VC里的45亿美元退出够不够？Harvey融资1.5亿美元，以及为什么该买入 Google、卖出 Amazon

Boys, we are back. We have some big news this week, and we're going to start with some liquidity, baby. We're going to start with Navan's IPO.

In the theme of discussing and providing some context, Navan obviously IPO'd this week. Oren Zeev invested $150 million and returned $1 billion, and then it slightly cratered, down 20%. I'd love to start there. How did we analyze this? Take me through your thoughts.

Jason Lemkin

Well, look, I'll let Rory do some deeper analysis. I'll tell you, the whole thing—and Dev stepping down from MongoDB—made me kind of wistful. It just feels like the very end of an era, the end of the SaaS 2.0 era.

Dev stepping down after an incredible run at MongoDB—the company dipped to around 13% growth, and it's now back well into the 20s. It's a good run, handing off the baton. I posted on Twitter that it was the end of the era, and Brian Halligan was like, “Yeah, a lot more are going to retire soon, right?”

For me, Navan was kind of a bummer. Ariel is such a tenacious CEO. It's a great IPO. Rory's going to tell us it doesn't really matter if it was down or up; you got it done.

Bill Gurley won, but it's just kind of a bummer that a company at $700-plus million in revenue, growing 32%, struggles in its IPO at $5 billion. All of our portfolio companies have to do much better, right? They've all got to be Harvey or better. All of them.

It's not discouraging. I just felt wistful that the combination of Dev retiring and Navan's stumble at the IPO felt like some of the last pieces of the last era. So be it, right? We're in the age of AI. So be it. It's time to move on to the new era, boys.

Rory O'Driscoll

Yeah, I mean, wistful. I didn't have you down as a wistful guy, but I think there was actually a lot to unpack in this. First of all, let's start with the basics: it's a great company. It survived a near-death experience.

It's a travel company, and that's a tough place to be when COVID happens. It came back from that. The investors stepped up and financed the company. The CEO stepped up and kept the company alive. Now they have, zooming out a million miles, what is a great outcome.

At the day of the IPO, it's a $6 billion outcome; now it's a high-$4.8 billion, $4.9 billion market-cap outcome. Big picture, it's great. Jason's right: 2 years from now, this stuff will all be in the noise, and it'll be seen as a really solid outcome.

It's a good company. It's operating in a competitive space, even though the financials look a bit messy. That's the thing that counts. We can then talk about the specifics of the IPO.

Look, what happened is they priced the IPO in the middle of the range, and it traded down a little on the first day, and then, I think, on the third day, quite a lot. It's at $17 a share right now, so a pretty tough debut from its IPO to today.

Jason, you said something that I actually disagree with. You said Bill Gurley would be right. Actually, this shows that sometimes Bill Gurley is wrong, because his idea is that these IPO share allocations are “free money.” You get your IPO, it always goes up, and they're leaving value on the table.

This is an example of an IPO where, if you look at it from the perspective of the issuer of stock, not only did they not leave money on the table, they actually priced at what now looks like a high. It's proof that Bill Gurley's wrong, and that the buyers of IPO stock are correctly saying, “I want a discount on the good ones because every once in a while—and you don't know when—this goes wrong, and the stock goes down.”

I want the 20% pop on Figma, which became obviously a much higher pop, in return for the risk of a 20% drop on Navan. I would argue it actually is almost antithetical to what Bill's saying, which is that it's free money.

Most of the time, it is. It skews positive. The day-one pop, on average, makes you money. But it does show that every once in a while, for circumstances that we'll talk about in a second and that are hard to predict, things just go wrong and it blows up in your face.

Can we touch on something that people don't talk about often, but is often reported? There are the winners and the losers, and their stakes are always reported very loudly in the media: Oren Zeev, from $150 million to $1 billion; Lightspeed, from $257 million to $1 billion; Andreessen's stake, worth $635 million.

A lot of people take that at face value and think that is cash in the bank today for these funds. Rory and Jason, you've been through IPOs. That is not cash in the bank today for these funds. For people who aren't aware, what does the lockup period look like? When does that turn into cash in the bank? Can you just share some insight on that?

Rory O'Driscoll

Sure. The typical lockup period will be 6 months, which means the minimum time when you can start selling, other than in a registered secondary, is 6 months from now. At that point, investors can start to sell, and maybe they're allowed to sell or distribute their stock.

It's highly likely that it takes at least 18 months to get out of your position in a company like this. If you have normal appreciation, on average, that can actually end up doing better than you priced at the IPO.

But if you have, for example, the Figma thing, where you have the pricing and then a huge first-day pop to $140, everyone reports, “Oh my God, XYZ investor made $4 billion.” Then you fast-forward a year and a half, and they very happily make $2 billion, but maybe it feels like a little bit of a disappointment if you're mentally spending that $4 billion that you thought you had for 24 hours.

The IPO is significant, but the economic significance is limited. In fact, when we used to track exits every year, which we do, we used to track them as of the IPO, and then we used to call it the locked-in value 18 months later.

Our mental model was that when you want to figure out how much money people actually made, look at the market cap of the IPO 18 months later. That's probably a much closer estimate. In some cases, it goes up massively, and obviously, in some cases, it goes down.

Jason Lemkin

I don't know what Scale's policy is, Rory. I'd be interested to hear. When I entered venture, what I was taught back in the day was: base case, distribute for 24 months ratably after the IPO.

If you didn't believe in the company, you might pull it forward. If you thought you had a Veeva in your pocket or a Shopify—maybe Bessemer didn't—you might hold longer. For sanity's sake, and also to manage float, because you can only sell so much if you own a large stake, that was a rough rule.

We figured it would take you 6 months to get through the lockup and 24 months to distribute. That's 30 months total after the IPO before you're getting most of your carry and your LPs are getting their distribution. It could very well be into 2028 or 2029.

Now, there was almost $200 million of secondary, but I don't think the big guys sold any, as near as I can tell. The founders took out $50 million, which I like. I like that much better than in the seed round, by the way. I'd rather see them take $50 million in the IPO than after demo day. I think it's a better time.

For the most part, it looked like it was just the smaller guys that sold. I could be wrong there, but that's what it looks like.

Does this impact your price sensitivity today when investing? When you see $700 million in revenue, 30% growth, positive economics, a $5 billion market cap, and then you see some of the prices that we're paying?

Rory O'Driscoll

The problem is, the prices you're paying are for things going at a very different rate, which they should. Let's unpick that.

At a minimum, your operating assumption should be that, for mature companies—SaaS transaction-type companies, whatever—we are back to 6 to 7 times NTM. It's not just SaaS; it can also be non-recurring-revenue companies with a decent margin profile and 30% growth.

If you recollect, a while back I said, “That's kind of like the 10-year Treasury equivalent of SaaS.” That's what they're worth. If you own one of these things, that's how you should think about it: what it's worth.

That impacts how you think about your late-stage portfolio and how you think about what a little extra growth is worth. But that's not directly applicable to some company doing $50 million and 5X-ing, or doing $10 million and 10X-ing, because those different growth rates mean it's not a like-for-like comparison.

Jason Lemkin

When that 10x year-on-year growth rate decelerates to a 30% growth rate, then you'll probably trade at the same valuation multiple as the companies that are already trading at a 30% growth rate, which is 7x. In the end, if the growth rates of the new companies become much closer to the growth rates of the existing companies, they'll trade at the same rate. There's no magic there. Right? But right now they're not. Right now, AI growth rates for these companies are in a very different place, as you know well, Harry.

Listen, of course, it's very interesting to see Navan at the same time as OpenAI and Anthropic raise their estimates, right? OpenAI raised its estimate to $100 billion-plus in 2027, right? I forgot what Anthropic's was; we should look at it. They just raised their estimates way up.

So when we see this, let alone the Harveys—Harry, let alone the Harveys—when we see these, it's just, for me, I don't even want to take meetings with mortal founders. I don't even want to take them, and it's terrible, right? I mean, literally, I just did a deal with some founders I love, and it was very expensive for me.

I did a deal at $50 million post, with everything else in it. For me to make a 100x on that deal, with dilution—and dilution in many cases is higher these days for a lot of reasons—it has to be better than Navan, for sure. It has to be way better than Navan for me to make enough money on that late-seed deal, right?

But do I really believe this deal I just did is for sure going to be worth more than Navan? I don't know.

[laughter]

And when we started, Rory, these deals would be—and I'm not being curmudgeonly or anything—but realistically, they were easier to see. Actually, maybe it was always hard, but now you have to see these $10 million exits. If they're not utterly breaking the mold, if they're not breaking the mold, it's hard to really believe it's going to be worth north of $10 billion, is it?

Rory O'Driscoll

So Jason, I get you, but I've actually adopted this mindset from spending so much more time with you. And I bring it to the IC, and people on my team are like, “You have to turn the next card to see sometimes.”

It's not obviously a $10 million company on day one, and actually, value can accrue in increments over time. You could miss some great ones by being flippant and being like, “Oh, it's not a $10 billion company.” You will.

Jason Lemkin

It just depends on your fund size. If you're doing more second- and third-checks and the first check is smaller, you have a lot more flexibility, right? I mean, literally, you can. If your first check for me is a very large percentage of the fund, I don't have a lot of margin for error.

If I'm writing 4% to 5% of my fund as a first check, I don't have the other $145 million that he had in TripActions. I just don't have the other $145 million. I should, I guess, have developed more SPVs and opportunity funds, but for me, that first one has to work.

If it doesn't have to work, then you want to—I mean, you want to play more cards today. I do think it's a good idea.

Rory O'Driscoll

I recoil from the “I don't do mortal humans” thing. I think it sounds a little judgy, dare I say it, Jason, but what I do think—

Jason Lemkin

I don't want to judge.

Rory O'Driscoll

What I do think—

Jason Lemkin

To the mic. Just come closer to the mic and then judge me.

Rory O'Driscoll

I think the sobering fact here is that you now have to assume that $400 million to $500 million is the threshold for an IPO. If you say to yourself that you only want to do deals where at least half the upside of an IPO—the IPO potential—then the bar for what is a doable, successful venture-backed deal with upside has gone up.

Right? I mean, it's what we discussed. We talked about it in our “fewer but bigger winners,” right? At any stage, if you're keeping the stage the same—if you were doing seed before and you're doing seed now—before, that seed was an 8-year journey where 20% of them got to the end of the line, and now it's a 12-year journey; maybe only 10% get to the end of the line, right? And that's just mathematically true.

Now the real question is, what do you do with that information? As Harry said, you can have one of two approaches. You can either say a priori, “I'm only going to do the $10 billion ones,” which is one approach. The other, Harry's approach, is more: you never know up front which are going to be the $10 billion ones, so you do them and you look at the next card and you play it out, right?

It's a little bit—if you have optionality, you can afford to do it the Harry way. If you're picking and most of your dollars are going in when you're going in, Jason, then you're exactly right. At some level, even though I don't like it, I'm coming back around, and your comment is correct.

I don't like the description of mere marketing, but you do have to go into these deals looking for a higher, believable exit story, given that the exit bar's gone up. And you can't do clever little small markets that are going to top out, because you're probably only looking at an M&A outcome, and then you've just intrinsically eliminated the magic pixie-dust part of the alternative.

Jason Lemkin

This business has got harder.

Rory O'Driscoll

It has. It has. And the other interesting thing to note—well, yeah, but also people are getting richer at the same time.

Jason Lemkin

[snorts] Yes, in a very concentrated fashion.

Rory O'Driscoll

Yes. Yeah. No. And I think the other interesting thing just about Navan, to make it, is that what's really interesting—you cited the amazing numbers for Lightspeed, Oren, and all those guys. The other thing it brought home to me is the amount of dollars that went in.

If you look at it, I think you said Lightspeed maybe 5x or 6x—don't quote me. What that means is they did the seed and some of the seed in the A and the B, right? I'm willing to bet, to Jason's point, the multiple on those rounds must be 20x-plus, right? At least, maybe 30x-plus.

But what you're seeing is, instead of being a $20 million investment getting diluted down, while still getting a magnificent 30x return, you're following that $20 million up front with $200 million more on the mid- and late-stage rounds. Your overall blended return is a 6x, but it's a 6x on a lot of money.

You probably have some early-stage dollars that are 20x and some late-stage dollars that might even be a loss, given the last-round price of $9 billion. Overall, obviously, it's a wildly successful strategy. It just brings home again that you're diluting your early return, but in return, you're doing that because it goes back to what we said last week.

When you have one of those winners from your early fund, and you have the amount of dollars these folks are managing, you just have to put every dollar you can into your winner. They did it here, and it worked—worked successfully.

Even more impressive is that Oren Zeev did it with a smaller solo fund and ended up with $150 million, via, I'm assuming, a bunch of SPVs into his biggest deal. Good on him, and across many different vehicles, which is awesome to see for Oren. I mean, Lightspeed was $257 million to $1 billion, so it's just under a 4x blended.

The horrible question—maybe we don't spend all the time on it, but the horrible question in venture and startups, it is horrible, is: Is a $4.5 billion exit good enough today? It's a horrible thing to say, but if we're talking about VC and inside baseball, it's a bona fide question. Is it good enough today? $4.5 billion.

Jason Lemkin

It is. It is. It is. They took $200 million—I mean, the interesting thing is, they took $200 million. No, it's not if your fund size is $1.5 billion or $2 billion. It's a third of the fund.

Rory O'Driscoll

No. Yeah, yeah, yeah. I mean, it's still a third of the fund, though.

Jason Lemkin

I mean, look, you're not going to have guys—

Rory O'Driscoll

I spent so much work, and I only got a third of the way to 1x, Rory.

Jason Lemkin

Yeah, but the point is, those are different.

Rory O'Driscoll

Good God, I spent 12 years with these amazing founders. We had a $4.5 billion exit, and I only got a third of the way to 1x. You never told me this was such a bad job when I joined the fund. This is the worst job ever. The perks are great, the dinners are fabulous. Tech Week was so fun, but this is a third of 1x? This is the worst job ever.

[laughter]

Jason Lemkin

But yeah, because you're looking at it through your lens of a seed-stage investor where you want your best deal to return the fund, right? The truth is, as you start putting these kinds of late-stage investments in—look, we just said it—the early-stage investment, your first dollars in on a deal like this, was probably a 20x-plus. Your first round probably did return the early-stage fund.

But then, instead of just stopping there, you decided—you being whoever you are—to raise a late-stage growth fund. You're not going to get a late-stage investment returning the fund without massive deal concentration. It's not a thing, because if you're doing 20 deals evenly, that's 5% each.

Unless you get a 20x, which you typically don't get in a late-stage deal, you're not going to return the fund. So these guys aren't sitting there going, “I'm raising...” They might be saying, “On my $300 million to $500 million early-stage fund, one deal can make it happen.”

But on their late-stage fund, they're saying, “We got $4 billion; we're going to put it to work. The average good deal will be a 3x to 5x. We'll have a low loss ratio, and over time, we'll get our 2x to 2.5x.”

It's not the small-number-of-big-hits business.

Rory O'Driscoll

It’s a very different business. It’s moving money at scale, and the example of Navan is that they’re doing it successfully. The embedded risk in that business is price compression, and you’re seeing a little bit of it here, right, in the sense that the $9 billion round lost money. Some of the people who bought at the IPO lost money.

So the risk you’re running is not catastrophic wipeout as much as you underwrote a 6X, and now the thing is trading at a 5X and suddenly your return is down 30%. So, for those who came in at the $9 billion price, are they down 50%? I believe so, because I think it boils down to—I read it as—they convert 1-to-1. I bought the prospectus.

But, yeah, a fair amount. Again, remember, that’s only the price today. If you want to go back to 2012, you can find a whole bunch of dumb articles about how Facebook is a crap company because the IPO and the stock price went down. It turns out it was a 10X company from there, at least, right? So it’s a point in time.

But you’re right: as of now, the last private round and buyers who bought in the IPO are down on the month. It’s a horrible short-term ROI.

Jason, sometimes I think Rory just sits back and thinks, “I’m so lucky to do this show with Harry and Jason.” I think that all the time, for so many reasons. There we go.

Listen, Jason, you mentioned Harvey. Harvey raises $150 million at an $8 billion valuation, led by our dear friends at Andreessen. I actually tweeted about us talking about this, and one of those not-so-quiet investors at Harvey shared with me that they had $150 million in ARR. Their DAU-to-MAU ratio is 40%, which I thought was astounding.

Jason Lemkin

The growth rate?

They didn’t share that one. Okay, $150 million—let’s go. $150 million in revenue, right? Outstanding ARR. People are really buying more: 170%. No one’s leaving: GRR of 98%, right? And DAU-to-MAU—you’re impressed, but it just means they’re logging in every day to use the legal tool. That’s table stakes to me. They use it every day, right? It would be a flag if they didn’t, right?

Hold on. Let me put it in the Saster AI $8 billion valuation calculator. What did it go out at? $8 billion. If it’s $400 million in forward revenue, that’s 20X.

Jason Lemkin

Agreed. That’s exactly what it is: $400 million next year. They’re at $400 million ARR. That’s what they’re predicting: $400 million ARR, not GAAP, $400 million ARR, that’s my guess. And $150 million raised.

For Andreessen coming in, this is very small dilution-wise for the company. I love these rounds of getting 9 figures for 1% or 2% of the company. Honestly, it sounds like I’m being facetious. I do love these as a seed investor. They’re great. There’s no dilution—no effective dilution—to these rounds. To Harry’s point, they’re great, right? Now, there may be a little pressure on the exit, but there’s no dilution.

They’ve executed really well in a core domain where LLMs were going to have a profound impact. What’s fun about it is that, up until now, legal had been a pretty bounded place. In terms of software sales, you have companies like Filevine and Clio that have built decent-sized businesses but haven’t yet gone public. Not a lot had happened in legal.

I think LLMs, by virtue of the fact that they manipulate language—which arguably is exactly the definition of what a lawyer does—are a perfect fit. I think they’ve done a great job. They’ve established market presence in Am Law very quickly, established a brand quickly, and executed well. The growth is clearly there.

When you start thinking about what you’re really asking—does “$8 billion” make sense?—what that really boils down to is a TAM question. They’re clearly in the lead. LegalZoom is clearly second, focused on lawyers in what you may call corporate law practices. The constraint there will be the TAM size, right? How big is the TAM? How many law firms are there? How much spend is there per lawyer?

There are 1 million lawyers in America, roughly half in-house and roughly half external. Does the math support a $24 billion, 3X-from-here company? Going back to what we said earlier, let’s assume that, in the end, it’s a 30% growth company like everyone else and it’s a 7X multiple. That implies a $3 billion revenue line.

Is there a $3 billion software business selling to lawyers and corporate law? Not crazy. Westlaw is bigger, selling information, but that’s the kind of scale you have to have. You basically have to be such a big automation tool for these lawyers that they’re willing to spend thousands of dollars per year on a subscription to make the math work.

It’s a TAM question. They’re clearly going to have to be number one in that market, and the only question, if you were underwriting that at $8 billion, is whether this is a $1 billion-a-year spend or a $3 billion-a-year spend. If it’s a $3 billion-a-year spend, maybe you get there.

Jason Lemkin

Totally agree. I think it goes back to your statement from episodes ago, which is that the core determinant of our success in venture with the AI transition is whether we’ll see the transition from software spend to human labor spend—or, sorry, from human labor spend to software spend. I think that’s the TAM question.

You’re right, because selling software to lawyers is a particularly shitty business. It’s a more constrained business. You’ve got to do more, help more, and speed them up. But it’s not all or nothing, by the way. I’ve been seeing a lot of good literature on it. It’s not about automating people; it’s about automating tasks. You’ve got to make them a lot more efficient, and they’ve got to be able to track that. Then, if that happens, it all makes sense.

We mentioned liking these rounds for their low-dilutive characteristic, or nature. There was a good piece from The Information this week about Benchmark lowering their ownership requirements, with MaC being the example. They only have 10%, where they normally always needed 20%. We always knew this was Benchmark.

In general, has AI seen a reduction in ownership across the board for this generation of venture?

Every deal I’ve been in—my God—ownership is being attacked at a level I’ve never seen. That’s my conceit investing today: giving up on that, right? For me, I feel like I can only make money if I own double digits of 2 winners per fund. I feel like, mathematically, that’s the only way I can make money.

The last 3 investments I’ve done are in the 6% to 8% range, even though that’s my rule. But what am I going to do? Not do the deal? We also know that’s the dumbest thing of all time, right? So I’m literally going to write my LP report up in a couple of weeks. I’m going to say my resolution for 2026 is to get my ownership up.

We’ll see how I do against my resolution, but that’s my main resolution. You can do it a few times, but if you do it every time, it’s tough. I don’t know what you do in this world if the companies are capital-efficient and they don’t need you. It’s complicated. In a hot company, you don’t really control the dial, to use kind of lame VC terminology. You don’t control the dial.

Jason, why didn’t you raise $125 million and then you could have more ownership?

Jason Lemkin

No, no, that’s—maybe that was a different mistake. But this is really just all there is. Agreed. If there’s a co-investor, I guess you could be a total jerk and say, “It’s me or nothing.” I’ve tried that once in my whole career. It’s just not my vibe. It backfired on me.

Of course, that company wasn’t that successful, but if you want to be somewhat founder-centric, the best thing I know how to do is say, “Listen, I just need to be the largest investor, and let me invest the maximum I can in the round, and that’s as far as I go.” If you’re not willing to put the extra dollar into the deal, don’t do it, for sure. But if they’re selling 10%, it’s hard to buy more than 8%.

It’s just going to be hard. I think it’s multicausal. I’m breaking apart your question about Harvey. The one question is: are VCs getting less, on average, in these deals? The second part of your question was “because of AI,” which we’re going to agree is a meaningless phrase and try to fix it later.

On the first, I think they probably are. But there’s no doubt in my mind—I think it is harder to get 20% ownership for the super-early-stage funds. It’s hard at our stage. Our target would be 10% to 11% on average, probably late A or B, some earlier and some later than that. It’s harder across the board. There’s no doubt that’s the case, right?

We can take Benchmark as a premier firm, and I will be validated again. As I said, reports of their death were greatly exaggerated. That sounds like an amazing 2021 fund they’ve reported. But, yeah, that’s an amazing firm courting a great company and only able to get 10%.

What are you going to do? If they only need to sell 10% of the company to fund their needs, then you’re only going to get a maximum of 10%. You can either decide not to play, which would have been a very dumb decision, or you can decide to take 10% and keep going.

Right? Now, it's interesting. Why is this happening? There are a bunch of different reasons, because you can't say it's because they're capital-efficient. Let's be frank: some of these AI companies are the least capital-efficient companies in the history of humanity, right? Meritech is pretty capital-efficient. OpenAI is claiming to spend more money than we thought existed in most entire continents, and they're not stopping yet. So it's not capital efficiency—not capital efficiency, right?

On the capital-efficient ones, if you're so capital-efficient that you don't need to raise a lot, and you become hot very quickly, then you've got leverage as a founder. On the other extreme, if you're so capital-inefficient that you need to raise $13 billion, then it turns out that no matter how much you put in—if you put in $100 million—you're still only going to have a couple of points, right? Both of those, interestingly enough, will be pretty good deals, right? Which actually—and I'm doing this in real time—makes you realize that your mental rules of thumb have been smashed to pieces.

You can only get 10%. There's a situation where you're going to get 10% because it's capital-efficient, and there's a situation where you only get 1% because it's capital-inefficient, and both of them have amazing returns. So that's just the way it be, right?

And I think part of it—so maybe that's the hard part. There's a continuum. I mean, logically, you'd say to yourself, if it's a continuum, there must be some sweet spot in the middle where they needed to get 20%, but then they were capital-efficient enough, so you make out like a bandit. I'm sure those companies exist, too, but I think that's the hard part.

It's a continuum here. It's a wider continuum than we've seen, perhaps—maybe not in the early internet, as I think about it aloud—but definitely wider than we've seen in the SaaS era in the last 10 years, right? Companies, as you say, being able to get to hundreds of millions of dollars on $10 million or $20 million, and then other companies needing $2 billion or $3 billion just to get a model out the door, right? Neither of them results in a standard venture ownership position.

Yeah, but maybe not to press on it too much, and I do think when I was reading ICONIQ's latest report, what they said is that when you look at the top-quartile companies in their extended portfolio—not all the companies they surveyed—yes, they burn a lot of cash, but the burn multiples of their top companies are much lower because they're growing so quickly.

So even if you burn a lot, even if you ultimately raise a lot, if your burn multiple is low, you're able to sequence capital differently. You could say, “Listen, hold off. I'm going to do just 10% now, and then I'll do 5%, and then at $8 billion I'll do 1.2.” You may raise a lot, but if your burn multiple is low, even if it takes a significant amount of cash, it lets you optimize how you sequence fundraising.

Agreed. And the reason—remember, the words “optimize how you sequence fundraising” as an entrepreneur: the inverse of that is, “Thou shalt not be optimized as a VC.” Exactly. A founder's optimized fundraising is a VC's below-ownership target. You're exactly right. I think that's right.

If Marqeta had needed $40 million to get rolling, then Redpoint would own 20%.

If they only need $20 million, then there you are, right? And you're right, they can incrementally raise another 2 rounds. Even the extraordinarily ambitious—the other extreme—the extraordinarily ambitious foundation models, not the new ones today, but early Anthropic and early OpenAI, were able to do incremental fundraisings in a way that, as you say, avoided significant dilution.

I mean, even—and I say it with huge respect—even Y Combinator is structured this way, right? Y Combinator's advice for most companies is to raise a maximum of 10% before, during, and after Demo Day. It's very thoughtful advice. It's very structured.

Yeah, it's biased to helping YC, but if you think about it, it's also saying, “Listen, we've run the numbers. Most of you will do better selling 10% at Demo Day and another 10% at 3 to 5 times the valuation.” It's not just in terms of overall capital raised. It's not just valuation; it's effectiveness.

And so they've institutionalized this low ownership, right? It's always been true, but I think it's been productized in the last couple of years. 10% at Demo Day, right? And if you want to do 4% from angels and friends, that's 6% left for a VC. Maybe you get 5% or 6%, unless you way overbid and basically do 2 rounds at once. That's how you get more ownership in YC: you do 2 rounds at once.

I absolutely agree with you. I see 3 on 30 instituted so well by YC as the de facto round. Again, with huge respect to them, it's good for them and it can be good for founders, but it's a challenge for us to navigate.

Yeah, I'm not even criticizing. It's just institutionalized. You asked about low ownership. This is institutionalizing low ownership. YC's always been about low ownership with VCs, and I get it, but now it's been institutionalized very effectively. 3 on 30, 4 on 40, 2.5 on 25. You just got to get single digits.

This is where I was saying, just do a world of Roger Ehrenberg there. We had him on the show. The world of high ownerships: go where others aren't, get 20% in actually reasonably priced assets. Let's not do AI dictation tools from YC and get 4%.

Harry, did you see that Anthropic is now projecting $70 billion in revenue in 2028?

I'd rather have a piece of that. Pretty good. I'd rather have a little piece of that. And, yeah, I mean, did you guys see Sam Altman's response to Brad Gerstner? What did you think of that? I was intrigued because it was quite a retort publicly.

It's a totally legitimate and entirely obvious question. You're doing $12 billion in revenue. How are you going to find $1 trillion in CapEx over the next 5 years? I'm sure there's an articulate answer he could have made. You're right: the answer, “If you want to sell your shares, sell your shares,” was a little snarky, probably because you're tired, you're worn out, and you're halfway through a 1-hour interview. You do a million of these all the time. I believe you just have a baby in the house. You're tired, you're grumpy, and you just make a snarky answer.

Maybe to say it this way: the question is substantively important. What did you learn about the plan to fund the trillion dollars from the answer? You learned a little bit about the persona of the person in a bad moment, but everyone has bad moments, right? If you want to know more about Sam Altman, then there's 53 pages of testimony now on the internet. You can figure out—you can all come to your own judgment on that.

I think the question itself is totally legitimate. It's going to be asked increasingly. There's a story that can justify it. It's all about revenue traction. If you get to $100 billion, you can support $60 billion or $70 billion in CapEx. But it's a totally legitimate question.

He did say, didn't he say to Sam, “If you want to sell your shares, I'll find someone for you in 60 seconds”?

Founders have only said that to me once in my career. I never said a critical word ever again. When that was said to me, that was a teaching moment for me. I'm like, “Okay, I crossed the line.” I didn't mean—I didn't realize I did. I never said a critical word ever again when I was told, “There's a market for you. Just let me know how much you want to sell, Lemkin.”

But no, I'm not doing the “Oh, damn you for saying that.” We all have bad moments, right? But it is kind of a [__] answer, right?

If that happened to me in a board meeting, if I were a board member and CEO, you might say to yourself, “Oh, I want to find—” If it's a legitimate, company-ending question, you have to have an answer to it. Because, stepping back, it's not just the health of your company, but bizarrely enough, the health of the entire U.S. economy depends on the answer to this question.

It turns out “Fuck off and sell your shares” is not an acceptable answer at scale. If I'd been a board member, I'd say to myself, “Okay, maybe I shouldn't have ambushed him in public. Maybe I should have said, ‘Hey, look, I just want to spend some time on this.’” You give your CEO the courtesy of not feeling ambushed. You maybe don't do it in a visible place.

If you're on the board of a company that's planning to spend a trillion dollars—I'm sorry, whatever it is, I can never keep up now—and you only have $12 billion in revenue, it's a totally appropriate board-level question to say, “How are we going to do this?” So I don't know anymore. I don't know if that's true. As silly as it sounds, it is beyond silly.

I think there's so much fear among VCs of getting out of step with the most successful founders. There's so much fear, and you guys are going to disagree with me, but I see it all across my portfolio. The better the company is doing, the more everyone's a grin-fucker. Just—never a critical word said.

Now you've poked the bear. I have a company that's going to [__], and it's going to [__]. Like, literally bust his balls. The board members will not intervene in any way to protect the shareholders because of the bad NPS that will come from damaging that founder relationship.

That, to me, is one of the most egregious escapes from fiduciary duty, and this is some of the most reputable investors. But you're agreeing with me, right?

I'm agreeing with you 100%. You've seen the same—

Same behavior, right?

[snorts] Yeah. Listen, OpenAI has had an interesting history. But if it were a normal history, a normal startup with the VCs we work with, and I had to come up with $1.2 trillion, I think everyone would be saying, “Sounds good, Sam. Sounds good, Sam. Keep going. Good month.”

Fortunately—and again, I don't know the man—but I've just been very impressed. I don't get the impression that Bret Taylor is a yes man, right? At one point, I'm not sure if he's still on the board. I should know. Larry Summers was on the board, and Larry Summers is many things, but not a yes man.

He may be off now, which would be a shame, because he would be worth being on a board just to hear him speak, right? I think he's a very smart man. But stepping back a level, there's a little part of me that's kind of sorry for Sam Altman because of where he's put himself.

Basically, every public analyst will say that he did what he had to do to raise this kind of money. But everyone covering the public markets will say the only thing between us and a 30% promote is the AI capex boom, right? And he is the poster child of the AI capex boom.

If the AI capex boom unravels and the world and media are looking for a villain to throw rocks at, it ain't going to be a big search. Genuine comment here: if this thing starts to slow down, it'll just get real hard, real fast.

I think if you're a board member in that, you actually owe it to your CEO to say, “Hey, dude, how are we thinking about this? Do we have good answers?” We can't just be glib. I hope that in the boardroom, Bret Taylor and those guys are sitting down and saying, “Okay, what are the cash flow numbers that say we can honor all our commitments? What are the cash flow numbers that say maybe we can't honor 2029 and 2030, but we can do the others? How are we going to do all this?”

When someone invents numbers at the $5 million or $10 million level and they're wrong, they disappear without a trace. When you invent numbers that are wildly overoptimistic at the trillion-dollar level and it unravels, you become the poster child in every economic history book for the next 200 years of the great AI crash of 2026. It's a pretty shitty place to be.

One of the things we've talked about in the past is that the job of the board, to your point, guys, is not to be a rabbit. It's actually to help the CEO avoid things. I often look back at some of these young founders, and we talked about this: yes, they were wrong, but the older board members who should have had more experience were, in one sense, less legally culpable, but morally culpable for not saying, “Hey, dude, are you thinking this through? Do we really have an audit, Sam Bankman? Do we really know where the money is? Do we really know where the customers are, Charlie whatever Javisser whatever her name is?”

You, as a board member, are meant to provide the guardrails. So I think when I look at this, I go, “Hmm, I hope someone's providing really good guardrails.” Because if it hits, it'll hit hard.

A lot of founders feel like VCs should just be their allies, period. Your job is to support me. That's your effing job. Some like the critical feedback, but I think it's far fewer today than you might think.

Your job—and if you're not supportive, you're a problem for me. You're making my job harder. I think that's how founders feel, and I think that's how Sam felt, is my guess.

The second thing that's interesting to watch, because I do think Sam's pretty transparent, is that it did show there is some stress around generating $1.1 trillion. I mean, not around raising it—he's committed to it—but around generating it. If there was no stress, it wouldn't have poked the bear, right?

I don't think it's actually a big deal. I do think they'll work around it and scale it back, but it did show there's some stress around hitting the $1.1 trillion. It's not even revenue, right? It's got to be gross profit at a 50% margin, so they've got to do $2.2 trillion or more to pay for it. They've got to do more than $1.1 trillion to pay off their capital commitments.

[laughter] Yeah. I can't believe I'm saying this. It's only cumulative, so it might be a little less, but yes, it's in the many hundreds of billions of dollars a year.

In a world where today only Amazon—I mean, there are a couple of companies. Google does $100 billion in a quarter, and so does Amazon. You've got to be at that scale. If you're doing $200 billion, $300 billion, or $400 billion a year, yeah, you're somebody. You can pay for these things. It's a real number.

You mentioned Amazon. Amazon crushed this quarter: plus 20%, AI shopping assistant Rufus, an additional $10 billion in sales. Things are looking up for them. They obviously have their ownership in Anthropic, which I think is at 7.5%. How do we feel about the state of Amazon today, where they sit considering last quarter?

The overall growth rate was, I think, 11% or something, but the real point is Amazon Web Services, their cloud business, grew at 20%. The story had been that they were the dominant provider pre-AI. They felt they'd slipped versus Google and Microsoft in the AI world. In fact, in terms of growth rate, they still have. Google and Microsoft were both in the mid-to-high 30s, but AWS kind of came back from, I want to say, 13% to 20%.

So the story was, “Oh, we're relevant too. We're not irrelevant in the land of AI.” The stock popped. I think the real takeaway from here is that this is the counterargument to all the cynical “It's a bubble” people. It would appear—and Microsoft said the same thing—that demand for these products is still exploding. If you have it, you can sell it.

At some point, maybe that won't be the case. I have my concerns, but right now the objective facts on the ground are that Microsoft was saying, “My biggest problem is I can't build data centers, so I'm capacity-constrained.” Amazon was saying, “My growth rate is up.”

Interestingly, they just signed a deal with OpenAI, because everybody signs a deal with OpenAI, to sell them more compute. So they clearly found some capacity for that. But overall, the story was continued high demand for compute, and if you're selling compute, stocks go up. That was the takeaway.

I don't have a sense of their consumer business and the Rufus shopping agent. I believe it's in limited beta. I haven't run into it. Have you?

The compute story was strong across the board. That was the aha. The people who want to articulate an “It's all going to go wrong” story right now still have to make that story prospectively and say it'll go wrong in the future. What you can't say is that it's going wrong now, because the demand was still there.

But to me, making a press release that says you're now the number 5 or number 6 partner to OpenAI for NVIDIA GPUs is not that impressive. It's just that you're behind Microsoft, you're behind Oracle, and you're behind Google. So it's interesting that they found some GPUs in a closet. I'm not speaking literally, but I don't actually think it means much to be late to the party to sell them.

It may mean a lot over time, don't get me wrong, but I think today there is a lot of AI performance theater. There's real, incredible growth, and there's growth in our retirement funds across the country, but there's still a lot of theater. There are a lot of folks doing a lot of work in AI who aren't seeing huge benefits.

This felt partly theatrical to me, but it's okay. I'm in favor of keeping the ball moving when you don't have all the answers.

That's very cynical, Harry. It doesn't rank them—I mean, it literally says they're like, “Okay, we called the CoreWeave guys. We called AWS. We called Microsoft, Google, and Oracle. We even called Benioff in case he had some GPUs or TPUs hanging around.”

[laughter] Yeah, no, I think the stock popped more because of the objective fact that the actual revenue growth grew than because of the OpenAI deal. But while you're right that it's not huge compared with the $250 billion for Microsoft, the $300 billion for Oracle, or the $400 billion for Broadcom, it is still better than not having one. Now at least we can say they don't have one. So, tick, done.

It is, but it's tough for AWS, having really created the category of cloud providers and being what we all grew up on. It's a major comedown not even to be above the fold on the leaderboard.

Totally correct. Look, you're exactly right. Five years ago, AWS dominated cloud compute. Now they don't, because cloud compute evolved from being simple compute to being AI-centric computing, and they didn't evolve fast enough with it. They let a whole bunch of people into their little oligopoly.

If they knew then what they know now, I think in 2019 they would have bought a lot more GPUs and been a lot more aggressive.

And, yeah, 20% growth is a lot, right? At this scale, absolutely. But Shopify, which is a competitor, right, on the other side of their house, blew it out: 32% revenue growth this quarter and 32% GMV growth, which is a close analog to Amazon. Shopify is reaccelerating at $8 billion of revenue, which is pretty crazy. Kudos to Amazon, but it’s also lucky that both of its core product lines are getting a general boost, right? There are a lot of tailwinds going on here.

Chad, I know where we went. Answer this one: Given what you just said there about them not being above the fold and losing market share in one of their core markets, which they used to own, and then considering the 20% pop, how do they sit in terms of being underpriced versus overpriced?

Look, at some level, this isn’t my strength, right? I don’t do a lot of public equity investments. Since you asked me, what I do think about as a B2B guy is that, for the moment, Google is underappreciated and Amazon is overappreciated.

Google was slow to AI. Google had to bring Sergey Brin out of jet-skiing and windsurfing in retirement to whip the troops into shape. But Google is really good now at all levels, right? It’s really good at consumer AI. It’s really good in search; search is back, and search is growing for Google again. The TPUs are good. It’s a good partner, and it’s the only one other than NVIDIA making real money.

It has the application layer. It has all the applications that we use. Amazon has none of the application layer, right? It has very little at the hardware layer. It has a niche search product, which is incredibly powerful but only used for e-commerce. In a way, it just doesn’t have as much going for it.

To be fair to Google, I mean, you used the word “underappreciated.” I think the stock has appreciated very nicely. The catastrophists at the start of the year were saying, “Oh my God, it’s awful,” and I think it’s up 53% from there.

I’ll admit it. In Q1, we talked a little bit about the companies, and I was positive on Google. Harry knows this. In retrospect, I thought about it, and it felt so contrarian that I even said, “Please don’t lead with that in the highlights,” because I felt a little stupid.

You fast-forward 2 quarters, and it’s turned out to be broadly correct that they have the key ingredients, a place to put them, and the ability to monetize them. Facebook is missing that last key element, right?

I still think, to your point, Chad, at some macro, zoomed-out level, if you had a monopoly in search and now you’ve gone to having ChatGPT as a de facto competitor, you’d still be better off if it hadn’t happened. You just can’t deny that it was a really good gig when you had this.

You’re not better off—you wish it hadn’t happened—but once it has happened, what you’ve got to give them credit for is getting almost everything done to be able to play in the new world. They had the key ingredients. It took them a while to put them together, but now they have all the boxes.

You’d still prefer not to have to do any of this and just optimize your 10 links from now until the end of human time, right, versus having to compete. But they’re competing well. In terms of playing in the new world, it was a bad week for Meta. I’m one big-ass Meta shareholder. Never argue with Zuck. Don’t doubt Zuck.

But wow: a $15 billion fine, and then the reaction to their commitment to capex moving into 2026 and the enormous spend that they continue to do and will continue to do—and the stock is down double digits. How do we think about how they fare in the new world?

I think it’s easy because the core business performed really well. Let’s start with that. The core business, I think, grew 20%. Despite the stock being down, it’s kicking off cash. It’s a great business, right?

The market is just—and we said this—entirely correctly saying, “Dude, you have this wonderful business, and then you’re taking the entire cash flow and building this AI stuff. Unlike Google, Microsoft, or Amazon, you don’t have an enterprise business to sell this shit to. And unlike ChatGPT, you don’t yet have an obvious AI-forward app that’s going to give you a lot more time and a lot more consumer engagement.”

So you’re basically spending $70 billion a year with no revenue attached. What the hell? Mr. Zuckerberg’s answer appears to be, “I think it’s relevant to be in this space. Thank you for your opinion. I refer you to the articles of incorporation. I control this company. Have a nice day,” right?

The market is just saying, “Hey, you’re doing the thing you did in 2021 and 2022, putting a whole bunch of capex into something that might not work,” and that’s why the shares are slightly down despite objectively pretty good performance. It’s entirely rational to say, “I don’t know why you’re making this spend,” and he’s saying, “That’s why I’m a founder.”

I personally don’t get it either, but I didn’t build Facebook. If he’s got a plan for it, we’ll see. It doesn’t appear to be the most efficient way to spend $70 billion, as judged from the amount of infighting, but we’ll see.

One that made me happy was Twilio. Twilio bounced 20% after beating expectations. The question is, when you look at that and you look at that generation of companies—your Dropboxes, your Boxes, maybe your underloved, underappreciated companies—are we going to see a series of bounce-backs on better-than-expected results, do we think?

If you undervalue, then you perform reasonably well. I mean, what happened with Twilio? Let’s have a Twilio: a $20 billion company doing $4 or $5 billion a year, valued at 4 or 5 times revenues. As you say, it had a decent quarter and bounced 15%, I think. Revenue growth was 15%, and it had a nice stock bounce.

It’s still in a bounded universe. The zoom-out point is this: these companies are in a much more bounded world now. They might grow 15%, they might grow 5%, they might trade at 6 times revenues, or they might trade at 4 times revenues.

If you look at the stock chart, the area of magic is gone, and now they’re just perfectly good $20 billion market-cap companies kicking off cash, with real strong cash flow performance. That’s what successful mature companies look like. They’ll be valued accordingly.

That’s the good news. Yet the bad news is that, as you look at the AI-first universe that exists, they’re not really playing in that 20-degree level of significance. In a world where you’re cash-flow positive and trading at 4 or 5 times, you can feel really good about yourself, and then you look up and realize, “Oh my God, Palantir is trading at 123 times revenues, and it’s cash-flow positive.”

They’re not in that ballpark. They’re not in that game. So that’s what a mature business with mid-level growth prospects looks like.

My take was a little bit different. Twilio goes from single-digit growth to 15%. That’s still significant reacceleration. We talked about Dev stepping down at MongoDB. He went from 13% growth 5 quarters ago to 24%—13 to 24%, right? That’s almost doubling your growth rate.

Here’s my point from both of these: now we’re heading into 2026, and you better have gotten a few nickels out of the AI expenditures, all the AI dollars. This is not new, guys. What is it, 60% of the growth of our GDP? You can’t miss AI.

Twilio said its voice-AI customers are a huge part of that reacceleration. It says voice AI is up 60%. We all know a million companies using voice AI, right? It said its top 10 voice-AI startups are up 10x.

The same AI is fueling the number of databases we use. Sometimes we use Supabase or Neon if we’re vibe coding, but a lot of folks use MongoDB. So, not to be the only guy in the board meeting that says something, but going into next year, you sure better have seen reacceleration because of AI, because there’s so much money there.

You got none of it, guys? You had 18 months to not launch a single feature, tap into a single trend, or reaccelerate. I’m not expecting you to go from 30% to 300% this quarter, but if you didn’t—if you’re not growing faster at the end of 2025 than at the start—as a founder, I give you an F-minus.

There’s so much money, and you don’t have to be Harvey to get a little piece of it. Twilio and MongoDB and Cloudflare and tons of folks have a little piece of that massive pie.

I think that’s totally fair, and that was well put, because I would almost amend what I said: you’re exactly right. They’re not the AI-native companies exploding at 2x year on year, but what you’re saying is correct.

There’s a big point spread between having no AI magic pixie dust and getting taken private for 3 times revenues, to be destroyed by your PE machine, and getting just enough AI pixie dust and lift to get that growth rate into the mid-20s-plus. We’re seeing some reacceleration, and we’ve just agreed, from the Navan comp, that you’re still only going to trade at 7 times revenues.

But it’s a damn sight better. 6 to 7 times revenues and some kind of forward story is a big point spread from 3 times revenues and being sold to PE. That’s the story we’re actually articulating to a lot of our private companies.

You can’t go from $100 million in revenue to just being Harvey. You’re not. But you better find a way to be relevant, and I like your expression: you can, Jason, in the age of AI, find a way to matter in this world and co-attach to the spend.

Even if it only takes you 10 basis points up, from 15 to 25, that’s a night-and-day difference in terms of your relevance and viability.

Frank Slootman

Actually, because I always like to come back from the public, where we have opinions but maybe it’s not our day job, to the deals we all work with, where it is our day job, I think you’re exactly right, Jason. That is the message for any of your companies that were pre-2021, that are pre-GPT companies. Maybe you can’t make yourself into the next Harvey or the next OpenAI, but by God, you better co-attach to that spend, because it’s the only game in town.

Jason Lemkin

I think you’re totally right, Frank. I think you should be preaching that to all your guys as we come into the end of the year and look at ’26. And look, sometimes it’s luck. The CEO of WorkOS, Michael Grinich, was posting—I think they went from, and I’m going to get these numbers wrong, something like $20 million to $40 million in 5 months.

He’s been working hard for years on WorkOS to be an OAuth and authentication layer, but all the AI companies used them. So, insane growth. We’ve all seen portfolio companies like that, and sometimes you have to make your own luck, but we’ve had 18 months. Tap into a little bit of the Harvey in there. You’ve just got to find it, and it’s at the edge of too late—not because there’s not time. I actually think there’s plenty of time for startups.

It’s because your team isn’t good enough. If you haven’t gotten a boost this year from AI, fire half your team right before the holidays. Give them a turkey and 3 months of severance, but they failed. Your team is not good enough. They had 18 months to ship a product that mattered in this world. Where’s your agent? Where’s your reacceleration? No more excuses after Thanksgiving.

Frank Slootman

But the irony is, it’s early.

At the same time, what we’re all learning is that we feel like it’s late, with all the things we’ve talked about—the Cursor, Replit, Lovable, and Sierra—but it’s actually so early in so many categories, right? It’s just 3 years since ChatGPT shipped.

Jason Lemkin

Yeah, but you got nothing. You did not reaccelerate this year? Fire yourself or half your team. Take your choice. Fire yourself or half your team at the end of this year. Don’t keep those folks who don’t have the answers around. You’re better off just not having them. You had time, right?

If we want to be critical—and I almost hesitate to say this, because we love these folks—HubSpot and Salesforce have to deliver in 2026 because they’re in play. They have the AI products. They built and shipped them, but they haven’t seen the bump that Datadog, Twilio, and MongoDB have. Maybe that’s okay; they’re not at the infrastructure layer. Maybe it takes longer.

But if I were Marc Benioff, I would fire half my team if I didn’t see real growth from that by the middle of next year. I’d just fire half of them. You’ve got the wrong people. There are 2,000 people building Agentforce, and we’ve deployed it. It’s pretty good. It’s quite good. It’s very competitive with any other agent you’re going to buy.

It’s time to monetize it in 2026. It works. It’s a good product. It’s not just smoke and mirrors; it’s really good. So, your team—maybe some of those folks have been hanging around for a decade—maybe they’re not the right people going forward.

Frank Slootman

I don’t know. Sometimes when Jason is cruel and harsh, I disagree. Then sometimes I listen and I go, “He’s absolutely right.” This is one of the louder ones. I might not say it as harshly, but I think you’re exactly right, Jason. If you’re not on this train now, you’re just not going to be relevant, and you will be sold for 3× revenues to a private equity firm that will smoosh you in with something else, never to be seen again.

You have your chance. It’s raining money in this space, and you need to navigate your product toward it. I think that’s probably pretty common advice.

As I think about our portfolio and all those kinds of companies we talk about a lot, that’s why the Navan thing was so good. All these companies that are doing $100 million to $200 million to $300 million, with sluggish 20%-plus-or-minus growth rates, need to figure out how to co-attach, or eventually you’re going to get tired, the VCs are going to get tired, your team’s going to get tired, and some are going to just not work out.

Jason Lemkin

The one thing I got a little bit wrong—I get a lot of things wrong, right?—is that it took me a while to see the data and believe that software companies could really capture dollars by replacing humans, for real. It’s not that I didn’t believe it. So much of it was VCs talking out of their ears, making stuff up, talking about how every human was going to be replaced with an agent when the software wasn’t very good. It just wasn’t good. It was a great dream.

When VCs say it, I believe it. But when most of the rest of the VCs say it, I think they’re just water-cooler talk. Now we’re really seeing it. Agents are better than mediocre humans, so you better be tapping into that revenue, too. They’re better than mediocre humans. Get going, guys. If the age of the copilot is behind us, how are you replacing humans? This is your job in B2B software: to genuinely replace humans.

I’m not a Harvey expert, but I do believe Harvey is partially replacing associates, right? That’s a lot of money there. Whatever Harvey is today, I guarantee you it’s going to be a better piece of software next year than it is at $8 billion, right? It will replace more mediocre associates who do terrible work, don’t want to do it, want to go home at 4:00 p.m., and don’t want to work on the IPO prospectus. It will just replace them. Find that money, and you can grow 50% faster.

Harry’s with me on this.

I am, but the trend I’m finding across shows is that Jason is becoming more and more right with his assertions. Have you seen this in the more recent episodes?

Frank Slootman

I agree with one comment here. I think the conviction that he brings from his use of the product is really valuable. I know all my colleagues who are engineers, and it’s just really great when you actually touch and use the product. If it’s an app, I try to use it. If it’s Cursor or something, my colleagues will be using it. I don’t want opinions from people who just saw it on PowerPoint.

When you use the product, it all becomes clear. He’s running a business where he’s literally had people, and now he has machines. That’s what this is all about, right? It doesn’t mean the humans go away entirely. It means they find other uses. It means they find other roles.

When you see it happen, you believe it, because if you want to be a good investor, I have this concept: you can tell when people are saying something they actually understand. It’s a very good habit to have. I’m listening to him, and sometimes I know the shit and I’m like, “Jason, you’re just talking out of your ass, and I don’t believe it.” When he’s talking about this shit, you can tell he has built these products, automated this process, transitioned out those employees, and is getting a better product.

There are areas where, right here, right now, you can replace a $40,000 worker with a $10,000-a-year agent and be better off, right? That’s why it’s not all just hype. I think the million-dollar question is how fast that diffuses into the whole economy, but there are places here and now where you should just be using AI. Jason is living that. He probably is right on the pointy edge, but he’s right.

If he’s the most pointy-edge Salesforce customer, in the next 2 years they have to get 20% of their customer base, maybe 30%, to be as pointy-edge as Jason in terms of automation. Otherwise, these guys are going to go somewhere else to get the automation. That’s what it’s going to take.

Jason Lemkin

This is what I’ve learned: if they don’t buy it from you, and you have a market position, they’re going to buy it from somebody else, right? That’s why I think Agentforce is so important. We have this thing on SaaStr.ai/agents where we share all the agents we use. It came out of nowhere. Now it gets around 12,000 views a month, and it’s not even highlighted, so that’s a lot of traction.

We’ve tracked it. We’ve sent millions in revenue to 2 vendors, Artisan and Qualified, that we use because we use them—millions in deals in a couple of months. We send it to Agentforce now, too, because we use it. If other folks are building these agents and you’re not building them, or they’re not on your platform, they’re going to find them somewhere else.

The demand in some of these categories is insatiable. Vendors, if you really can replace humans with software—not for pretend, not for San Francisco, but for real—and you have even a mini-brand, you’ll find you have more demand today than you can actually service. You can’t even onboard the number of customers that want to replace their sales team with AI.

There were blowups—the 11x situation and the blowups of last year—but Harry can talk to us about that. The demand is like something we’ve never seen. It’s insatiable to replace humans with software.

It’s all tied to this, but I was talking about the adoption of OpenEvidence. Open Evidence grew to $300,000 in 1 year, which is one-tenth the amount of time it took Doximity. I thought that was interesting, just in terms of market pull and adoption rate.

This goes to what we said last week, which was just everyone being in the market at the same time.

It is super interesting. You're exactly right. The fascinating thing is that OpenEvidence got there in 1 year, where it took Doximity 10 years. It speaks to this latent demand for AI. You just have to put in the negative, the nagging worry, which is the following fact: There aren't any more doctors as a result.

This is something Jason said last time: Everyone is in the market right now. You could have a world whereby you get every doctor on the platform in 3 years. And, yeah, that would have you saturate TAM more quickly, right? That's why, when you look at these hypergrowth rates and they're so compelling, you have to say to yourself: Once you get all the names, you have to be sure you have enough follow-on stories for those names. You don't want to be one and done.

Doximity is a really good company. It has a finite TAM associated with the number of doctors and the amount of advertising and stuff you can sell to them. OpenEvidence plays in roughly the same market, right? Probably, at its current privately held valuation, it will need to expand that market substantively to offer a return from here. Totally doable, but it's not just going to be getting there quickly and then stopping. Does that make sense?

Totally makes sense.

Because, as I said, TAM counts. One of the things we started to do with all these tools is that you should know, for each profession, how many people in that profession exist. How many lawyers? How many doctors? How many bankers? How many wealth advisors? That's your TAM, you know? It's really the number of people times the amount of their work you can automate. In the end, these are finite money piles.

Great adoption still speaks, and what it means is— that sounded a little more negative, but let's put the positive out there. OpenEvidence seized the moment. If you come along to Jason's point a year from now with a slightly better version of OpenEvidence 2.0, no one will care, because you missed the moment when 90% of the people who are ever going to adopt a tool like this are probably in market now. In the next 1 year, they're all going to make their initial decision pretty damn quickly.

You've shot up the S-curve super fast. If you've missed your moment, you've missed your moment.

Having said that, let's go to Mark Benioff's point, which I think was a good one from before when he was on the show. There are categories where you'll miss your moment, right? For sure, if you are a little slow, maybe find the areas that are slower. Mark's point was that only a couple percent of his customer base is fully ready for AI today.

If you are a little slow, maybe play to your strengths. Maybe go into retail or manufacturing, or areas where it's not that AI isn't there, but it hasn't changed overnight, right? Play to your strengths.

I agree, Jason. It's interesting, because OpenEvidence, just like ChatGPT, is an individual-adoption product. I think the velocity of individual adoption you see in Lovable is explosive. Corporate adoption is much slower. You're exactly right. I don't think everyone's going to buy Agentforce in a year. It might be 5 years, not 10 like SaaS. It might be half the time of SaaS, but it still could be 5 or 6 years.

I think the fascinating thing is that the consumer has shown they're going to adopt in a year, right? That's what you saw with OpenEvidence. It was interesting: If you looked at some of the general AI search and AI science tools, the number-one user was doctors using general research tools to look up very specific people and very specific medical questions, presumably meeting some patient with an odd disease. OpenEvidence caters directly to that need. It was a perfect product, and the adoption has been enormous and super quick.

That's what you're seeing with all the individual users. We can talk about the pace of adoption in corporate law with people like Harvey, but I can tell you every individual associate is legally or illegally using ChatGPT to help them format the masses of writing. Every time you do any kind of market research, you confirm that. Even if they don't have a corporate use case, they just have the laptop open and they're cranking along.

Okay, fantastic. It has never been harder to do Series A investing than today. Agree or disagree?

I'll tell you why I disagree. Ignore some of the stress around pricing and otherwise: This is the best of times to be a Series A investor, because there is just an explosion of seed AI startups. There are so many, and they can't all get funded. I know it's stressful and I know it's hard, but it is a gift that thousands of founders are in San Francisco raising seed funds from multiple accelerators. YC, Neo, South Park Commons—all these folks are creating hot, very smart, good candidates.

Yes, it's hard and you have to hunt and it's competitive, but the funnel is better. The top of funnel is the best it's ever been. Even if it is hard, this should be the best of times to be a Series A investor, because the funnel's the best. The layer above you on your funnel is the best.

I think it's the worst for seed, because everyone wants to be a seed investor even more each year. It's not just the Chainsmokers anymore, and Jared Leto—it's everybody playing at that hot seed startup level. That's Sasquatch to footballer. Everyone's in. So it should be a gift to be an A. It should be the best stage, stressors and prices aside.

The other thing that's actually helpful is that the direction of travel is now clear. If you think back to the last couple of years before ChatGPT, we were at the end of the SaaS era. It wasn't obvious what was going on. A lot of the deals you did then turned out to be evolutionary dead ends.

I do at least feel now, since ChatGPT, that the architectural direction in enterprise B2B for the next 10 years is pretty much obvious and a given. It's some form of agentic software—I hate that word, but it conveys a lot. It's some form of rearchitecting the enterprise stack to enable AI to do more of the work. That's the mission. The task has been assigned, and the only question now is which verticals first, which tasks first, and who will be an early adopter and who will be late. The direction of travel is pretty clear. That's the good news.

Unfortunately, I think the bad news is that there's just a lot of capital doing it. We're finding, from a world of competition, speed, and the need for speed of execution, that it's a lot. But that's life. It should never be easy to make a lot of money.

The most recent person I lost to was Andreessen Horowitz. Who was the most recent person you lost to?

I would say earlier this year: a very talented seed investor, Chamath Palihapitiya.

Perkins. Great name. Can't argue with it, you know, right?

So, yeah, I mean, I think the relevant point there is I would have to expand on that versus just having it be a litany of shame. Five years ago, I would have said those firms would be slightly earlier, and we would have run into a slightly different peer set.

But what's happened now is that we typically do revenue A's and B's, so it's kind of early product-market fit. Given the fund size, all the large firms that were typically seed and A are doing all those deals. The competition set has stiffened, right? You're up against tougher, better firms, and you've got to bring up your A game.

There's no doubt, if I was to list things that worry me, it would be that. In the face of that, you've got to do all the things you have to do. You've got to work on your relationships with the entrepreneur earlier. You've got to see the deal earlier. You've got to be more decisive. You've got to play to win when you want to, which means you have to know what you want to win.

I share with my LPs that the talent of the people we are competing against has gone up markedly. The good news about this is that we're fishing in the same ponds as some of the smartest investors on the planet. The bad news is that you're competing against some of the best investors on the planet. You just have to find a way to win.

Nice, man. Final one: I would rather be in Kalshi at $5 billion than Polymarket at $9 billion. Agree or disagree?

I actually think the real—my comment would be this: I wish I was one of them. And I don't mean that just glibly, because they're good, right? Consumers are typically not our focus, but I really like that space. I really like the idea of prediction markets. I think it's a very clever and good idea.

I think there are going to be a lot of issues around the sporting— I mean, the sporting-gambling side of that. You can get troubled by that. I know in the States we can be troubled by that. Coming from Europe, we bet on sports all the time. As you know, Paddy Power is an Irish company. There's a lot of great sports betting that goes on.

I do think the fascinating thing about Kalshi and Polymarket is the whole Brian Armstrong thing: You have these prediction markets, and then the person can tilt things, put their finger on the scales of who wins—in this case, Brian Armstrong—by using a certain set of phrases during his earnings call. He basically dictated that, I think, it was a Kalshi bet one way or the other.

There's going to be a lot of weird stuff that happens as a result of this. Some of these policy bets are situations where you just know an insider is making a trade: 1 hour before the administration announces something, you see it hit on Kalshi and Polymarket. So, there's a lot of fun stuff. But as a deal to be invested in, they would be fun. At some level, that's worth having.

So, Kalshi. Why? I mean, as a customer, yes. If it is accurate to say that Kalshi is fully U.S.-compliant in all 50 states today and Polymarket is still in an ambiguous position, even with Nasdaq's investment or Dice's investment—if that's true; I don't know if that's true. I think it's changed now. I think under the current administration, pretty much everything is legal.

Yeah, what I was going to say is, politics aside, there is a chance there will be a new administration that will be less sympathetic to this category. Based on my limited knowledge, I'm going Kalshi because I feel like it is a safer long-term bet than someone that is riding the current political vibes, which are all in favor of everything here. It was a lot different a couple of years ago, and it could be a lot different in a couple of years to come.

So, if I could minimize—if I could slightly de-risk the regulatory side, because it's CFTC-approved or D.C.-approved, I would take that bet just because I don't know who the heck's going to be president next. I don't know. I don't know. But the first act could be to undo everything that Sacks and his buddies have done. That could be January 1st, whatever.

I mean, these edicts, it's all going: crypto's out, Kalshi's out, Polymarket's out, everything's out. The next administration could say, “It's all gone.” Just to make a prediction, which is in keeping with the idea here, I predict that any re-regulation won't happen because of any new administration.

I think the real challenge to sports betting like this will actually be the leagues themselves wrestling with the fact that when you have sports betting, you have sports cheating. If it becomes endemic, like in some of the European soccer leagues, it'll be a problem. So, I actually think a fun problem for the next baseball commissioner, basketball commissioner, or NFL commissioner will be: what the hell do you do about this thing when you have these very particularized bets?

Not, “Will the Cowboys win by 7?” but, “In the third quarter, will the quarterback throw a second thing—a throw that misses?” The possibility for cheating just becomes high. So, I actually think that problem—the next administration will have plenty of other things to deal with. That particular problem will be the purview of, as I say, the sports folks. That's my gut. I hope you Americans don't watch any Pakistani cricket, because that'll really show you the way to do it.

But, guys, this has been a joy.

I don't spend a single second watching cricket, you know. It's just torture. But we respect that you guys love it, even though you're not good at it anymore. Well, do you know what, Roy? I would love to take you to Lord's. Come to London. We'll sit and watch a 5-day game, yeah? We'll watch it every day, and then it's going to be a draw at the end.

No, absolutely. Yes—a product that would not be built by them. The one thing you know about cricket: it was not designed by an American TV executive. Hopefully, mobile phones are collected outdoors, too. That's my hope. You have to put it in a basket so that nobody can be on their phones.

Silence in the stadium. Yeah.