Anthropic Inference Costs Skyrocket |TikTok Deal Closes |The IPO Market:Wealthfront & EquipmentShare

1. Brex Acquisition by Capital One for $5.15BN

Boys, it has been a big week of news. I was super happy when this news came out because I got tagged in so many messages saying, “Can you do an emergency pod?” I thought, “Well, that’s a great sign of product-market fit for what we do.” So, we’re going to start with Brex’s acquisition by Capital One: $5.15 billion, 50% cash and 50% shares.

Diving right into it, how did we analyze the announcement, which came as quite a surprise to many of us?

I thought it was a great outcome for the company. By the way, everyone who’s on this pod has probably heard the whole thread. First, you have the people saying it’s a great outcome. Then you have the people sneering, saying, “Oh, that’s a disappointment from where they were.” Then you have the counterparts saying, “Grow up, kids. Anyone who builds something from nothing to $5 billion—it’s a great outcome. Shut up.” I think that’s the right response.

Let’s assume everyone’s already caught up on that so we can engage from there. Going back to the first point, I think it was a great outcome, and I think you built something from nothing to a $5 billion outcome before your 30th birthday. That’s a heroic result and should absolutely be praised. I think it’s a smart acquisition for Capital One, too, by the way, and we can come back to that later.

On the second point, I think there are 3 different sets of comments. First, is it a great outcome in the abstract? Of course it is. The second big-picture question is whether it’s a great outcome relative to the $12 billion raise in 2021. My co-podcaster Jason did an awesome piece on that, and we should talk about it next: hubristic financings.

The last thing we’ll circle back to is whether it’s a great outcome relative to Ramp and the competitive dynamics. I’d like to throw it to Jason and say, “I thought your post on hubristic financing, and the pros and cons of raising at $12 billion and selling at $5 billion, was really good.” So, over to you for your thoughts.

Guest

Look, I don’t know everybody as well as Harry does, but as someone who was a smaller investor in the company, I asked what she thought of the outcome. Her response was, “Given where we are in the world in 2026, it’s a good outcome.” That qualified answer was interesting, and that was my thought.

Some folks are taking potshots on X. That’s the way it is. But why do we have these weird feelings? Why aren’t we sure this is a good exit at $5.1 billion in 8 years? I would have loved to have led the seed round. Maybe it’s not good enough for Harry, but Rory and I would have been happy to have led the round.

So, why do we have this feeling? This is nothing new. This is the era I call hubristic financing. You’ve got to keep doing these Harveys and Loras and OpenEvidence going from $1 billion to $12 billion, to keep up. If you don’t, you put yourself at a competitive disadvantage.

But then you set yourself up for disappointment because these companies that are fundraising to the nth degree—the Thinking Machines today and the Brexes back then—are promising, as Rory says, 100% growth ad infinitum, not just a couple of years of sustained growth. Brex was basically promising it would own all of business finance at some point, and that was the bet at $12 billion.

So, it leaves us with a weird feeling when you put the latest-stage investors aside. On paper, all the stakeholders have a great outcome. Even the liquidation preference—it’s not like that ruined the deal, right? They raised a billion-something in equity and debt. This wasn’t one of the ugly deals where $7 billion went into the company. This was a great outcome for everyone on paper, except for the late-stage investors.

But versus the promise—the commitment made to everybody, customers and late employees, for 4 years—you have to make these promises to win today. Do you have to make them to win today?

When you take money at a high price, you run the risk of subsequently exiting at a lower price and having this weird, odd feeling for a day. But I think the broader point is to step away from the weird feeling. Why did you raise money at $12 billion, going back to 2021? Who’s going to say, “Investors are offering you money at $12 billion? No, I’d prefer to take it at $6 billion. Thank you very much”? I’m an idiot, right?

No, you raised the money because you needed the capital. If they hadn’t raised the money, they would have run out of money. That would have been dumb. Once you need the money, you’re going to raise it at a market price.

I thought what you said was that there are pros and cons to that. You get the buzz, you get the momentum, you get the employees, and you get to compete in a world where other people are doing it. In return, you pay this weird 1-day tax on the day you exit, which is that you’ve just had a heroic, world-changing, life-changing event, and then you just feel weird.

But given what’s on the field, there’s no way to avoid that because you had to raise the money in 2021. Therefore, you raised it at the market price at the time. The bad feelings last for a day, and the $5 billion lasts forever. So, you’ll get over it.

Guest

No, we’re actually going to forget about Brex in 24 months because I’m going to get a Capital One card in the mail as soon as this deal closes. We’re going to even forget whether it was 2 X’s or how to spell it. This is our world. We’ll forget, right?

I will say just one thing on the topic. It’s a great outcome for the founders on many levels, a great outcome for early employees—not what they thought they’d make in 2021, but still a great outcome—and a great outcome for Ribbit.

The one thing, in addition to my post, is that I saw something not too long ago, but a while back, from Ali at Databricks, one of the most successful companies to use this strategy. He said, “I never wanted to raise more than 2 years ahead of the valuation I was confident I could hit.”

I never wanted to raise not just 2 years ahead—the classic VC thinking is, “1 year ahead for a hot company, 2 years ahead for a great company, maybe 2.8 years ahead for Thinking Machines.” I don’t know. But that’s a thoughtful response. If you believe it, did Brex believe they’d be worth $100 billion? Probably. I don’t know.

Things were loopy in 2021, and maybe they’re loopy in 2026, but—

They probably did because, look, it’s a sin of extrapolation. The growth rate was probably 200% to 300%. You extrapolate. I mean, it’s the age-old truth.

Most of these so-called insane valuations can actually be explained by the Ali comment: “Provided I keep the growth rate up for 2 years, then I will have grown into this valuation on a revenue-multiple basis.” It’s a perfectly innocuous sentence. But buried in it is a whole debt trap, because the minute 2022 came along, the growth rate faded.

Then your capital gets more scarce, so you have to try and converge on profitability. The growth rate goes down even more. What was a totally legitimate belief in 2021—that 3 more years of 300% growth would make me worth $12 billion—becomes utterly insane in 2024. It’s just the cost of doing business in this crazy game that everyone plays, right?

What I’m trying to do is avoid this moral judgment of, “It was good,” or, “It was bad,” and all these emotion-laden reactions. If you play the game of, for example, paying up massively for hypergrowth, and it doesn’t work, you get a 1 X. As long as maybe 30% to 40% of them are 1 X, you don’t have any major whoopsies where you actually take a loss, and you get 3, 4, 5 or 6 X on your good ones, then overall the math works.

Even in a mediocre year like 2021, you end up with a sub-2 X but still a perfectly fine fund. In other words, it’s just the nature of the business. Things prove up in the end for what they really are, not what you delusionally think they are at one point in time.

In the end, a financial services company was always going to trade at a financial services multiple adjusted for growth, and that’s what happened here. Growth came down to still very impressive but normalized levels. Capital One leaned in and said, “$700 million in revenue, give or take, at 7 times revenue—that’s a good deal for me. Done.”

2. Does Brex's Acquisition Help or Hurt Ramp?

How does this change the game for Ramp? Ramp obviously raised at $32 billion.

Guest

I think this is a great question because, remember, I said there are 3 things: is it, standalone, a great outcome? And then, of course, is it a great outcome for the investors versus 2021?

Rory O’Driscoll

You're raising the third one, which to me is the interesting one, because Ramp—you know, and some of—actually, I've got to say, the CEO of Ramp did a fairly classy post. It was like, “Hey, congratulations.” He had sold a company to Capital One. It wasn't awful, right? The Founders Fund guy did a dance on your grave.

Dude, I love Eric, and I do. Yeah, I like the team. It was a dig. I know some good Londoners.

Rory O’Driscoll

It was a dig, but perhaps, as is so often the case in America today, I'm judging quality by the opposition. The Founders Fund did a straight dance on your grave.

If I'm Ramp, from an operational perspective, the two things are true here, guys. From an operational perspective, this is further validation that, quote-unquote, “I've won.” I started later and I'm doing $1 billion; they started earlier, and they're doing $700 million. Good news: you've won, right?

But the bad news that you just can't ignore with a tweet is that when real money decided not to buy 2% of this thing in a secondary sale, but to actually write a check for the asset, they said, “We're going to multiply by 7.” And if you multiply Ramp's $1 billion run rate by 7, you get $7 billion. Now, they're growing faster than Ramp—maybe double it, $15 billion.

What it points to is that we live in this crazy land of VC valuations where they're made once a year, when only 1 person buys and no one can substantially sell. They're very thin markets, and we hope they're roughly right. Sometimes we're surprised to the upside when they go public, and then sometimes we're surprised to the downside.

If you're doing your mark-to-market on Ramp right now, how do you factor in a recent transaction at 7 times into your multiple of 30 times? It does at least—I mean, I'm not saying it's dispositive, because you're growing faster and you did, as you put it, win—but it does make you think: maybe when you go public in 2 years and you want to monetize a, let's say, at that stage, $2 billion revenue company, maybe you're still growing a little faster, maybe you get 10 times. I don't know.

But if I was the investor in Ramp, or if I was the investor who just wrote the check at a $32 billion valuation, I'd at least pause and go, “Hmm, let me check my assumptions 1 last time here,” right? Maybe it can still work, but I've got to be a big company. I mean, you ain't going to get the M&A outcome anymore.

You've just got to be the big company and trade in the public markets at a significantly higher multiple than the other financial services companies. The only way you can do it is if you keep the growth up—if you keep the growth up. So it's not like it's impossible, but it's just a significant data point that weighs the other way as you think about value.

If I'm the Brex founders, right? If I'm Pedro—especially Ramp—we talk about Ramp, but we also look at nonpublic companies, and I have another comp, going to Rory's point. Let's say the 3 of us were the founders. I'd be like, “Jesus, we have a comp that's basically worth the same as us. Basically the same revenue.”

Now, it's got debt and other issues, but I'm like, “My God, guys, we could work for 3 more years to an IPO, suffer lots of dilution in the IPO, stress, and basically economically be the same place in 3 years.” Now, that's cool if the 3 of us want to build something much bigger than this, right? But I don't even know if $10 billion is worth it if we don't want to do it—if we don't want to build this as a generational company on our own for a decade—because the non-comp's a tough one.

It sort of says to me, we're going to grind it out for 2 to 3 years and be worth the same. So we better want it. And all my public-company CEOs—they're pretty grouchy today. I always say 80% of the public-company B2B CEOs are now—they may be thrilled when the next generation IPOs, but this isn't the happiest cast of characters, is it?

Rory O’Driscoll

The public-company CEOs—they're not happy today, but it's a super-good outcome. I think Capital One has played a very shrewd hand here, because remember, all these businesses—Ramp, Divvy, which my former company I was formerly involved with, Bill.com—they all monetize on interchange, right?

Most interchange is Visa and Mastercard, which is a third-party network. You have Visa, the issuer bank, and the accepting bank. Capital One bought Discover Card. Discover has a closed network where they get all the money in the interchange, right? So that's a really powerful asset for them.

Now that they have Brex, they'll probably be directing as much of that money flow onto their own rails, as the bankers call it, as they can. And what that means is they'll be able to extract a lot more of the value from it. So this could be an example where the asset is worth more to Capital One than it was on a standalone basis. I think it's a very shrewd acquisition for Capital One.

In the last 5 or 6 months—we forget—in the last 5 or 6 months, they bought Discover less than 6 months ago. I think it closed just recently, for $35 billion at announcement, $50 billion at close. And now they bought Brex, which they can fold onto Discover.

So they're making a real push into this space, which is another thing you think about as an independent: you're sitting there going, “Hmm, I'm going to be playing against the A team now with a structural cost advantage.”

You had 3 pieces of news today, investors in Ramp. You clearly won, and the other guy said you won. That's good news. Put that in the positive column. Second piece of information: people think a company going a little bit slower than you and in roughly the same business is worth 7 times, and you think yours is worth 30 times. Hmm. Put that in the negative.

And then, lastly, a well-funded public-company competitor is directly entering your space with a structural cost advantage. Now, you can total all of that and decide. My impulse at the end of that is to tweet and say, “Well done.” But deep in your soul, you kind of go, “Hmm, not sure that was the best day out there for my stock.”

At least stay private longer.

Rory O’Driscoll

You're going to have to at least believe in the dream while private. Let's do another secondary, boys. Lads.

3. TikTok Deal Completed: Who Won & Who Lost: Analysis

Okay, moving on, boys. Another very, very significant bit of news. We've all been waiting for a long period. Is the TikTok deal finally done? U.S. investors will own 80% of the company. The algorithm remains controlled by the Chinese owners, which is interesting. How do we analyze this deal getting done now, and how do we think about it?

Rory O’Driscoll

I think you can't—I mean, you can't approach this deal economically. In the main, it's a political/geopolitical decision to force TikTok to divest, and when you have that and then have a very directed purchaser program, let's do the economics first of all.

It looks like a very attractive deal. I think the company is doing $15 billion or $16 billion in U.S. revenue, and they bought it for $14 billion, like 1 times revenue, plus or minus. That's a wildly cheap deal compared to anything else.

Now, there's a term that says some portion of the opex is a license fee or some payment back to the Chinese parent. So you don't know the full economics, but my sense is that it's a wildly accretive deal for the lucky, chosen investors in the new oligopolistic capitalism that we now practice.

It is worth pointing out that, fundamentally, from an economics perspective, ignoring any questions—ignoring any other questions—I wish I had some of that in my 401(k), you know? The most addictive, popular application in the United States social-media marketplace at 1 times revenue. Put me down for some.

At first when I saw this deal, I thought—and Rory can play historian here—it hearkened back to me to when Andreessen Horowitz got off the ground doing Skype. Yes. And the reason it was a structurally weird deal was that they got a good deal. They took risks. Skype was an aging platform, but they bought it from eBay, right? Is that because eBay needed to divest it? They'd had enough of the deal.

Rory O’Driscoll

There was no synergy, and Andreessen went all in. They didn't have that money, but they went all in and 3x'd their money in 12 or 14 months.

This felt like another moment in time where you could get a great deal. The only thing I don't get is, why didn't those guys show up? Why aren't Andreessen, Sequoia, and Lightspeed in this deal?

And you have weird ones: Oracle, which also has infrastructure; you have UAE sovereign wealth funds. Where are Sequoia and Andreessen? That's the only thing that made me pause, like maybe it's not such a great deal, because those guys are just in the business of minting money now. Why didn't they each at least throw in $1 billion or $2 billion into this deal? They put it into everything else.

I don't know. At one point they were in it, and then they were not.

Rory O’Driscoll

There's a reason they're not in that deal. It's free money otherwise, right? We're missing something.

And they're not averse to structurally challenging deals like X, with Elon taking over, which they all engaged in.

Rory O’Driscoll

No, totally. And I think you're exactly right. I remember them doing the Skype deal. It was a very shrewd deal, and I want to say that the PE firm brought them in because they had venture expertise—which is, again, don't quote me—I think this PE firm was in fact Silver Lake again, but I'm going from memory there.

4. Anthropic Inference Costs Higher Than Expected

But I remember, as you say, Andreessen came in on the deal. It was spun out from eBay; there were some issues around licenses and IP, and it was a little bit risky. They cleaned it up for 12 to 24 months and then sold it to Microsoft and made 3 times the money on a ton of money in the first fund. So, super shrewd, that was. And if someone does the same thing here, it'll be interesting.

Let’s move on. Let’s discuss Anthropic. Anthropic’s inference costs are 23% higher than expected. Are there economies of scale in AI after all? How should we read this?

Rory O’Driscoll

You know, look, there’s a lot here, but I think this is so important for B2B companies. I was literally at a board meeting of a B2B company with a powerful AI agent costing $100 million, and I was seeing some of the dumb points in this board meeting: “Hey, guys, in 2026, we’ve really got to drive down inference costs now.” I’m like, do you realize you have 6 mega-funded competitors? The only differentiation is who has the best agent now. You’re going to cut back on your inference? It doesn’t make sense, right?

This is the point Amjad was making so many times. I’m sure Anton from Lovable has made his own version, but Amjad’s always been like, “No, it’s going to—everything’s going to get more expensive because as soon as we figure out how to do this stuff, we’re going to burn even more tokens. We will actually burn an infinite amount of tokens if we can.” It even happened to Anthropic, right? It happened to everybody.

I think a lot of folks, especially folks that aren’t quite growing at the OpenEvidence levels today, or at Ramp, are thinking, “God, what am I going to do with these inference costs?” And I’ve got to tell you, the idea that you can use cheap models and cut back on your inference and still be competitive—that’s the thing. Still be competitive with the hot Andreessen-funded company? There’s no chance. You can’t be competitive without that inference.

Guest

But I do think it’s important not to lose sight of the fact that, even though you asked the question, “Oh, my God, Anthropic’s inference costs were higher than expected. Is there any leverage with your theoretical economies of scale, Harry?”—the truth is, remember, last year they had a negative 94% gross margin, and this year they have a 40%, plus or minus, gross margin. Now, it’s not 50%, so clearly the gross margins are improving substantially.

I think the real, middle-of-the-road, boring comment is that there is significant leverage in inference costs, and the P&L is getting a lot better. But it may not go all the way to the—you know, it may take longer. We thought we’d be at 50%; now you’re at 40%. It may take 2 years to get to 70%, or you may never get there. You may top out at 60%, right?

I think it’s just the nature of the beast in AI. You’re dealing with this totally new business product and a totally new market. You don’t have a history; you have a hypothetical. However, I don’t doubt the fact that a profitable business—and I define that as free cash flow or operating income—will emerge from something like Anthropic.

I mean, Anthropic is not going to not have a profitable business model because this clearly is converging. It’s just a question of at what scale it converges and what operating model it converges at. Is it a 10% operating-margin business or a 30%? Right? So, it’s getting better. That’s—but it’s getting there. It’s getting there a little more slowly than you might like, but it’s still massive. I mean, from negative 94% margins last year to positive 40% this year, that’s a big move.

Can I ask Jason specifically? You said at the end of the year, when we did our quiz show on the roundup, that 2026 would be the year where we would see inference running for 24 hours a day for a small portion of the knowledge economy. I thought that was a really interesting takeaway. When you think about that—

Guest

I actually tried to build it over the weekend.

When you think about that combined with these increased inference costs being higher than expected, how do you think about those 2 together?

Guest

It’s easy to say we’re going to use 3 orders of magnitude more inference in 24 months. It’s easy to say it. It’s potentially more than an order of magnitude by the end of the year. The cost decline that we’re also seeing, despite the Anthropic thing—it’s hard. Token consumption is increasing, and there’s deflation in the per-token cost.

We haven’t seen—we’ve seen that improvement, I think, at Anthropic, to Rory’s point, and I might be getting it wrong—but we haven’t seen this break point where we’re getting a break. Maybe the margins are getting better, but we’re going to keep burning more. So, I’m waiting for the moment, but I’m not seeing it. I just see it accelerating.

We’ve talked about memory, but ChatGPT and Claude don’t have that much memory if you try them. They don’t remember much, do they? They remember this little bit. So, I built a version of Claude over the weekend called REN. You can try it at ren.chat.ai. It remembers everything. It compacts everything, so it never forgets anything.

I learned a lot of things, but one of them is that it’s going to burn a lot of tokens if you want to save every chat you’ve ever had, every discussion over all time, and reference it for years. What if that runs 24/7? Maybe I’m rambling and not answering your question, but I don’t see it—I just see it accelerating.

Again, maybe I’m not sure. My biggest concern for founders out there, especially for folks who are not quite in the top 0.01%, is that they’re mis-modeling this—especially folks in ops on their team or folks who aren’t close to AI. You need to model in that your inference costs are going up this year, not down.

There’s no way—if you walked into your board meeting and said, “Hey, good news, guys. Inference costs are going down 30% this year because our IT team’s really good at managing costs,” I would throw my mouse at the monitor.

Rory O’Driscoll

You are right. But, at the risk of zooming out, I have 2 zoomed-out comments on that. First of all, when you look at the problems you’d like to be wrestling with as a business—an individual business or an industry—the problem of, “I have infinite demand for this digital good, which is still quite expensive to produce, so we’re going to have to figure out how much to charge for it and how to ration it,” is a wonderful problem compared to, “No one wants to buy this digital good. I don’t know what to do,” right?

The big thing—you’re right. What you’re saying, Jason, is correct. The demand for inference, the demand for tokens, can be almost unlimited in some cases because the more you can deliver, the more you can do. So, metering that demand relative to the cost to produce is, as it were, the business challenge, and you’re seeing that across the board.

In Anthropic, you’re seeing all these plans. They have their $200 plan, the $20 plan, and then they have those few people who are doing $1,000 of tokens on their $200 plan. What do you do about that?

But again, that’s what’s happening out here: everyone’s trying. And that’s if you’re the model producers. If you’re a SaaS vendor, like an AI apps vendor that you and I are investing in, and inference is one of your biggest costs, that changes things.

It used to be, 5 years ago, that AWS would be one of your biggest costs. It would hit 10%, and everyone would lose their minds in the board meeting. You’d say, “Let’s get it down,” and you’d manage the process and get it down to 8% or 9% with efficiencies. Now, you’re right: inference is the big cost. If you’re a high-priced app, maybe it’s 10% or 15%. If you’re a coding-type app, maybe it’s 50%, 60%, 70% of your revenue. If you don’t manage that correctly, you don’t have a business.

But I do want to get your thoughts on what I worry about. Maybe at a practical level, I worry there’s this middle category. These are mature companies, not hyper-mature: $50 million, $100 million, $200 million ARR B2B companies, and they finally got a decent agent built.

It’s taken them a while. They have 10,000 happy customers, and they’re pushing this agent out. They got to break-even last year because Scale and 20VC aren’t going to put any more money in, even though they’re supportive, because the growth’s not there, right? They’ve got $30 million left in the bank. They’re break-even at $40 million or $50 million ARR.

Now I’m competing with Decagon or Lovable or whoever, and I need $20 million more of inference this year or it’s game over because I can’t compete. Their agent is better. The way I’ve deployed mine is great, but it’s $2 per interaction, Rory. It’s $2.50 per interaction. I need 50 million interactions this year. That means I need $100 million, and now I’m bringing in $50 million.

What do I do, Rory? I can’t be competitive. You told me to. You told me to get to break-even. I did that. Thank you, guys. Then you told me I had to be more AI. Thank you, guys. Now I built it. How am I going to fund the $50 million in inference, right?

Rory O’Driscoll

OpenEvidence has the money. You have the irritating habit of asking exactly the right question.

No, I’m mentally thinking of some companies that have gone through that. You’re exactly right. It’s like, “Hey, your SaaS product isn’t enough. Let’s get profitable.” Okay, you got profitable, but nobody cares. You need an AI product. Oh, my God, you’ve delivered an AI product. Your customers love it. And now the next shoe to drop is: how are you going to finance this thing? Because you’re up against people who can raise $200 million on a dream.

I acknowledge that. Rather than telling you I have the answer, it’s an issue I’m wrestling with. I can think of 2 boardrooms in the next 2 months—maybe 1 month, as we do annual planning, right? How aggressive can you be in this market? Because if you play defense—if you try to meter it to your cash constraints—you’re going to get left behind.

So, you’d better be aggressive. The gut-level test—and this is how capitalism, I suppose, is meant to work—comes down to this: if your customers are getting value from your AI agent that they can’t get anywhere else, and you can make that value clear, then you can charge enough to pay for your tokens. Yay, you.

If you’re not giving value, or if you’re locked in a war with someone else who has infinite capital and is willing to give it away longer than you, then you’re probably going to lose, and you should figure out how to exit now.

And to some extent, the midyear coding wars, where, oh my gosh, likely Windsurf looked to exit—there was a little bit of that dynamic going on. Is this war escalating with a level of token intensity that you just can't keep up with? I acknowledge that at the app level, a couple of my companies are wrestling with those issues right now.

Guest

Yeah. I'm honestly worried. We talk about whether SaaS is dead or what's going on. I worry this is the next final act: you did all the right things, right?

I did everything you told me to do.

Guest

You did it. You're growing at 0%. Your customers don't hate you. You built an agent, and the final nail in the coffin is: we just can't afford the inference. We just can't build a competitive product. We can't.

Even Canva, which will be one of the great IPOs—and even Cliff Obrecht teased that I could build Gamma—but I can't burn the way Gamma burns those tokens. Now, maybe he'd say something different today, but it was the same point that echoed in my head that now we're seeing across boardrooms, across B2B companies. I think it's the final nail.

Is there a way out, though? If you can't raise the money to compete, but you can't not spend?

Guest

Well, Rory O'Driscoll hit the way out. There is a simple way out, which is: you build an epically good agent. Typically, one where, let's say, your product's $5,000 a year or $10,000 a year, and you're able to charge $20,000 a month for your agent, or $10,000 a month because it replaces 20 people. It's that good. It's not pretend-that-good. It's not that good on a sales pitch. It's literally so good that the ROI is measured in weeks, right? That's your way out.

But the problem is, a lot of B2B companies are just struggling to get par. The bar is so high, and that's the answer. But it's exhausting because it was so much work just to get here, to get to profitability, to get to an agent. Now you have to beat the agents that OpenEvidence, Lora, and Harvey—whoever we're going to talk about—represent. You have to have a better agent than them to earn the $20,000 a month. Your team better be the best.

To take another example that probably applies to even a large public company like Salesforce, which, yes, has infinite money but also doesn't want you to dip into the red, you make sure that you have the advantage of the data that you possess to make it a better agent, to make it a more efficient agent. Maybe you have to do less processing. Another thing is, I've seen so many companies using the open-source models for a lot of it, so you can leverage that and get cheaper processing. Yeah, you've got to do all those things, but more than anything, I think, Harry, you're right: you've got to deliver value such that you can charge for it.

But in the end, the dirty little secret is that everyone's going to have to deliver value greater than the amount of money they are spending. You're going to have to be able to charge more than the amount of money it costs to make the thing. OpenAI may get to do that for longer than anyone else, but in the end, the wheels of capitalism do grind fine, and we're all going to have to pony up and be cash-flow positive.

But the thing is, listen, I don't have everybody's numbers, for sure. The other advantage that the new entrants have is that if you have the best agent and you have the kind of market demand we see, then, for you, your inference costs are a marketing cost.

Guest

The established players don't have that luxury. They're already spending massive amounts on traditional sales and marketing.

I know, I hear you. Versus, you know, I've had Harvey and Legora on. Harvey went to $200 million last year; OpenEvidence, $100 million a year.

Rory O’Driscoll

One of the ahas from this is just the demand for inference and, by extension, the demand for compute. What does this say about it? I always apply a certain discount factor to what people running the large AI model companies say about demand, because they're talking their book. Even the poor fools like Oracle, who are investing to chase that demand and sell them compute services, I'm like, "Maybe you're getting fooled by these other guys."

But I always think the guys running TSMC are sharp, right? They've been around a long time, and they're cynical. I'm sure you all saw that piece about 12 months ago. They were fairly skeptical when Altman was talking about, "We're going to need to raise $1 trillion." They were like, "Yeah, yeah, yeah. Go away, AI boy," right?

They just did their earnings call, and the comment was basically that demand for compute is effectively infinite right now, and they're raising their capex. Remember, these are not folks who say, "I'm going to spend $50 billion." The peak before was $40 billion; the low was $22 billion 2 years ago. They're raising their capex budget for next year, and they're basically saying, "We think the demand is real right now."

To me, that's the point, because we've all been wrestling with this. Is there going to be a day when everyone says, "We're not going to invest as much anymore. We're going to slow down just a little"? You're so far out there on the—going back to the Brex comp in '21—we're so underwriting hypergrowth that even the slightest slowdown would be pretty brutal for the markets.

And this was the biggest tell of all, because these are the guys who spend the capex with a 2- or 3-year lead cycle, that services NVIDIA, that services the hyperscalers, that services OpenAI, that services the AI companies—the very bottom, the very first step in the AI pyramid. The guys running that are saying, "We're going to need a whole bunch more compute here. We're going to need a whole bunch more capex."

It was just interesting because I think keeping an eye on TSMC, as the people who would own the problem if they overinvest, is important. You can cut employees, you can turn off your GPU, but if you dig a big, deep hole in the ground in Phoenix and a big, deep hole in the ground in Japan and put a fab in there, and no one uses it, you're out $20 billion—and they're leaning in right now.

That inference demand is pretty clearly there according to all the tells. For those who think about the AI bubble, does that not completely denigrate those risks of an AI bubble bursting? When you look at them, when you look at the improvements, when you look at Dario Amodei coming out today saying, "Hey, when you look at the improvements, we'll be replacing everyone's job in under 5 years."

Not every statement that says it's going to go on now has to be equally correct. A bunch of people who have the money, starting with the foundries, going to the chip companies and going to the hyperscalers, have all said, "We're going to spend this money this year." So I think it's highly unlikely that this is going to be the year when people get terrified and say, "I'm not going to do it."

At some point, I think they will, because I think we probably are overinvesting at some level. But right now, people are saying, "I can see logic to this thing for the next 12 to 24 months," despite the massive gap between the capex—which is now $600 billion—and the apps' revenue, squinting, is $100 billion. So you're still $500 billion a year in the hole. But right now, people are saying the return is there. That's all you can conclude right now. If you knew when it was going to happen to the day, you'd be trading NVIDIA puts, and you wouldn't be talking to me.

Sorry, I'm the least intelligent on this call, which is why I love doing it. When you look at the cost of inference maintaining its high price, and when you look at what Jason is quite rightly saying—that inference will be running 24/7 for more and more of the knowledge-worker population—why is that not just continuing evidence that NVIDIA has so much more room to run and is actually underpriced today?

Guest

Take, for example, that statement. The cost of inference—the cost per token—goes down enormously quickly, and it's just that demand expands and people use more and more tokens to get to the same dollar amount, right? Just to be precise, I think the only argument against what you're saying is some version of: as the numbers get bigger and bigger, you start encountering GDP-type limits. Your total U.S. capex is X, and you're now 30% of the total capex. Can we really stop building tractors and buildings and put all our money into great big data centers?

There are some people who articulate that vision. There are people who articulate, on the Dwarkesh podcast, the trillion-dollar data center. Well, maybe. I'm not sure that happens. I don't think you have to believe in the trillion-dollar data center or that all human beings are going to be unemployed by AI to believe that, at the margin, for the next 12 months, it looks more likely than not that people will continue on roughly the same investment trajectory.

Those are both statements that have the same conclusion for the next 12 months, but are very different in terms of their grandiosity. Do you understand me? And I'm not making the grandiose statement. I'm just saying, let me make it really tangible.

At some point, NVIDIA puts will be a great buy because every semiconductor cycle for the last 40 years has ended up in a massive downswing. I ain't buying them today. That's when the rubber hits the road: when people are done talking, do they want to say they believe? I don't have that conviction yet, because people who have money and conviction are saying they're going to spend.

We all know that some version of the bubble will pop, even if it's well after SpaceX IPOs and we have 1,000 data centers in space, which is Elon's dream. There are so many interesting things coming—24/7 inference—but it will pop someday. If we can't see it reasonably popping in the next 24 months, I don't know that, as investors, as employees, as management team members, we can have dinner conversations, but I'm not sure there's much we should change.

We all got caught around December 2021, when 90% of tech thought this was going to last longer, and then, bam, we got it. It just bit us, right? HashiCorp went public, and then it just stopped for 2 years. But this is different, and there's just no upside in betting this is going to slow in the next 24 months. There's literally, at least for 99.9% of us, no upside.

To ground it in practicalities, the only thing you can advise people is: think about a scenario plan. Think about whether you would have a plan if it were to change. Think about your funding strategy, especially if money is cheap, to make sure—going back to Brex—you’re really glad you raised that money in 2021.

All you can do is play to the current scenario, but have a plan so that, if the world changes, you'll know how to change, and then you've raised money to be able to survive that.

5. OpenEvidence Raises at $12BN from Thrive and DST

Speaking of playing the game on the field, we mentioned them a couple of times. OpenEvidence raises at $12 billion, led by Thrive and DST. It's a 12× valuation step-up from where they raised at $1 billion from likely Sequoia at the start of the year. Revenue growth has been amazing.

Pharmaceutical ad spend in the US on media is $22 billion a year. If you think about the transition of that to their business model and assume a reasonable take, you can see them being a $4 to $5 billion revenue business in that alone. That doesn't feel crazy, but then, in other aspects, it does. How did you guys read this one?

Guest 2

I think it's a great company. It's a perfect use case for AI, and it's one of the use cases where the general models are good, but the combination of specific relationships with the New England Journal of Medicine and all that, plus restricting access only to medical professionals, plus HIPPA compliance, means you've got this really nice product to allow doctors to do decision support.

You go and check online, “What's the recommended treatment for some obscure disease I haven't seen?” And then the obvious thing you do with that is you sell them ads, right? The obvious people to advertise to those doctors are the drug companies because they want to sell to the doctors, right? So it's a perfect business, and they've escalated to, I believe, $150 million in revenue.

I was actually impressed that you led with the market size, because the only thing that's clear here is that they're the winner in the space, right? Doximity is the old pre-generative-AI competitor. But in terms of doctor media mindshare for doctor-like things, Doximity helps you a lot with thinking about salary and thinking about jobs.

But I have a medical question to which I want a highly technical medical answer. They appear to have commanding market share, so you've won that business. The only question is: how big is the market?

You can say total media, total drug-company spend on quote-unquote drug advertising is $20 to $30 billion. But Harry, a good half of that is TV ads to consumers, right? For a lot of these drugs, especially for long-term conditions, the advertising is not going to the doctors; it's actually going to the individuals who are wrestling with the disease so they can build consumer preference. That halves the market, right?

On top of that, if you look at pharma companies' spend on trying to reach medical professionals, actual direct-to-doctor advertising is a $2 billion to $3 billion marketplace, which is now getting a little bit smaller, right? You then have a whole bunch of these infamous pharmaceutical reps. A lot of this marketing is done in person.

You have folks just calling on doctors, bringing donuts, saying, “Hey, here's a sample pack of my nice new arthritic drug. Give it to your consumers.” So, for OpenEvidence to get to that valuation, what they have to do is one of 2 things.

Either, A, they have to blow open some of that budget away from pharma reps calling on doctors and move more of that budget online, which, by the way, is a totally credible thing to do, right? But that's what they have to do. Or they have to expand into other services to doctors.

I think Doximity added some SKUs, for example. A product they added that was really clever was a scheduling app with a kind of phone number that doctors could use that wasn't their personal cell, because doctors want to give out their cell so people can reach them, but they don't want to give out their personal cell. So, some nice little doctor products.

To get 3× from $12 billion, you probably have to do some significant TAM expansion. It's credible that they do it, but they've got to do it.

You've got to have $5 billion in revenue, don't you, at a 7× multiple?

Guest 2

To get to the $35 billion. Yeah, that's where it's open.

I'm not a total expert, but based on what I do know about OpenEvidence, if the deal was priced right, anyone would want to do it. It's got the market share. It's very valued by physicians. It's one of the largest.

They haven't figured out the true TAM, but the notional TAM is about as big as it gets. Of course, you'd want to do this deal at the right price, right? Anyone would do it.

If you were a growth investor, would you do it at $12 billion, Jason?

Guest 2

Well, that's the question. This is back to hubristic fundraising in the Brex round. Who, at OpenEvidence, ElevenLabs, Lovable, and Harvey, is going to do the Brex round at $12 billion? Who's doing that round? Is this that round, or is it the round in March at $30 billion? Because this is hubristic fundraising.

OpenEvidence will probably do a round at $30 billion or $40 billion next year. I'm actually going to suggest that Thrive is very smart, and they've probably done the math. This is the right insertion point for them, they believe in it, and someone else is going to do it at $30 billion to $40 billion next year as it goes to $40 billion next year or $50 billion. Someone's going to do that.

So who does the $12 billion Brex round here, where there's nothing but greatness but someone gets caught with the tail end of hubristic fundraising?

The first line really resonates with me. This is such an obviously good deal in such an obviously good market, with a wildly high-quality founder who's had a win before. He sold likely Kensho to S&P. Big-brain, PhD, AI guy, AI-native from his first deal, which was a financial AI company. This is an impeccable background here, great connectors—there's nothing not to like here.

And so you're right. Let me give you a clue: you're not going to find the discount here, people. But you don't think this will be the $12 billion price round that Brex was, where the music stopped and it was that last high-price round?

Guest 2

It's always a tricky question, because if you played back, remember, they had a round at $3 billion and I think a round at $6 billion. So this is the 4th time in, and every one of those rounds you'd have said, “Maybe this is the one that's going too far,” right?

But when you step back, they 10×ed revenue this year when they 10×ed their valuation, plus or minus, and so the revenue multiple is the same. I think that's the market we're in now, and at some point someone's going to be left with it.

You're right, Jason. It's the Brex risk: the tide goes out, it's still an amazing company, but maybe you're doomed to a 1×. Is this the round that happens?

Guest 2

I might have said the $6 billion round was just, given the core of the TAM and market size. Yeah, that's the bet, right?

What's so hard is, in the moment, it never feels that hubristic. I remember with Brex, I had Henrique and Pedro on the show back in 2020 or 2021, and they were talking about Amex and the fragility and how they could build a $100 billion business, and $12 billion did not seem that crazy, does it?

Guest 2

Hubris is like that.

And even more, these latest-stage deals of great companies are very easy to talk yourself into when times are good.

Guest 2

When times are tough, they're still hard to talk yourself into, but times are mixed today. The good stuff is so good, it's so easy to walk into the partners' meeting and advocate for OpenEvidence, isn't it? It's just so easy.

6. a16z Companies are 2/3 AI Revenues

“Guys, it's a yes. It's a little expensive at 50 times revenue or whatever it is, but you can't argue that this is a generational company.” And Mark Andre says, “We do generational companies at any price.”

We just buy as much as we can. They only go up overall. Not all of them, but they only go up overall. This is a generational company. I know it was $12 billion last week, but I proposed $1 billion at $35 billion, guys, this week. It’s a generational company.

You said Mark Andre’s firm released a report this week, which I thought was astounding for a couple of different reasons. Most importantly, they put out $8 billion invested in 2025. This is a16z’s report, by the way. To give some context, a16z did a report with incredible slides.

I thought a16z and Air Street did great reports this week.

Guest 2

But in a16z’s report, they said about $8 billion was invested in 2025. The stat that blew me away was that 2/3 of private AI revenue is generated by Andreessen-backed companies. OpenAI, Databricks, Cursor, Harvey, Replit—the list continues.

I was astounded by that. I don’t know if you have takeaways from it, but I thought it was interesting for the audience to hear.

Rory O’Driscoll

I thought it was an excellent report, and I thought there was a lot of substantive, good economic analysis up and down the report. I thought that slide was probably the least astounding one when you think about it for longer than clearly you did, because, yeah, it was a great sound bite—and those guys are the best marketers. It’s a great sound bite, right?

But objectively speaking, if you add up all the AI revenue, you’re going to get $13 billion for OpenAI, $4 billion for Anthropic, and everything else is in the noise, right? $200 million for Harvey—whoop-dee-doo. They’re amazing companies. They’re going to be great, but my point is, to a rounding error, the sum of the revenue is the sum of OpenAI and Anthropic.

And then, actually, if you were to lump a third one in, it would be Databricks, which has a massive market share. So, to a rounding error, another way of saying the same slide is: a16z has money in OpenAI, and OpenAI is 40–50% of total revenue.

I’ll tell you what I found interesting about it: this, and then Gary Tan again saying that venture should be 10 times bigger. Smart guys, right? Is this really an asset class?

The classic take in venture is that it’s not really an asset class; it’s a weird niche of private equity. Yes, the top quartile—certainly the top decile—performs, but the rest is a disaster. So it’s not an asset class if the bottom 75% isn’t even worth getting out of bed for.

If a16z has proven that this penetration and AUM are repeatable, right, like clockwork, and Y Combinator is doing it at the low end, is venture finally an asset class? If it is, that’s Garry’s point: put 10 times as much money in. We have access to the early-stage funnels, right? And a16z is saying, “We have 2/3 of private AI revenue.”

There’s an asterisk and a dagger to Rory’s point, right? Because it’s weighted on 2 names. But still, the point is that if it is an asset class, then you can deploy the maximum amount of practical capital into it efficiently.

Rory O’Driscoll

I could agree with your conclusion on it being, at some level, an asset class. I might even argue 2 asset classes. I’m not sure that I agree with your conclusion that, therefore, you deploy more.

Well, Garry said that, not me.

Rory O’Driscoll

Got it. At the Garry level, I agree, to be clear. Now we’re going to jump around a lot, but let’s digress onto Y Combinator. The slogan of Y Combinator from day 1 is to make it easy for startups to start. It’s some more elegant version of that, right?

At the margin, there’s no meaningful capital cost to giving someone $250,000 or $500,000 to have a go. So the more people who start and try new companies, at the margin, it’s a great thing for everyone, including the people involved. Worst case, you do it for 2 years, you fold up, and you go back to college.

That’s exactly right. You’re golden. You’re fine.

Rory O’Driscoll

So, as far as Y Combinator is concerned, and encouraging startups, the more the merrier. I think in terms of where I disagree with you, I think venture is actually 2 asset classes.

It’s the traditional early-stage venture that’s existed for 20 or 30 years, and this new late- and later-stage venture asset class that used to be called small-cap growth and is now privately held. So there are 2 asset classes.

I don’t think in either case they benefit from excess capital, because I do believe that Martin Biggs has said it before: there’s no investing business so good that excess capital won’t ruin it, right? I do think that excess capital will make this business harder and, to some extent, erode the returns.

You’re seeing that. It’s funny: they said that 2021 was a very active year. I think 2025 was the most active year since 2021. It was a reasonably active year in ’21, not nearly as different. We do roughly the same number of deals every year.

In retrospect, I wish I’d just gone home, right? If you think everything in 2021 was either priced wrong and makes a 1x, or was early and just totally wrong and makes less than 1x, let’s just say, right? Other than a few companies that were the early precursors of AI, excess activity is not necessarily the best thing in an investing class.

I saw the likely Druckenmiller quote, and it’s for public investing, but it’s been sticking with me all week. He said something like, “Most of the time, we sit around here waiting, reading, and thinking.” I thought that’s a real investor, right, who understands that activity is not everything.

What is true for them is that what they figured out is that the byproduct of doing early-stage really well would allow you to deploy 3 or 4 times more dollars in later-stage, and the combination of the two could be effectively managed and would be disruptive up and down the chain. That’s the aha from them, right?

I did the math 2 weeks ago, and you do it again. If they’re 18% of the funding last year, annualize that over 2 years. They’re 10% of the Series As. They’ve got to be 10% of the good deals, they’ve got to be 10% of the great deals, and they’ve structurally figured out a way to make that happen, right?

So, yeah, I think that’s the victory lap from this. That was probably one of the things about them that struck me the most. The other stuff was all about some version of what Jason was saying: we’re all still fine. The valuations are fine. It’s expensive, but not ’99 levels. Yeah, we’ll see.

Guest 2

Not to be a little glum, but what happens if Mark or Ben step down? Especially Marc, for a variety of reasons. What happens? Another option is to wait them out.

This is an autonomous firm, and things happen. Even health scares happen, right? People get tired. You think people are all excited, and then Dustin Moskovitz quits Asana out of the blue. You don’t know. Everyone’s smiling; you just don’t know.

Can Andreessen Horowitz—I know everyone’s going to say there’s Martin, there are all these great people—survive at this level, going to Harry’s point, at this elite level, a generational transition? Can it survive that, or is it always going to end up being shirtsleeves to shirtsleeves in 3 generations?

Rory O’Driscoll

Yes, I would assert vigorously that the answer is: can it survive? Of course it can. One of my favorite quotes—I think I’ve said it before—is that the graveyards are full of those indispensable men, right?

Let me just recite the names for you: Kleiner Perkins, Caufield & Byers. Firms that proactively manage succession planning can make it happen. They’ve gone to 2 or 3 generations.

Yeah, Kleiner still exists. Mamoon’s doing a nice job. He’s not Kleiner.

Guest 2

But this one is so—like, the world’s changed, right? This one is so iconic.

Rory O’Driscoll

But, on the other hand, yes, but in a weird kind of way, it’s harder. I’m actually going to push back the other way. It’s actually—and this is one of a16z’s big insights—harder to be someone like Benchmark, small and brilliant, and manage generational transition, which is why it’s awesome that they do it, because the asset is the brains of 4 or 5 individual people.

The beauty of what a16z is clearly trying to do, and why I think they’ll be able to manage it, is that they’re basically trying to transcend the individual by just being an institution. Their fundamental bet has been that venture capital is going to go the same way as investment banking.

It used to be dominated by individuals and small partnerships, and now it’s dominated by Goldman Sachs. It used to be Salomon Brothers and Mr. Solomon, and they all went public. They’re all just a very different business. That’s the bet they’re making.

Guest 2

Oh, yeah. I’m not saying that if there were an unexpected transition, people wouldn’t make money, right? The question Harry had was: this apparent dominance right now, could that survive the loss of Mr. Andreessen?

Beast, right? Could it survive the loss of Elon Musk? The question—I don't know.

Rory O’Driscoll

Yeah, I would not like to think of the Tesla stock price if Mr. Musk decided to move back to South Africa and retire.

The shadow of Mark Andre is long, even when he's quiet on social media.

Rory O’Driscoll

A generational transition would be harder for a firm like Andreessen Horowitz or Coatue Ventures.

Guest 2

Look, there's an insider-baseball thing here that I don't know. I'm not gossiping enough to know. I honestly don't know if Vinod Khosla wants to build a generational fund. When I'm just guessing as a brand guy, when I look at how it's named and I look at some of the—he has some of the best talent on the bench.

Rory O’Driscoll

I thought Vinod's plan was not to die. So none of this matters.

Guest 2

It might not matter. It's a good plan.

Rory O’Driscoll

Yeah, it's a great plan. I'm with him. And if he can figure it out, it's a good idea, either for real or in the GPUs, one way or another, to not die.

7. Salesforce Wins $5BN Army Contract: The Last Laugh for SaaS

Guest 2

If anyone will, he will. So there we're not. Harry, we're going to avoid your question because I think we have a sense of what the answer would be.

You're never going to die and you love the game.

Rory O’Driscoll

Never going to die.

Rory, you'll remember that Jason said—

Rory O’Driscoll

And he could just invest his own capital infinitely. If anyone disagrees, as long as he's going to live to 300, he can just invest his own billions, right? You don't need any years of longevity. So it plays very nicely into that.

8. Wealthront IPO Disaster: Is $1.5BN IPO Too Small?

I do want to discuss public markets because we saw EquipmentShare's IPO pop 33%—an $8 billion market cap—growing 47% at $4 billion in revenue. Great IPO. The IPO market's open. Are we feeling great about this?

I think it's a good IPO. I think that, again, it points to the need for scale and profitability, and it's a very different IPO. Just for everyone's background, EquipmentShare is a technology-enabled equipment-rental company for construction. If you're a builder in, you know, pick any U.S. city, and you're doing a job and need to rent diggers, conveyors, or whatever other equipment you need, these are the guys to go to, right? A great story, a 10-year story, real critical mass, making money. It's inherently a physical business with a digital overlay. At the end of the day, there's nothing digital about a piece of construction equipment. It's a large yellow or green-painted thing that digs up dirt and moves it around.

It's a grounded business, but they seem to have built, in large part using digital technology, a pretty compelling business. So, go team. It's probably good news for all the other $2 billion digital construction companies out there, right?

Rory O’Driscoll

Growing 47% at that scale and profitable.

Guest 2

If you're at billions in revenue, growing 40% and profitable, with outlier margins for your segment, then you can IPO in an effortless fashion. I view this as an effortless IPO, which was interesting. It was really oversubscribed. You just IPO, you trade up—there's no drama. This is what an IPO is supposed to be.

And fun to see. It was a Y Combinator company from 2015. I would love to go back and look at everyone's notes as they sat through Demo Day in 2015 and what they said about the equipment-rental company from the heartland, because Y Combinator and Lead Edge both made a lot of money on that one.

Does this start a floodgate of this size of outcomes going public? Do they see the 33% pop and a good IPO, as Jason said, and say, “Okay, the market's ready now for us”? Will this start a flood?

Guest 2

Well, Rory was saying—when I pointed out in the notes, I thought the contrast to Wealthfront was: here's one that wasn't good enough for the markets. It's a very good company, a company whose software we admire, that has done a good deal of good in the world, making more efficient investing very, very easy for people. It doesn't seem to rip consumers off in a lot of ways.

This was an IPO that the market said shouldn't have happened. It's a deeply broken IPO. It's trading down 30% or 40% from its IPO, and it's subscale. The markets are saying, first of all, this wasn't worth remotely what we IPOed it at. It's only worth $1.3 billion, not $2-something billion. It's down 36%.

A billion dollars, you know, that's nothing for OpenEvidence or friends, but that is not—you're barely public. You lose the liquidity. You lose analysts. We can say they IPOed, but it's going to be a long haul for everybody to get their money out of this company, right? For employees, maybe it's fine, right? But it's barely public.

Rory O’Driscoll

I still have more than vestigial affection for this company. And, yes, it does kind of suck. Some part of it may be timing, but it does point to the low end of the market-cap space being a perilous one, because you fall a little below it and you do end up in that, you know, $1.2 billion to $2 billion range. There are companies doing $5 million in ARR that are raising at $1.2 billion, right?

And here's Wealthsimple: hundreds of millions in revenue, billions under management, in the same space. I think they will compound out. I actually like the company and have a mental note here to go check on it, see the valuation, and maybe buy some. But you, Jason, are right: it's not going to be a liquidity event in the short term, because there is not going to be the liquidity.

Again, it gets back to the—you can say it's fortunate or unfortunate, but it doesn't matter what your subjective opinions are of it. The objective fact is $3 billion, plus or minus, appears to be the point at which it's easy to go public, and it gets a lot easier the more you go up from there, right? Maybe $3 billion is a cutoff. When you do something at $2 billion and then you slip even a little bit, you're down into who-cares land, which sucks.

My worry is, honestly, if companies are created, maintained, and grown by the people within them, do the best talent really want to go to Wealthfront in a 30% to 40% down IPO? I didn't mean that horribly, but is that a magnet for the best talent today, given the many options they have? And if not—

Guest 2

Genuine comment: I think if—look, if you're an AI engineer, no. If you're actually interested in finance and investing, I think it's a very compelling space to go, because I think the things they're doing are super interesting.

Do you really—is it a top-5 place? No offense.

Guest 2

Well, let me give—can I give 2 things? I think 2 things could happen. If you have a deeply driven and charismatic ex-CEO on a mission, you will at least find a way to attract a handful of leaders to even a company with that struggle. You will find a way if you're utterly tenacious. I believe you will.

They may be failed founders themselves, which is the hottest recruiting category in tech right now, right? Failed founders. You may find them other places, but you will find 2 or 3 folks that can move the needle, and it's all you need. You only need 2 or 3 leaders at a company of any scale. The best ones will find 2 or 3.

At the same time, I have to tell you, when I talk to companies like this—and I've done several of them recently for the start of the year—I feel like people are just blinking at the camera. They joined these companies to not work.

Rory O’Driscoll

They joined these companies so that—I had to argue with one of these companies that I'm just friends with that they didn't want to get a big release out this year. There was a lot going on to get a big release out this year. You're going to get destroyed by the competition.

So, on the one hand, you can do it, but you better be, in my opinion, this CEO on a mission and reboot the company and find those folks. But realize 90% of your folks, if you're not careful, are just going to be blinking at the camera. They're just going to be blinking at the camera: “We need to slip that release. Well, these next quarters look soft. Actually, Harry, Q4 is looking great at the end of the year. This year, I know Q1 and Q2 are going to be down, but we'll make it all up in December.”

I'm going to push back on this because we live in a power law in terms of outcomes. We say only a few outcomes matter. Therefore, all the other outcomes don't matter, right? Which is—yeah, which is mathematically true about company results because that's the distribution curve for outcomes.

But the distribution curve for humans, just for the record, is actually pretty much a bell curve, right? So the idea that even in the good company, not everyone's going to be exceptional, right? There's an implied statement behind what you're doing, Jason, which is all the great people are in a great company and everyone in the okay companies is mediocre.

I actually think they're 2 different distributions, right? Probably the great companies skew a little better than the average, but most of the time, once you're up to 1,000 people, you have a fairly representative subsegment of whatever class of people you're hiring, right?

So I don't believe all the people in something like—well, yeah, a solid-outcome company like Wealthfront or even EquipmentShare are mediocre, not trying. I don't think that's—I think it's an overgeneralization, along with Harry, who's got his confused face on, and I can't explain it better right now.

Guest 2

And of course you're right. The reality is, even at the best, 80% of folks are not contributing significant value mathematically, but you've got to have these epic leaders and ICs to compete today.

I totally agree. But great leaders are everywhere.

Guest

It’s so competitive. You better find this founder who can truly bring this talent in in a magical way. And it does happen. We’ve all invested in a company, and often it’s a company that plateaued and then reaccelerated. The CEOs find a way to hire through that plateau, right? That crappy 6 months. So, we’ve all seen it, but you better not pray it’s there because you like the product or you like how you used it in 2023, because brutal.

We said about my last one—we said about small and subscale IPOs. Ethos, the insurtech funded by USAA, a provider of life insurance, is going public today, when this show airs, on the 29th, valued at $1.3 billion, at the high end of the range. The last valuation privately was $2.7 billion, at its peak valuation. Is that subscale, and as a result, should they not be going public?

Guest

They should. What do you mean by “should,” Harry? Why shouldn’t it? What’s your alternative plan for this company if it does not go public? Just curious.

Continue being funded by its existing investors. But if its existing investors think that the return profile from here is more akin to what a public investor would want, then they can go public. Does this not slightly feel like being fed to dogs?

Rory O’Driscoll

I mean, we all know that. No, Harry, again, this is the standard. It’s the edge. It’s a discussion we have every freaking week.

I don’t plan, despite his $4 billion, to run my life on where Chamath’s lines are, right? I think it’s true.

That’s a good tweet.

Rory O’Driscoll

Yeah, please don’t. I’m not trying to be argumentative; I’m actually being complimentary. Look, I think what is true is that at this kind of valuation, at this kind of market cap, it is harder to get liquidity, as Jason said. But you don’t know what’s going to price, you don’t know what’s going to trade, and maybe they don’t want liquidity now. Maybe it’s a process of starting, and over the next 1 or 2 years they perform and grow. You grow 20% year on year, whatever it is, 30%, and you just, over time, build up your market cap.

I don’t think we can all stay private forever, being passed around amongst us, especially if we’re not growing at—I mean, the venture cost of capital should be around 30%, and the public cost of capital should be around 11%. There does come a point when you’re better off in the public markets.

Guest

And if it’s subscale and cheap, let me tell you what will happen: people will buy it and they’ll make what’s called a capital gain. Just like you—if I believe, I mean, if I have the courage of my convictions, I should go away, look at the Wealthfront numbers, and say, “I believe it’s cheap at $1.2 billion. I should buy,” because in the end—

So, maybe a different version. It’s going to go out on the 29th, Jason, so we can ask Rory now because he’s just said about a capital gain.

Rory O’Driscoll

That could come.

Will this have a pop, or will it have a drop?

Guest

Well, I think it’s going to have a drop. It shouldn’t—I mean, notwithstanding Bill Gurley, my personal view is IPOs should be engineered. Things matter. Just like the start of the conversation on Brex, it does matter to feel good. There are a lot of benefits to feeling good about the IPO.

I think it won’t, but I do think the interesting question is—maybe this is what you’re asking, in a sense, Harry—Rory may disagree, but in a sense this is a capitulation by the investors saying, “We’re not getting back to that $3 billion valuation or $2-something-billion valuation of years ago. It’s okay, right? We need liquidity. There needs to be an exit path. No one’s offered to buy us for a good price in the last 3 or 4 years, or they would have taken it. So, it’s time. It’s time, boys, to IPO.”

If this one works, will we see a flood of these? Even if they’re like Wealthfront or others and trade down, if the market will absorb them, will it be time to flush our 2021 unicorns out the door and down the drain—but out the door—in 2026 and 2027? Maybe it’s time to just let them go, no matter what price it’s at.

Rory O’Driscoll

You’re veering on doing the thing you condemned others to do, which is sneering at a $1 billion, $2 billion, and $3 billion outcome, right? It’s the low end of the public-company market cap, but it’s a perfectly good outcome and to be congratulated. To start a company from nothing, get to $1 billion or $2 billion in value, get it public, and have the chance to compound for 5 or 10 years—it’s an awesome achievement.

And you’re right. We’ll all, at some point, look back and go, “The winners were the ones who got a crazy valuation in 2021 and then were able to get out from under that valuation via a down round, a down IPO, or a down M&A.” The bad ones are the guys who are still sitting there looking at their 2021 valuation and thinking that’s ever coming back, right? So, I give everyone credit. I give anyone credit who’s making progress on clearing the logjam and doing what it takes.

I remind you, there are 7,800 unicorns valued at more than $1 billion, and SVB did that analysis. I think something like 3% of them have a decent growth profile and scale—call it 200. At 1 a business day, it’ll take a year; at 1 a week, it’ll take 4 years. So, yeah, I think this has to start happening.

One of the things that’s true, Harry, is price clears all markets. In other words, there’s a price at which public investors will say, “Yeah, I’ll buy that.” Maybe it’s not $2.1 billion. Maybe it’s not the price he wanted in 2021, but if you want—I mean, you call it being fed to the dogs, which clearly was a very pejorative statement—maybe it’s just figuring out what price it takes to clear the product.

I’m an adviser to a unicorn that’s doing hundreds of millions, growing pretty well—not being killed by AI, but not necessarily benefiting from it, but enough to eventually compound to a better IPO than that, but not an OpenEvidence. We did an M&A review a little while ago, and I was shocked by who was on the block. I mean, everyone’s for sale, and I was shocked that folks worth much more are willing to be acquired by someone with a fraction of their revenue. These are some—I didn’t see Databricks among the ones that were passed around the room virtually—but I was shocked by the companies that have been on 20VC that I did not know were in market and are aggressively looking for an exit.

Which is why when you see people pull off an IPO or pull off a freaking $5 billion exit, the correct response is, “Yay, well done,” and then, B, “Goddamn, I wish one of my guys could do that. That’ll be great.” It’s a great outcome, right?

Guest

Yeah.

You know what? If we’re going to end on that, I do think the hat tip deserves to be given to our friend of the show, Mr. Marc Benioff, who the Army just awarded Salesforce a $5.6 billion contract over 10 years. Mr. Benioff, hat tip. Well done.

Rory O’Driscoll

Totally. Yeah. Yeah. SaaS is not dead. And now SaaS has an army. I love it, man. SaaS has the army, right? Yeah. Take that, doubters.

First of all, you’re exactly right. Great outcome. And some sales rep at Salesforce is getting the mother of all commissions here, and good luck to them. But I also think it speaks to the whole zeitgeist of “AI is going to eat everything.” I think the correct pushback has been—and you mentioned the AI—the correct pushback, I think, has been people saying, “All these systems of record like Salesforce are going to get replaced.” Yeah, they’re obviously wrong, because that’s not what’s going to happen. It would be an incredible waste of talent to do that. You should just live off the systems that we have, and I think this is an example of that.

It’s a separate piece. The AI piece is: how much value can you build in the AI-first world as a system of record? That’s a totally legitimate question. Yeah, can Salesforce get its mojo and growth back, or is it a utility of SaaS stocks for the next 5 years? That’s a fair question. But I think deals like this put to rest anyone who thinks that they’re going to vibe-code their way to a product that can replace a $500 million Army order.

I’m really sorry. I’m dumb as rocks still, after many shows with you. My question to you is: when AI sales reps work and you have distribution to the scale that Salesforce does, I don’t see how they don’t regain growth and become a dominant force again.

Rory O’Driscoll

Interesting you should say that, because I was commenting on the people who are extreme, who are kind of—look, the market is saying again right now, “Oh my God, SaaS is dead.” The trading multiples are down, and you’re saying almost the exact opposite.

Again, I would remind you that you can take some of your ill-gotten gains and bet them on the public markets if you want to. I think I’m kind of in the middle. I don’t think they go away, but I think it’s what Jason said: they don’t go away, but it is hard to do that innovation that gets that new product out the door.

Guest

One thing that we do get confused about is we think it’s all directly AI, and AI is the biggest issue because the lion’s share of the new CIO’s budget is going to AI, right? That’s where all the discretionary budget is, and price increases.

Guest 3

And price increases are almost self-defeating because they only work for so long. If you're not tapping into the AI budget, that's—you know, Marc, when he was on this show and on his own show, was saying how much better a deal Palantir got. I think that was echoed in some of these Army contracts, because maybe he had to take a haircut on those deals and get the budget where it is. And so it's there. It's just that there are so many other issues, unfortunately, that are attacking SaaS.

Seat contractions are existential. Workday said seats are perpetually under pressure. Shopify has held headcount flat for 3 years and grown 40-some percent in that time. We're not hiring anybody, and we're going to hire fewer people.

Price increases have become destructive because SaaS product prices are up 40% over the last 3 to 4 years, and that's great for a CRO to make their plan this quarter, but it crowds everything out. There's no room to upsell or anything when price increases take up everything.

And so there are all these different issues. If we're not buying as many seats and we're radically increasing pricing, there's just multiple ways the old model is getting attacked. And if it were just as simple as adding an agent, that would be hard enough. But it's not. Our workforces are shrinking, and what we expect from our workforces is shrinking.

And so SaaS will adapt, and sure, we will get there, but just magically charging per token doesn't necessarily change the fact for many providers. Just tweaking pricing models doesn't change how much folks want to spend for a product, right? That's a fallacy. That's what consultants do. Their pricing consulting is great, but if no one wants to pay more than $20,000 a year for your product, you can't force them to with a clever pricing model. So there are a lot of issues to deal with that the new guys don't have to deal with today.

So you end up in that boring kind of quadrant: it ain't going away, but it ain't exploding.

Guest 3

It's just—there are so many threats attacking. It's so hard because there are so many threats attacking. If it was just AI and it was nothing else, yeah, you put 2,000 people, like Marc did, on Agentforce, and it works or it doesn't, but it mostly works, right?

But if at the same time folks are contracting seats, right? If at the same time they're cutting budget for existing investments, right? If at the same time you're beholden to price increases to make your plan, it's just being attacked from so many sides. I worry that, for all but the best, there are too many daggers out. It's being attacked from so many sides.

We just got used to these 130%, 120% NRR years that were magical and often didn't even rely on price increases, right? Slack never raised prices and still grew at 140% NRR. I don't know if those days are ever coming back, no matter how good our agents are.

Okay, boys.

Guest 3

Time to rock, baby.