was how quickly the jury came back. Literally, 2 hours. She said they went in, had lunch, picked the foreman, and said, “Look, we can decide on the technicality of the statute of limitations. We can be done in here in half an hour, or we can waste a whole bunch of time arguing facts beyond that. What does the vote say? We're done.”

Rory O'Driscoll

I'm going to push back and say, first, my wife—it's been years since she's practiced, but she would say generally that juries are smarter than you think, and it's kind of patronizing to think people would just do what they want. There have been some places where I could argue that's happened, but in the main, juries are pretty sensible. They listen and try to do their civic best. I actually feel—this is where I'm going to be a little touchy-feely—I feel people try to do their best, right?

I think in this case, the real truth is this: it was a no. Elon knew that they were talking. Stepping back for people, the question came because the conversion had already been blessed by Delaware and California. You can't say the conversion was illegal. So the case you were making was that the other 2 guys were fraudulent when they didn't tell Elon about the planned future conversion and took his money on false pretenses. They were alleging fraud, right?

That fraud took place in 2016, 2017, and 2018, and there's a statute of limitations on a fraud claim. So if Elon only found out about the conversion to a for-profit when it happened in 2023 or 2024, then it's within the statute of limitations and his claim could proceed. But it was obvious to anyone with the brain of a pea that, given he was discussing a conversion to a for-profit back in the day, it was a ludicrous allegation that he didn't know.

That's why it went to a jury, and also to a judge, versus just being a black-and-white thing: it was about knowledge. When did you know? In the case of fraud, it's when did you know you were defrauded? That's when the clock starts for the statute of limitations. But it's pretty clear he knew.

I think the real truth is that anyone sane wouldn't have taken that case—they wouldn't have been the plaintiff. Elon just didn't care. He didn't do it based on the probability of winning. He did it because even if he doesn't win, he can damage the other side. He's really angry and pissed off about what happened. And if he were worth $800 billion, so what if I wasted $40 million on a case? I yanked everyone's chain.

I'm happy. So my take on this one is that justice was served. Elon got his pound of flesh for the $40 million or whatever he spent on legal fees. He's going to appeal it, but it won't get a second past appeal and it'll go away. I actually think it came out exactly as planned.

But then Jason actually found one of the news stories that came off the back of this, which I didn't actually see, which is that Elon spawned dozens of other investigations into Sam Altman's finances on the side, creating more problems for Sam.

Rory O'Driscoll

I think that's true. A separate comment is—and these are words I never thought I'd say—I feel empathy for Sam, in the sense that he hasn't taken any equity in OpenAI. He's been asked about that, and he said he's had no economic interest in OpenAI. We probably all know he doesn't own any equity.

But what's going to happen now is that a whole bunch of people, including in congressional testimony, are going to say, "But you have an ownership interest in Y Combinator, which has an ownership interest in OpenAI. You have an ownership interest in these companies that are selling to OpenAI, and therefore you're nefariously trying to get the money."

In one sense, he may have been factually incorrect to say that; in the other sense, obviously he's not, because if Sam wanted to get 4% of OpenAI, the board would have given him 4% of OpenAI. Whatever he gets indirectly is minuscule compared to what he could have gotten.

The reason all this is biting him in the ass is this whole, "We're doing it for the good of the world. I'm not getting paid." What I find enjoyable is that all these good intentions are biting him in the ass. It's kind of unfair, right? Had he been Larry Ellison and said, "I'm doing it for the money," it would all have been clear, right?

So yes, Elon is going to be able to continue to make OpenAI's executive team's life a misery, which clearly makes Elon happy. Not only that, but other people are going to be able to pile on too because of the complex nature of OpenAI's structure, in a world where a much simpler structure wouldn't attract any attention.

I mean, to be fair to Sam, he was the founding idea behind OpenAI; he convened that meeting. If he'd taken 10% ownership on day 1, or 10% ownership when the conversion happened, no one would blink an eye, right? It's one of those things: good intentions bite you in the ass.

Jason Lemkin

Well, maybe let me add just 2 final thoughts. We could go forever. I've said I'm on Team Sam now. More importantly, I'm on Team OpenAI.

But he did have consideration. He set up an entire venture fund where he got all the carry and claimed it was OpenAI. If Elon keeps this going, this will reverberate forever. The idea that Sam took no consideration from OpenAI is the biggest load of malarkey, because he set up a venture fund on the side, probably without telling the board. This is probably why he got fired, and he kept all the carry.

And you know why he did this? Because he didn't think OpenAI would be worth anything as a nonprofit. So he said, "I want to do this. I'm deeply passionate about it, but I also want to monetize it. How do I do this? Do what I did at YC: set up a fund on the side and keep all the carry. Call it the OpenAI venture fund, make all the investments, and keep all the carry. That way I can at least make $800 million like I did on Stripe."

Rory O'Driscoll

Ironically, if in fact you're correct, it's evidence of a belief that OpenAI is not going to make any money, which ironically would actually have helped his case. He could say, "Hey, that's right."

What it really points to is complex arrangements. Complex arrangements bite you in the ass, because I go back to my comment that once it became a for-profit, if you wanted to just be a paid CEO, you could have gotten your ownership. You didn't need to do all this other stuff.

And you're right, Jason: when you do all this other stuff and then you make an enemy of the richest man on the planet, who is clearly malevolent and willing to go to the mat for this over and over again, you're in trouble.

Then add to that something we haven't talked about, but which I think goes back to where you're wrong on your comment on the jury. They didn't find for OpenAI because they found OpenAI more sympathetic than Elon Musk. I think it's becoming painfully clear now that no one in America, other than us here in California, likes the AI trend.

If you asked the jury what they thought of all the people involved, they would probably say, "A curse on all your houses. What a nasty, obnoxious, arrogant, entitled bunch of shits. But we did our job, we followed the law, and I hope I never see these buffoons again. I hope only bad things happen to them."

Probably that was the jury. Yeah, that's probably right.

Rory O'Driscoll

That was the jury, right? And now, can we get our lunch and our daily stipend? Are we done?

Did you guys see Eric Schmidt? Eric Schmidt got booed. I think that goes back to my comment here. We've spent 3 years with the leaders of this thing telling us how it might destroy humanity and put us all out of jobs, and then we're shocked to discover that people don't like us.

Oh, and by the way, your electricity is going up in the meantime, but have a nice day, right? In general, we have people who are brilliant scientists who politically are utter morons, and the people who are utter morons at AI but brilliant at politics are going to have us for lunch.

That's the movie in the next 3 years. They're going to have Sam for lunch because he's lied to them, as far as they're concerned. And they're going to have the AI industry as a whole for lunch because we're firing people left, right, and center. The politics are going to be brutal.

Jason Lemkin

Yeah. This is why I'm on Team Sam. I think he's doing the best balance he can here. I think it's mostly a positive image. He's not doing the Dario thing.

Rory O'Driscoll

And people still shot at his house.

Jason Lemkin

Yeah.

Rory O'Driscoll

It's not funny.

Jason Lemkin

It's not funny. Eric Schmidt got off light. I mean, it's not funny.

Rory O'Driscoll

It's very telling because 3 years ago—maybe 4 years ago—I was at my son's graduation, the year after ChatGPT. It was the exact opposite. One of the speakers made a semi-nice reference to ChatGPT, and all the kids clapped in a totally knowing fashion that basically exuded, "We've all cheated for the last year using this product. We love it."

It was a really sweet moment, including my son. I'm like, "Oh, I get what just happened here," right? We've gone in 3 years from graduations clapping about OpenAI because it was, "Oh my God, that got me my final essay done in 24 hours when I didn't do it," to now booing Eric Schmidt.

You might want to think about the trend here and the direction of travel if you're representing AI. That's why there's been a message shift that Dario hasn't got, but most people are now trying to emphasize the positive.

It's going to be hard to do that because, as we speak today, Meta is laying off 8,000 people. That's 8,000 lives impacted because he wants to put it all into capex. I think the politics are going only one way.

Jason Calacanis

What did the CEO of Standard Chartered Bank say? "We're getting rid of 8,000 jobs, but we don't have job losses. We just have job reductions in favor of the machines."

This is the greatest graduation speech of all: no job losses at Standard Chartered—7,800 reductions. We just have job reductions in favor of the machines. This is a level of honesty that I think is as disingenuous as it gets.

He's not even seeing them as job losses because they're no longer necessary. They're just in favor of the machines. This statement should echo through history.

I don't want to end on a negative, but Cisco cuts 4,000, LinkedIn cuts 875, and Meta cuts 8,000. To add to it, I think—

Rory O'Driscoll

I will give LinkedIn credit. They specifically said it's not caused by AI; it's realignment. But yes, the trends are tough here.

Jason Calacanis

Intel's a big one, too. Old-school Intel: 16,000.

Wow. Yeah. I mean, the politics here are going to be interesting.

Jason Calacanis

That's why I honestly think we could talk forever. I don't mean that repetitively; I actually think we will need to create policies in tech to rehire these people. I think we need to reflate.

First, we're going to get fit. We're going to realize reskilling doesn't work. We're out of time. We've got to be better than Figma. We can't screw around anymore.

We're going to get fit. We're going to replace our workflows. We're going to have AIs. Then we're going to have a social obligation. The Eric Schmidts can't just go to a graduation and say, "Fuck you." We're going to have to reflate and hire thousands and thousands of people per tech leader to avoid social unrest. We're going to have to do it.

I've had this conversation with a number of high-flying AI companies. At first they think I'm ridiculous, and then they think about it, and then they're like, "Well, maybe we need to have a 2021 social charter where we just double our headcount and have nothing to do but play on ChatGPT all day."

Rory O'Driscoll

No, I just want to flag that that's only the case if, in fact, AI is capable of replacing these jobs. There is a scenario I want to put out there in which people are overestimating what AI can do, and maybe it's not 20% of R&D headcount; it's 5%. Therefore, the amount of efficiency that AI creates might be less.

In which case, you have to ask yourself: Are these people being laid off because they were surplus to requirements all along? Are they being laid off again because you just spent all your money on capex? Or are you shocked—have you cut too deeply and have to claw some of it back?

Jason, let me be clear: If the math—if the reality is as you articulated—then you're correct. If the tech industry really does put, let's say, 20% to 50% of white-collar jobs out in the next 5 years, then you're going to have to do something massive on the social front. Otherwise, they'll be forming the guillotine in the square in San Francisco, and I have a long list of people I'd suggest putting in the tumbrils.

I actually don't think that'll happen, and I think we're exaggerating the impact of it. But you are right.

Jason Calacanis

What you cannot do is what's happening right now: laying off a whole bunch of people, saying it's AI, and then acting surprised when it bites you politically in the ass.

I mean, a genuine question. Going back to something you said, how do you think those 8,000 ex-Facebook employees are going to vote on the wealth tax next week?

Rory O'Driscoll

They're going to vote. You know, it's worse because, first of all, let me say this: We can take 5% of Zuckerberg's money and he might leave the state. I'm in.

Yeah, I agree. That's why I think we have to have this reinflation of hiring.

Jason Calacanis

First of all, one thing I know: it's N = 1.

Rory O'Driscoll

You do that, he'll just leave. He may have left. He may be a citizen of Nevada.

Jason Calacanis

You're right. My point is, politics, when you're calm and rational and you can talk, I think it's a horrible idea. You can talk about it very rationally like this because, if they leave, you lose all the tax, right?

Politics, when you've been laid off by email at 4:00 this morning because the CEO of your company has decided he'd prefer to buy, to Jason's point, $100 million of machines than $100 million of people—that politics becomes very different, and I don't think you think as much. I think you are pretty pissed off. That's my point.

I actually think it's worse than that, Rory. I think these are going to be, by far, the layoffs that Harry just rattled off, and the ones for this year, I believe, are going to be far worse than any layoffs in our lifetimes. And I'll tell you why. I know this is brutal, okay? No one's going to hire these people. No one wants them.

It has always been a scarlet letter to be laid off from a tech company, but it is a double scarlet letter today. These people are going to be angrier.

No, I'm interrupting. I'm interrupting because—late-breaking news—I have to leave in 1 minute.

Rory O'Driscoll

But I just saw a headline come in that says OpenAI might file as soon as Friday. What this says to me is they've figured out that the money—the last trains are leaving for Money Station. I don't know if it's true or not. I'm literally responding in real time here, but I think when you look at the S-1 market, you say to yourself, “Go, go, go.”

Jason Calacanis

Right.

Rory O'Driscoll

Yeah. So when Sarah said, “We need another 12 months to start the process,” what we really meant was, “We're going today.”

Jason Calacanis

If it does happen tomorrow, we might have to do another supplemental podcast like Cursor. We'll find out.

Rory O'Driscoll

No, it'll only be a filing. It'll be a confidential filing. You'll learn nothing. All you'll learn is what we've just learned—

Jason Calacanis

—which is that they're at the point of starting the process. It's the S-1 period for SpaceX, not the IPO. What really counts is the IPO, and the IPO is happening first, so we'll see SpaceX tomorrow with no OpenAI filing. Two months from now, they'll do their IPO. But there you go.

What an ending, Rory.

Rory O'Driscoll

There you go, man. My God.

Okay, I’ve got to go to just a little board meeting and try and make a buck. Goodbye.