20VC：Anthropic 的100亿美元融资轮 | 拆解 Klarna IPO | 走进 a16z 的72笔种子投资机器 | Martin Casado：共识投资是唯一玩法吗 | Satya 为什么在 SaaS 应用消失一事上大放厥词，嘉宾 Marc Benioff

And even when you look at other data clouds, like Snowflake, Databricks, or even Palantir Foundry, they're all at the $3 billion to $4 billion revenue level. They're in my sights, so I'm on it.

Soccer is a game played by 22 people. In the end, the Germans win. In the same way, venture is a game played by 6,000 people, and in the end, Sequoia wins. They won here again.

Amazon's AGI head said there are just 1,000 AI engineers that matter. Marc, I wanted to start with you on that one.

1. AGI Hype Meets Reality

AGI head—that sounds like an oxymoron. You're talking to somebody who's extremely suspect if anybody uses those initials, AGI. I think we've all been sold a lot of hypnosis around what's about to happen with AI, not that it couldn't happen one day. We've all seen those movies. Peter Schwartz, who wrote Minority Report and War Games, works for me. He's our chief futurist. But just realize that isn't the state of technology today.

What made you realize that? What was the penny dropping there?

I think when you look at large language models, which are kind of the state of the art of AI today, and prompt engineering, which came out of our Salesforce AI research team, large language models are 2 things. First, they are a finite set of algorithms, which have gotten a lot better—incrementally better, for sure—over the last 5 years. Second, they're a relatively finite set of data that has come off the internet. Those 2 things together really provide the state of the art of large language models today.

When we work with these LLMs, it's very cool because you're thinking, "Oh my gosh, it feels very intelligent." It kind of felt that way when I was using ELIZA when I was 16 years old on my TRS-80 Model 1, too. It was like, "Oh yeah, this kind of feels like a person."

It was pretty accurate.

Yeah, it was like, "Oh, this is like a person."

But it's not a person, and it's not intelligent, and it's not conscious. It doesn't have a childhood, it hasn't suffered, and it doesn't have compassion. It's not a being. I think there is some hypnosis around the state of the art of AI and what is currently possible or what is about to happen, and I'm extremely suspect around that.

I'm trying to bring people back to the reality of the current state of the art of AI, which is amazing, but let's actually use it for what it can be used for and also realize the major issues with it. I tweeted about this after I read 2 articles about doctors who are using AI. They're so over-reliant on an AI that's inaccurate that all of a sudden they're giving their patients bad advice and becoming intellectually lazy at the same time. I think that is a huge warning sign for all of us around AI.

Back to the AGI hat at Amazon.

If we separate the finite from the infinite, the thing that everyone feels is finite is talent. Zaki is paying up for talent like no one has seen before. You're seeing your peers get offers at $1 billion with, bluntly, very little to show for it—no disrespect to her—but other than the team. Do you feel the pressure to enter this talent-buying frenzy in the way that we're seeing other large incumbents?

2. Salesforce Builds An Agentic Enterprise

No, and we're not. We're very focused on defining what the next generation of the enterprise is. Tactics must dictate strategy over time in enterprise software.

The first thing that we've been talking about now, for only about 8 or 9 months, is that we have help.salesforce.com. Help.salesforce.com is our agentic layer around our support. This agentic service means that there is an Omni-Channel Supervisor paying attention between my human support agents and my digital agents.

To that point, I've been able to reduce the number of human agents I have in support from about 9,000 to about 5,000. That's important because I've been able to take that headcount and rebalance it into other parts of my company where I need more help and support, because we're still growing. It's a huge change in how our company is structured and how our technology is built and delivered to our customers. We're customer zero.

Let me give you one other crazy story to that point, and you'll be the first ones to hear this story. Over the last 26 years, Salesforce has had more than 100 million people contact us whom we've not been able to call back. They're just leads we've not been able to call back. We simply haven't had the people. That's all there is to it.

We have these people we call SDRs—sales development representatives—and we just don't have that many of them. We have about 15,000 salespeople, but we don't have that many SDRs. Well, we have agentic sales now, and not only are we doing support, but this agentic sales system is calling everyone back and having conversations, then deeply integrating through the Omni-Channel Supervisor into our new agentic sales product, which you're going to see at Dreamforce.

In your body language, you're saying you think you're gonna sell a lot of software powered by agentic AI in the next 1 to 5 years. Is that the summary message here?

Well, I don't think that there will be a piece of software that we sell—

Right.

That will not be agentic.

You're willing to say, just as it was never on-prem again in 2000, you're pretty much saying it's never non-AI agentic in 2025?

When you get to Dreamforce, you'll see that our promise that humans and agents will work together, it's not just in our Sales Cloud, it's not just in our Service Cloud, it's not just in Slack.

If we look at the impact of AI today on the business, it hasn't maybe led to the lift that one would think so far. Do you think that's fair? And how do you think that changes over the next year?

It's so untrue, and that's the funny thing. Number 1, our AI is part and parcel with our Data Cloud. Our Data Cloud, love it or hate it, is based on the idea that you need a data cloud that's federated to all of the data sources in your company—and why that is so important—is so that you can get all your data harmonized in one place, which is why we also bought Informatica, so that everything is together and now the AI can be more accurate.

Go to the front of my website and you'll see Agentforce now at the front of our website. It has done as many customer interactions as our support agent. Why is that? Because we put our whole website into our Data Cloud, and now people are just using this agent at the front of our website instead of clicking all the way through the website. It makes total sense, right? So that idea is really important.

3. Agentforce Clears One Billion

The Data Cloud and AI together now are more than $1 billion in revenue. We talked about that on our last earnings call. It's our fastest-growing cloud product ever in 26 years. We've talked about how we have thousands—I won't go through the exact numbers—of customers now on Agentforce, as well as the number of deployments and all of these pieces.

This is a product that a year ago we hadn't even announced. This is a product that wasn't even shipped until November of last year, and customers are still getting their heads around it. What software in the history of enterprise software has ever grown at that level of scale? I would cite to you—okay, Harry—none. I will say that this is incredible.

Now, you can talk about any other new company, whatever existing one; we can go through whatever it is, but this is a product that's breached $1 billion. And even when you look at other data clouds, like Snowflake or Databricks or even Palantir Foundry, they're all in the $3 billion to $4 billion revenue level. They're in my sights, so I'm on it.

I am like the guy in Star Wars, my favorite movie, in my TIE fighter: “Stay on target.” I see where I'm going, and data and AI are a huge focus of the entire company and our products, along with the fundamental aspect of humans and agents working together. That's how I look at that. Thank you for letting me address that directly.

4. Salesforce Takes On Palantir

When you're looking from the TIE fighter, what do you think of Palantir's growth? How do you think about it from the Salesforce perspective? We can all look at the numbers; the numbers are great, right?

Mm.

We can talk about defense and who knows who's spending these contracts, but how do you process that? Because it was growing 15% or something in 2013, right? It's crazy.

Oh, it's very cool and amazing, and very inspiring to me, that the idea that a data cloud, which is called Foundry, integrated with analytics, can be very exciting for a company.

So I will say that our Data Cloud plus a new agentic Tableau, plus Informatica, plus looking at a product like MuleSoft together is our data foundation. And with that idea, we need to have all of the government certifications, and they sell into parts of the market we don't sell into.

We really have reassessed: Where are we selling? Because the US federal government is already my largest customer, right? We run Veterans Administration, the GSA, and we just won a huge US Army contract. We beat Palantir.

But in some of the areas that they sell to, and some of the people that they sell to—and I won't go through all the details because it's not appropriate—we have not traditionally sold into those groups. So it got our attention that they're closing these deals, and their products are so expensive. Have you seen their price lists that are out there online?

They're good at getting the deals.

It's like, whoa, these prices. I'm like, whoa, my prices are too low. I'm actually automating the whole VA at this price? What would they be charging? I mean, my prices are low compared to theirs, and my products are much easier to use.

Yeah.

That's how I think about it.

Yeah. So no, he's not ignoring that $300 billion market cap.

How do you think about it?

No.

Yeah, well, that got my attention. I'm like—

Tell me.

How do I get that 100-times revenue multiple?

I don't know.

It's $4 billion in revenue, so let's keep it in perspective. It's an order of magnitude smaller than we are. But I just realized that it is a... As someone who was 4 billion in revenue once and is now 41 billion, it's two different companies.

Can I ask you one related question? I don't mean to go—

No.

Harry, you take the agenda. But related to Palantir, one of the things Palantir's gotten everyone's attention with is forward-deployed engineers. Do you think that's a new concept? Are they the same people at Salesforce deploying software for the last 20 years, and is it different? How do you think about this FDE concept?

Oh, what a great question. I think that it's both. I think at one level, we've always gone to the customer to try to solve their problem, listen to them, and do our best. We have a large sales organization and a large systems engineering organization, and we're out there talking to them, working, and building the prototype.

We also have professional services, and then we have partners, and all of us are in there. But we don't have that kind of branding of, “These are our four deployed engineers,” where now we're gonna start building your product before we've really signed a deal.

I think that idea is very cool, that all of a sudden you're in there kind of saying, “Yeah, we're gonna make a bet that we're gonna start doing business together, so we're gonna start building now.” I think that's something that we can all embrace and adopt and say, “Yeah, well, let's have more of that engineering resource start right at the beginning with the customer. Fantastic. Let's do that.”

Come back to Palantir in a minute, but just going back to the first comment, because the truth is, mathematically, Harry's right on the growth-showing-up thing. But I think it's the law of large numbers.

Look, when you're doing $40 billion, you said, Harry, it's “not showing up in the growth numbers.” When you're doing $40 billion, 10% is $4 billion, which is the entire revenue of Palantir. The problem with this poor guy is—

Thank you. I didn't miss that last part, Rory. I'm very hard of hearing in my left ear. Can you repeat it again?

No problem. I speak quickly with an Irish accent. My comment is, when you're doing $40 billion, Harry's giving you grief for growing at 10%. But I'm making the point that when you're doing $40 billion, 10% growth is adding $4 billion, which is an entire Palantir every year.

So the comment on growth, which is mathematically true—Harry, you're right—the 10% growth is what these guys are now; you're just dealing with scale. And I think it speaks to one interesting thing, which is you added 9 figures of revenue on the AI deals in the last quarter. I mean, it's a $400 million AI-only startup, which would be fricking amazing if we all owned it, right?

You're just up against the law of scale here, which speaks to—even if AI is amazing—what I liked about where you started, being grounded. Even if AI is amazing, I think some of these people who think it's gonna transform $100 billion market caps in a week are just way overestimating what it takes.

Can I ask another one? When we think about MCP, and we think about how it changes how we engage with different products, do you think we'll want to log into SaaS apps in the future, or will we just want our data inside of ChatGPT and Cord-

Harry's just gonna keep coming at you.

Try and ignore those.

5. SaaS Apps Survive The Agentic Shift

No, I think this is—I’ll open my heart here and just say—I think this is one of the greatest disservices that has been done to our whole industry and to all CIOs and all CEOs of software companies in the last 12 months: certain executives, who will not be named, have said that SaaS apps are just gonna be CRUD databases, and CRUD means create, read, update, delete.

And it's like, really? Do you really think that? Because if you really think that, wow, you are really wrong, and that is crazy talk. That is not how it works, and I don't know what software we're talking about, or what applications, or if you use computers anymore, or if you use a phone. But right now, in the current world—the world that I'm in here in 2025—I'm just saying that I need apps and I need agents, and I need them to work together.

And yes, if you can make my job easier and better through AI, then give it to me. But to say that all of a sudden all of those apps are no longer relevant and that humans don't need apps—that's what we just said, humans don't need apps—that's not true for any of us on this call, and it's not true for anyone on Planet Earth. That is why I think it was a huge disservice to the industry and got everyone anxious, because certain people—they'll remain nameless—have a lot of credibility because they are great people, actually, and great executives.

Oh, my goodness.

But to say these things is nonsensical. Why Microsoft has 3% CRM market share? Because of nonsense.

But I do want to disaggregate 2 things, because it's worth it. One, it's going to be a CRUD app and we're going to vibe-code it. Take it apart. We agree. No one's going to build a big, sophisticated app vibe-coding. Let's just discard that discussion entirely.

I think the interesting question is, how much of the real estate on top of Salesforce do you guys own? How much do you allow other people to own? As Jason said, there's a bunch of startups; we've all funded one, God forgive us. Assuming Salesforce is a given, you are the infrastructure—not this bullshit comment that you're going to be replaced. Ignore that entirely, but assume Salesforce is the infrastructure.

Maybe the sales rep in their daily toil can have a better tool than Salesforce to do some of the work, or maybe even an agent that's not owned by Salesforce can be doing the work and coordinating with Salesforce on the back end. To me, that world is much more realistic. Do you want all that front-facing real estate on top of the Salesforce data? Do you allow other people in the ecosystem? How do you make those choices?

There is going to be a level of application functionality that is going to be required, and there's no question that these apps that our users are on today are still going to be very much a part of how they get their work done, and that they operate in the flow of work—in sales and service and marketing, and all the examples.

Then, at the third level, there is going to be an agentic layer that's going to interoperate with those applications and that data. Yes, there will also be an ecosystem that is going to fuel all of these things as well, and the connectivity is going to happen and it's going to be open.

You look at, like, the Slack ecosystem or the Salesforce AppExchange. The agentic layer is a huge investment opportunity for the whole SaaS ecosystem, and I hope that it's going to be built on Salesforce.

Now we have several agents that give daily updates in Slack.

I need a demo of everything you're doing, because the first time we were talking, you're like—

Tell me.

“Yeah, I have this agent. It's with me on the sales calls, listening. It's coaching me.” That was very inspiring to me, and now you're like, “And I have a dozen agents.”

There is going to be a radical explosion of small and medium-sized businesses like yours, because entrepreneurs like you can do more than ever. So while the enterprises are trying to figure out whether they're going to DIY it and whether they're going to do this or that, look at you and all the entrepreneurs like you who can boom, boom, boom, go right into the future. We're going to see an order of magnitude more SMBs, because SMBs can do more than ever.

They will. They will.

Marc, every week we hear from Jason that SDRs are screwed, that if you're 23 to 35, à la poubelle—in European terms, to the trash. You don't have a future. You have said before in this conversation, “Human and agent,” and very much suggested a pairing between the 2. Jason has presented an idea that in the next 12 to 24 months, actually, we'll see this mass exodus of the SDR class. Do you think Jason's wrong?

6. Salesforce Redeploys Its SDRs

Well, like I said, I think that we have all these leads that we've just—

Totally.

We have systematically not called them back, and now we are. That gives me the ability now to rebalance my headcount and to really say, “Hey, I want to take all these folks and make them sales folks.”

I think that in all of the segments of the business that we do business in—not just government, which was 1 segment, and not just the enterprise, the high-end enterprise, the 5,000-plus world—but the mid-market and the small business, we're a company that's going after all of those segments. We don't leave any segment behind.

So you're saying that the SDRs will remain, and it will just allow you to cater to the ones that you couldn't cater to before? Because, Jason—just to be annoying and British, but it's mid-Atlantic time—no, I think Mark might be saying, Mark said his support team went from 8,000 to 3,000 and he redeployed them into other areas, I think about that number. I think the same thing happened with the SDR team.

9,000 to 5,000. That's exactly right.

Yeah, I think we'll redeploy. I don't know how many entry-level SDRs Salesforce has, but I bet you redeploy 70% of that headcount into enterprise reps or forward-deployed engineers. That headcount just becomes more valued with agents for sales. I bet you don't need 70% of them.

What you're saying is so important, Jason, because what you're saying is that the fundamental architecture of an enterprise software company in the future is not exactly as it was in the past. The fundamental architecture of the company will be different.

All of us grew up in SaaS and applications and all this over the last 25 years, and so we saw how the applications have changed and evolved. But now what we're saying is, it's not just that; it's also the companies as well, and that is different.

So, Harry, did you get the answer to your question?

Well, yeah. Redeployment—

I think—

Yeah.

No, I don't think it's bullshit, though. I mean, it's because we have this discussion every week, Marc. Jason is basically the grim reaper and thinks not a single 25-year-old will ever work in this town again. And, you know—

Well, I think it's grossly overly optimistic to think that you can redeploy 25-year-olds who aren't that passionate, don't have that many skills, and are entry-level SDRs.

Oh, we're off on this one again.

No, but when you're at Salesforce's scale, it's about headcount. Marc's budget is fixed. He's got 80,000 heads on a spreadsheet. That's—well, I don't know, when I was at Adobe, it was 20,000, right? If you can move those heads up the value chain, Salesforce can be a much more efficient company.

Yeah. Yes. It's a more optimistic—

That is exactly right, Jason.

—view.

It's a more optimistic view, A, than Jason's taken in the past, which is why he's contradicting himself, but it's a good view. I actually notice time and time again Marc's—I won't say spin, but approach on it. When Jason did his thing about how he only has 3 people in his company, Marc's take on that was, there will be lots more entrepreneurs because of that.

It's super additive, which is the only way you're going to sell this AI revolution. Otherwise, there will be another fricking revolution if we keep pushing on this. I like the upside-related focus. As we've discussed over and over again, if they're not any damn good, they're on their own, but it's at least a vaguely upside-y approach, Harry, versus Armageddon here.

You can go to our website and see who we're hiring, and this narrative that we're not going to hire any more kids out of college—this is also bullshit.

Marc, I'm aware that you're going to have to run.

Oh.

I do want to ask 1 final thing, which is just, in terms of unfair questions, Rory loves me for this. You have—

You're such a dick about this.

No, I'm not. You just always comment. You have OpenAI at 300, and you have—

Oh.

Anthropic at 170. Which would you prefer to buy?

Well, I think both are actually great companies. Salesforce owns 1% of Anthropic, so I'll just—it's obviously a great company, very focused on the enterprise. OpenAI also is a great company. I'm a big fan of their leadership and what they've done.

I don't know what you're paying your media-training person, but you should pay them more. That was a master class in how to handle Harry being annoying. Basically, Harry, thank you for your question. I've complimented everyone. I love it. He just won. You should just fold, Harry.

No, I'll be practicing that next time. Be nice about everyone and shut up, Harry. Good job.

Marc, you are a hero.

You really are.

Thank you so much for this, and thank you for putting up with my pressing questions. Jason's free to present at Dreamforce about how he's changed Salesforce.

I'll be there. I'm even going to Metallica this time.

Marc, thank you so much. You're a star.

All right. Thanks, guys. Great to see you.

Take care.

Bye-bye now.

Thank you. Bye. All right, are we ready? Now I'm excited because I want to dive into this. Nat Friedman reporting to Alexandr Wang after not a huge amount of time. How did we analyze and interpret this news of the new structure that's come to be in Meta's AI division?

7. Meta Reorganizes Its AI Empire

I thought the consensus when we talked about this deal at least 10 days ago—14 days ago—was that it was fine to give up billions of potential carry and funds to be in the game, to be a player rather than to be on the sidelines.

I don't want to be critical, but, man, then you're essentially getting demoted in a reorg. Maybe it doesn't feel that way, but a hiring freeze and a total reorg within 30 days—it’s a lot to process. I might rather be running my own fund.

At least it seems like a vaguely sensible org structure where you have 1 person in charge and then the 4 divisions. You have pure science, LLM foundation models, AI applications, which I think is where Nat is running, and infrastructure. You read the org structure and go, “Yeah, that’s probably how you should run it,” and you’ve got 1 guy in charge.

It’s the same thing—stupid example—when you get these soccer teams where they just have so much money, you hire all these people in on the transfer market, and then you’ve got a bunch of drama, and then someone’s got to be the manager and figure out who’s going to play what position. I don’t know what promises were made. I don’t know who’s bent out of shape, but it seemed like a sensible thing to do. You spent $20 billion on talent. You now need to tell them what position to play and who’s going to play forward, who’s going to play striker, and who’s going to play fullback.

I just don’t get it. I feel naive here. I don’t understand why you’d do it if you’re Nat. I understand you want to be in the room, I get that, but then you’re reporting to someone else who’s not Zuck. For anyone who knew Nat and knew Microsoft, he was really in the grooming position to be the next CEO of Microsoft, as many understood.

And now it’s like, to then report to someone who’s not Zuck in this structure. You’ve got Yann LeCun also reporting to Alexandr Wang as well. Daniel Gross is reportedly not really there day to day. I’m just confused by the whole structure, and it just feels like, wow, you gave up on probably one of the best funds.

I’ve got to push. Just be logical, Harry. You’re not confused about the structure. You’re confused about why he’d do it.

Yeah.

You’re confused about why someone who was highly autonomous would sign up to report to someone who reports to the CEO.

Yeah.

That’s what you’re confused about.

And the rationale around that, for me, would be: guess what? Elon goes to Zuck when he wants to buy OpenAI and Sam Altman. It’s pretty cool being in that room, which Nat would be with Alexandr Wang to have that discussion, and you’re not if you’re just another fund. That would be the reason why you’d do it.

I think being in the room for that a couple of times is fine, and then I’d rather run my own shop.

You know—

There are only so many rooms I need to be in. It’s pretty fun. It’s like the first IPO you’re a part of. It’s great, but I’m not sure what it’s like as a VC to have 20 IPOs. I might rather have more carry than show up to ring the bell. I don’t know.

My honest comment is, I totally get why someone who’s a great operator would choose not to be a VC, because I think if you are a good operator, I always tell the great operators who talk about coming into venture, “Don’t be crazy. Your highest and best use is operating.” If you had the ability to be the next CEO of Microsoft or be a VC, my strong advice is: go be the next CEO of Microsoft.

I get the transition from venture to operator. The question you’re raising is the level at which you make the transition. It’s giving up autonomy, but again, as I say, I don’t know what was promised.

How did you think about Meta more broadly being hit hard? I mean, they were down 6%. They’ve had a pretty meteoric, continuous rise. This was a blip.

The big picture here is their core business is doing extraordinarily well. They have a very tenuous link between their core business and their AI initiative. They talk about how AI is optimizing their core business, but even from the discussions, I think they said that’s much more old-school AI than any of the LLM stuff. So you’ve got this core business that’s kicking off cash, and then you’ve got the CEO with untrammeled power deciding to invest all this cash in this new business.

So if you’re trying to value the stock, your entire day is spent thinking, “WTF is this new business worth, and is it going to eat all the cash flow?” It’s like Kremlinology. When you’re looking at who lines up in Red Square and trying to figure out who’s in charge, you just look at this announcement and say, “I don’t know what this means, but maybe it means something bad, so maybe I should sell the stock off.” There’s just no data, and there’s no way of knowing.

At some point, someone’s going to have to explain what they’re doing with this $60 billion or $70 billion and how it’s going to change their business. If Zuckerberg is right, like he was with Instagram and WhatsApp, everyone will go, “Yay.” And if he’s wrong, like he was about the metaverse, everyone will go, “Oh my God, what were we thinking?”

So you don’t think this is the beginning of a cooling of the excitement of the AI market, a dampening of market caps, and a dampening of public markets in a way that some people are worried about?

How the hell would I know? I mean, I don’t think that implies I’m a know-it-all. It’s just not knowable. Let me tell you: you’ll know when it’s happened because it’ll hurt.

All you know now is that things are pretty lofty. When things are trading at 15 times earnings, you don’t have to agonize all that much because, you know, if earnings blip 10%, the stock blips 5%, and no one cares. When things are trading at a very pricey level, then everything that goes wrong, no matter how tiny, gets magnified through the stock price.

Things are trading at a high price now. You don’t know whether that’s going to change in a week, a month, or a year. It’s going to be an angsty time, and either the growth comes to fill the earnings gap or the stocks go down to reflect that. When that happens, who the hell knows?

Listen, I’m not an expert, but Meta has a 1.59 beta. It’s a volatile stock. So I don’t think we can read anything into these ups and downs because the beta is so high. I mean, Nvidia is 2.3.

Oh.

These are insane numbers, right? So abstract away from that. When you look at the amount of volatility Figma’s had since the IPO, it hasn’t even had a quarter go out. These high-beta stocks, I don’t know.

Totally.

You’ve got to be smarter than me to figure out what even a 7% or 8% movement means. The beta’s too high.

Aligned with what I just said, which is the cooling or the lack of cooling, Anthropic goes from a $5 billion to a $10 billion raise. Is demand just completely inexhaustible for this? I heard it was 4× oversubscribed. How did you guys react to the move from a $5 billion to a $10 billion raise and the reportedly 4× oversubscribed round?

8. Anthropic Raises Ten Billion

Good for them. Demand appears to be pretty damn high. It looks like you can raise $10 billion-plus in a single financing in the private markets. Yeah, OpenAI was $40 billion. As you say, appetite for the AI story is extraordinarily strong, and most of the public comps aren’t a pure AI story. They’ve got AI blended into something else.

You have Facebook, Google, and Microsoft, which have at least something. Apple has nothing there. Amazon has little there. So there’s got to be huge demand. If you’re a Fidelity-type manager, you’re like, “How do I get me some AI action?” There are 2 obvious at-scale candidates.

And yeah, you probably can sell a lot of that stock right now, and they’re going to sell it. The good news is they know what to do with the money. They can buy GPUs.

But is this then, Harry—you would know this better than me. Maybe Rory knows it. Iconiq is a lead for this round, and Lightspeed led the last round?

Yeah.

I mean, maybe the underlying LPs and money are from sovereign wealth funds or others. These are the standard cast of characters who can tap into vast amounts of money and charge and keep a vast amount of economics on top of it. Of course, they’re going to go from $5 billion to $10 billion. If I can deploy, why wouldn’t I deploy another $5 billion if I’m Lightspeed or Iconiq? Why wouldn’t you?

Instead of Lightspeed putting $2 billion in, if its LPs will give them $6 billion, why not? At a GP level, it’s the same amount of risk, isn’t it? If I lose $2 billion or $6 billion, what’s the difference? But I mean, I can make so much more money.

And they could also be right in that call that it’s going to work from here. It’s an interesting exercise to try and take the Anthropic numbers and say, “What do you have to believe to believe in a 3× from here?” It’s frankly not impossible. A lot has to go right, but a lot is going right.

I kind of did the thought experiment a while back. The growth rate over the last year or 2 is so fast that 1 of 2 unprecedented things is going to happen in the next year. Either A, it decelerates at quite a normal rate relative to its current growth rate. It’s going to hit $50 billion in revenue plus, because things that go from $1 billion to $9 billion or $10 billion probably grow next year. I don’t know. That’s a 10× growth. Do they go 5×? Do they go 3×?

Yeah, it could end next year at $40 billion in revenue.

Exactly.

It’s possible. If it ends this year at $9 billion—

Yeah.

—from $1 billion to $9 billion.

So—

You’re much better than me, Rory. What if you just do your trailing velocity? What does that end up—what’s the—

That’s exactly it. You end up with an enormous number, and you go, “Wow, that’s not crazy.”

Enormous.

And then either that happens—which would be unprecedented because the amount of revenue would just be so big—or they slow down faster than anything has ever slowed down. If you go from 10x growth to 2x growth, and once there's 2x growth, it's amazing at that scale, but it would be such a deceleration.

So when you look at the stock and the price they're paying, as I say, it's not crazy to say that if the growth only slows even 50%, it's still got a kind of trajectory and growth path to tens of billions of dollars in revenue, and that gets you into the valuation. So, processing through that, you say to yourself, "Hmm, at some point it's a market-size question."

If there's enough revenue out there, these 2 guys are going to get it. And thus, in the end, as highly priced stocks do when they're really leaning into growth, it boils down to your assessment of whether there's $50 billion of demand for foundation model APIs or $500 billion of demand for foundation model APIs. If it's the latter, they're probably going to get 40% of it, and it gets them $200 million. If it's the former, they're going to get $20 million, and a lot of these people are going to be sad.

What do you think it is, Rory?

Because it's a really hard question. I think it slows down more than people are...

I mean, it's something Jason said 3 or 4 shows ago where, if you start running out the numbers on what... I mean, Marc—let's talk about $100 billion of revenue. Salesforce is doing $40 billion. So at $100 billion, you're saying it's kind of 2.5 times the size of Salesforce, which effectively has dominant market share in the CRM space.

Coders have to get what Jason said a couple of weeks ago. These agents have to be worth $10,000 or $20,000 apiece for that market size to get to that scale. If all it is is $2,000 an engineer, I don't know if you get there, right? That was my big aha when I did the math.

You actually need these things to take vast chunks out of the labor budget and be worth $20,000, $30,000, $40,000—almost apiece—to the enterprise for the math to work. And Jason said, in some cases, it will. There will be some use cases where an enterprise will part with $20,000, but there'll be lots where it won't.

So you can tell my lack of certainty here. I don't know if I get to that $100 billion-plus in revenue because I just run the math and I can't find the TAM, but I could be wrong in underestimating it. My guess is no, and it slows more than you think, but it's not a crazy call.

I can't shoot from the hip and do the math, right? Because it's so much money. It is so much money.

Yeah.

I mean, listen, we just had Marc Benioff here, who's saying at Dreamforce they're going to launch an AI SDR that I guarantee you is going to take 6 to 9 months to scale up, but it's going to be bonkers. Everyone's going to turn it on. That will tap into a vast amount of budget, a vast amount of cycles, a vast amount...

Now, maybe some of it will be their own LLMs, but it doesn't really matter for purposes of this. I mean, we're just starting this cycle, right? And it's hard to predict how much human replacement and how many new applications—

But let's do that exercise, though.

Yeah.

Listen, Salesforce is doing $40 billion a year. I think $12 billion of that is Sales Cloud. Let's say they turn this on and it's a 30% uplift. An AI SDR on top of the core Sales Cloud, which is $12 billion, would be $3.6 billion of extra revenue, which, as you point out, only gives the poor man another year of 10% growth.

But $3.6 billion in revenue—let's just say LLM costs as a percentage of revenue are expensive, at 20%. So that's $720 million. You've just had the second-largest software company on the planet turn on the most labor-saving device for its core marquee product, and when it filters down to LLM revenue at $720 million, round it up to $1 billion. That's when you kind of go, you have to sell a lot of labor replacement to get to $100 billion.

You do.

Now, maybe I'm underestimating. Maybe the 30%—

You—

...percent is wrong. Could you see yourself, Jason, paying 4 times what you pay for Salesforce for an AI SDR on top of that Salesforce?

Listen, we're a tiny group, right? But we have 4 seats of Salesforce. So what do we pay—$300 a month? $1,200 a year for Salesforce? Nominally, we're paying $500,000 for 11 AI agents. So what's the ratio? I don't know whether that makes sense long term. I don't know if it scales.

If a portion of that ratio were to hold—

You do.

...then it's a pretty cheap round.

Yeah.

But it's a crazy ratio, isn't it?

It's crazy. Because let's just even do 2 to 1. Let's just say for every dollar you spend on Salesforce, you spend another dollar on top. That's $12 billion. Let's assume 20% goes to the LLM; that's $2.4 billion. It's real money, but it's only $2.4 billion. I just—

Yeah, but I'm spending $500,000 versus $20,000. That's more than 20x more, right?

Agreed.

To be clear, if 20x is the ratio, then you're right. Then it is a cheap round.

I think the problem—and Harry was teasing at this, but we have to go gently with the CEO of a $40 billion run-rate company—is that you're doing the right thing, Rory. Salesforce may not capture that incremental $120 billion. That's the challenge. Workday may not capture it. Palantir appears to be capturing it. That was why Marc was impressed with them.

If the big guys mostly don't seem to be capturing this agent dollar, if they do, great. But today, when we're recording this—

I—

...it hasn't happened yet, right? They're not capturing much.

And I'm saying, even if they do—and I think they will, I think they're well poised to capture some of it—I think, as I say, when you apply the 20% ratio and you get back down to how much revenue it is for the LLM, you struggle to add it all up.

And then I'm going to make the argument against myself: you look at the explosion in revenue in the last year, and I've never seen something grow 10x, 9x from $1 billion in 1 year. I always joke that Newton's law of motion applies to companies: things in motion stay in motion.

I can never remember anything going from $1 billion—even $1 million to $9 million and then flattening out to $12 million, let alone $1 billion to $9 billion. Just the trajectory alone implies $30-something billion the following year, which would be a significant slowdown. You'd have gone from a 9x year to a 3x year.

I'm just worried that the Magnificent 7 today have so much concentration of value in the public markets, driven by AI hype and excitement. But it's very valid, as we see with Anthropic's revenue growth, like you're talking about there. I don't feel like we've ever had the concentration of value tied to AI in 7 companies as we have today.

And I am looking at it now going, "Ooh, I really hope there's not a blip here. Dear Lord."

So basically, you've done all your analysis just like everyone else, and then the last sentence says it all: "I don't have the stomach to sell, crystallize my gains, and move it all to value stocks. Instead, I'm just going to let it ride and pray a little." Nice, Harry. I'm not going to argue with it. It's what I'm doing too, but—

Yeah, that is exactly. And I ask you—

Where the rubber hits the road is when you do that analysis and you have to say to yourself, "It's unprecedented. Do you want to make a trade? Do you want to sell down? Do you believe that it's going to revert to the norm?" And—

And you don't now?

I do. I do believe it's going to revert to the norm. I'm more pessimistic than some, right? I do believe it's going to revert to the norm.

So you're crystallizing your gains now?

I'm actually looking at it right now. In fact, I had a long conversation with someone about, given all the other dynamics, what's the best ETF for core commodities, which are the only things that survived the '70s. But it's a 5% play, not a... I mean, I'm not going to go down that rabbit hole.

But I think something can be amazing and still overpriced. That's perhaps the sentiment. So I'm looking at this going, all these companies and these opportunities are amazing. I don't want to be down on them because any growth from here will be just astonishing.

I didn't plan to come here and talk about stock prices, but you asked about the Magnificent 7. Eventually, you get reversion to the mean, and we're at the highest point we've ever been in terms of concentration.

I mean, it hit us hard in 2022, right? Reverting to the mean hit everyone hard. Hit everyone hard, right? 2023 was worse, but we've already half forgotten the precipitous drop in 2022. I mean, not everyone has, but it was brutal.

And 2022 was even worse because the revenue growth was still there. The cloud companies were still growing at a decent percentage compared with 2021, but the valuations fell 66%. It was brutal.

9. SaaS Reacceleration Returns

Okay. So we have this realization. We understand that actually good times sometimes end. And then when you look at MongoDB, up 27% today on amazing numbers, we have Box up. We have Okta up.

Jason, can you turn up the volume where the party's going? I'm ready to put on my DJ set here.

I need a little time to process it, but thank God, because you were just asking Marc Benioff why they weren't getting the lift from AI. I'm glad to see that just literally this week, we are seeing Mongo, even Okta, which had been struggling, and Box. The other day, Zoom, which is not exactly a rocket ship anywhere, saw growth reaccelerate because of AI. It's like, thank God, the cavalry's coming just in time to help.

The public guys need it. So I think it's heartening, but to your point, this is not Anthropic growth. It is reacceleration. Reacceleration at scale, to Rory's point, is always epic. We owe everyone a hell of a lot of kudos when they reaccelerate at scale because it's so rare. It's just not like Palantir reacceleration.

Exactly right, Jason. And maybe, actually, the thing the two things have in common is just a reminder that changes in stock prices happen when you get a difference between the expectation of what actually happens. What you're seeing in some of these Mongo bounces is that when people have the SaaS-is-dead story, and the markets buy into it, these things start trading at, you know, 5.5 times revenue.

Suddenly, it's not like you grow 9%, but you beat expectations by a couple of percentage points, and suddenly you can get a nice bounce in your stock because you're trading at a value where, once the upside shifts, the stock's only going one way. It's almost the mirror opposite of what happens to these super-high-priced things when all the good news is priced in, when even one piece of good news goes out of the deal, you fall fast. If Salesforce had a 13% Q/Q GAAP revenue quarter, you would see that stock bounce like you haven't seen it. It would be, "Ooh, we priced in 10, and we're suddenly getting 13. We're getting 13 at scale. Oh my God."

I'm not a total Mongo expert, but when you look at companies like Lovable and Replit, which we over-discuss, right? Let's call it $300 million or $400 million of ARR already this year, plus everybody else, right? Every time someone's using the app, they're spinning up multiple Neon or Supabase databases. The load on both of them is massive. They've never seen demand like this. It's massive.

That's great for them. I mean, Neon got bought by Databricks for $1 billion, right? I didn't even understand why at the time. Now I get it, right? Supabase is probably worth much more, right? It's kind of a bummer, in air quotes, if Mongo doesn't benefit from that. If it's all the Harveys and the Supabases and, like... I guess it's good for VC, but it's also a terrible stability point if none of the incumbents benefit, right?

Where is Atlassian benefiting from this AI wave? Where is Monday benefiting? So it's heartening at a meta level to see Mongo benefiting from AI deployments. It's heartening because it means maybe the revenue is a little more durable. Maybe Barry's Replit investment—Lovable—will go 10X rather than crash and burn next year because all this stuff's enduring. Right now, it still feels so fragile, doesn't it? All this revenue feels fragile.

No, it feels very durable. Thank you very much.

It does? Oh, well, did you go from—

Very, very durable.

Did you see what Wix said? What did they buy? Base44? What's it called?

Yes.

The one they bought?

Yeah, for $80 million.

Yeah. Now they did $1.2 million last week.

Probably a good deal, then.

Oh my God, deal of the century. I tried it. I actually took my site and had it rebuild it. It looks like Claude, but not as good. I get it. But they're working out all the issues.

It's just interesting: if Wix can buy an 8-person startup and then achieve that revenue velocity, it's impressive, but it also makes you think about durability.

It totally does. Rory, you said if Salesforce grew 13%, not 10%, it would bounce like never before. We had Klarna file to go public in a $13 billion to $15 billion range. It was lower than people thought, and it was lower, I think, largely because of the 20% year-on-year growth, which isn't great. It's good, but it's not great.

Jason, how did you interpret Klarna finally going out? We know that it had a $45 billion priced round before, SoftBank-led, then a repricing to $6.5 billion, and now going public at $13 billion to $15 billion.

If they were growing 24% last year and now they're filing and they're growing 20%, what's the inverse of the Mendoza Line, Rory? The opposite. When you fall below, you can't file.

I could be wrong. It might be that Klarna's filing just in time in the midst of this IPO wave, because 24% to 20% is not the reacceleration that we're seeing in some of the folks. Even Netskope just filed with modest reacceleration, from 30% to 33%. Thirty percent to 33% may sound modest, but it's a lot of work. Twenty-four percent to 20%, you know—

It—

Mav's hitting the hard deck again. You better pull up. Pull up. Pull up.

Yeah, but I think—

File, file, file. Pull up, pull up.

There is a level of growth below which it's hard to file. But just to be clear, the bigger you are, the lower that growth threshold. It's simply that there's a transaction level below which the Wall Street math doesn't work.

For example, if you're doing $10 billion in revenue, they'll happily take you public with a 7% growth rate because you're just big enough to matter, right? But you're right: for the typical venture deal, somewhere around 20%, you're starting to get to where the multiples don't get there. But clearly, at $14 billion, it's a perfectly doable deal. So I don't think it's a question of you're going too low to matter. I think at $14 billion, it's a valid transaction.

I mean, it's just a very different business from Netskope. It's very much a financial business. Interesting, they did that little bit of overstatement on AI and they're going to automate everything, and then backed off on that. But that was interesting but not important. I think the more interesting fact was some of the early comments on lending losses earlier—I think it was earlier this year.

So it's a financial services business, and it lives and dies on financial services metrics. And yet, once you start to lend, you gotta be good at lending. I mean, we talked about Nubank last week, which appears to be bloody good at lending, right? Klarna will be just fine. It'll trade, it'll go public, whatever. It'll be valued like a relatively mature financial services business.

I haven't studied the S-1 yet, but the guys who priced it at $6 billion were right, which is Sequoia, and the guys who priced it at $45 billion were wrong, which is SoftBank. You'll recognize this, Harry. There's a saying: Gary Lineker used to say that soccer is a game played by 22 people, and in the end, the Germans win. In the same way, venture is a game played by 6,000 people, and in the end, Sequoia wins. They won here again.

Still, 24% to 20% growth at less than $4 billion in revenue is still incredible, right? But deceleration.

Yeah.

They're hyping, as they're going public, their $1 million in revenue per employee. That's implicitly saying, "We're finding our Rule of 40 in the bottom line, not in the top line," isn't it? I mean, that coded message of $1 million per employee as growth decelerates—

Again—

—is fairly clear to Wall Street, right? We're mature.

Yeah, we're mature, but also you're a financial services company. It's not the same metric. I shudder to think what Jane Street or Citadel's revenue per employee is; one day, it's in the tens of millions, right? This is the classic fintech company trying to make software noises.

But let me give you a clue: you're a fintech company. It's all fine. It's a totally worthy thing. You're a fintech company at huge scale. Well done. You built the category. You'll get the medium-growth fintech valuation, and everyone but the last round will make money.

Does SoftBank just get washed?

First of all, no. It boils down to the details in the document. So, 2 comments. One is, at the time of the last round, I remember thinking, people are going, "Oh my God, you paid $45 billion and now you're raising money at $6 billion." Yeah, you look like an idiot, but you only took 10% dilution.

If you were an investor at $45 billion, yeah, it sucks to take that dilution, but a down round at $6 billion didn't kill the economic value of their investment. In fact, it preserved it by keeping the company alive. Now fast-forward to today. You still overpaid. As we've discussed before, it boils down to what's in the docs.

My guess is they don't have a block because Sequoia are not dumb people and wouldn't have left it in. So, yeah, they just overpaid, and they're going to get converted, and they're going to trade at 30 or 40 cents of what they originally paid. Their hope is just that it bounces up from there, so they don't get washed.

They just do what's called losing money. It turns out when you buy a stock at $45 and it trades at $15, you're down.

I totally get you. I love that also in terms of the 6,000 players and, in the end, Sequoia win. That's the intro for sure. And they're gonna pay you for that one, Rory.

I know the head of marketing there. She's gonna be like, “Rory, go. Woo.” It's amazing.

I started here 31 years ago, and they were doing great. You fast-forward 31 years, and they're still doing great. There's something in that. You gotta hand it to them.

I remember thinking when the Klarna round went down—and obviously there was a bunch of drama after that with Sequoia that we'll just leave out for now—that it was a shrewd call. You just let them raise money at $45 billion a year and a half ago, and now you're stepping in at $6 billion. I remember thinking, “Good investment,” and it's gonna turn out to be that.

Not that it helped them. SoftBank did have a ratchet in WeWork. This deal was not that far off at a similar valuation. They could have a ratchet here. I need one more day to find out, right?

Yeah.

You don't think so?

Yeah. I don't think—

I mean, if they got one—

I wouldn't have been in—

Another deal at about the same price, at about the same time—

It's no—

At least it was discussed.

No, it's—

At least it was discussed.

You're exactly right. It is knowable.

Yeah.

And when I get off here, we will—

We'll find out.

Feed the S-1 into ChatGPT, and we'll know in an hour.

Yeah. What will Netskope go out at? $700 million ARR, growing 33%. Last valuation was—

It was around $7.4 billion in 2021, and after that they raised some kind of weird convert that's harder to track. It's a good company. It's not making money like Figma. It's losing money, but it'll be at or close to it, my guess—or maybe even up from it.

The 2021 round can exhale there. It's not quite out of the woods yet, but if you've got a company at $700 million growing north of 30% with a little bit of reacceleration, it doesn't take more than a squint to see a $7 billion flat round to 2021 as being doable.

Yeah.

And good for them. Great company. They've been around since 2012. Congratulations to Lightspeed, who own a big chunk of this, along with Accel. I'm not going to tell you where it's gonna trade day one, because as we've proven with Figma, that's not knowable.

What we were right about on Figma was the step-up in the process. They'll file at $5 billion or $6 billion. They'll get demand. They'll walk it up. My gut would be $7 billion, $8 billion-ish. Where it trades on the first day, who the hell knows?

Do you think it could be a bounce like Figma?

The answer, of course, is no, because I believe, as I said earlier, in reversion to the mean. Figma had the largest bounce of any large-cap IPO since, I think, 2000, so I sincerely doubt they'll copy that.

It was funny, actually. I got an email from one of the many millions of bankers. You know those marketing emails they all send out the next day saying, “We priced XYZ IPO.” The headline was, “We successfully priced the Figma IPO.” I just so wanted to email back and say, “You priced it, but it's not clear you priced it right, my friend.” “Successfully” might be a reach here.

10. Venture Embraces Consensus Bets

Going to the other end of the spectrum, guys. I don't know if you saw this, but it was astonishing for me. It was a mapping of seed rounds segmented between megafunds and boutique funds.

Yeah.

The number-one megafund seed investor was Andreessen, with 72 seed deals.

Yeah.

Compared to number two, which was 27—Sequoia.

Which was Sequoia.

Exactly. Rory, how did you analyze that? Is Andreessen just playing a totally different game?

You have to say they're playing a different game. The words “ipso facto”—the words speak for themselves. If everybody else is doing 27 or less and you're doing 72, then by definition it's a different game.

We saw it again in the other interesting analysis that someone did on the Series A rounds. They are the successful quantity provider at every stage in the thing. They're the largest capital raiser, I think, other than Insight, but Insight is obviously slightly more later-stage. In their pure Silicon Valley universe, they're the largest capital raiser at every stage, so by definition they're doing the most deals and being the most aggressive.

Do you think it will work out? When you look at some of them, and we've mentioned the Databrickses of the world and how much that will return.

The truth is this: if it does or it doesn't work out, it won't be because of their seed program. And that's the big aha. The seed program could get lost in the noise.

It will work out if, by virtue of their seed and Series A program, they get the small number of absolute outliers. They stated this right back in 2009, so give them credit for wild, wild consistency. As long as they get that small number of companies that are absolutely outrageous upside performers, and they stuff $1 billion into them like they did at Databricks, and they do it at the right price, it'll work out fine. Everything else is a loss leader.

The seed program is basically like cheap milk in the supermarket. It brings in the crowds, right? It's the loss leader.

Seed is for suckers, apparently, Rory.

No, no. We said that. Jason said that last time.

Yes.

I think it's—

Yes.

I think it's consistent.

I've got a friend who's—

I think it's consistent.

I've got a friend who's a complete dipshit, and he's gonna make a huge amount of money from a $100 million SPV into OpenAI at $200 million.

Well, he may be a dipshit, but he's got good sales skills because he got in. There are different ways to win in this business, and sales is part of it.

There you go.

Yeah. Sometimes you just gotta sit on their steps, just sit outside of OpenAI's office all day long.

Mm.

Grab Sam Altman 11 times—the classic Sequoia playbook. Sit on your steps until you get the meeting. Don't leave without the term sheet.

Now, guys, do we have any other news items before I do a Tweet of the Week, where I just want to talk about one tweet I thought was particularly interesting, that grabbed the zeitgeist, and I want to hear your thoughts on it?

What's the tweet? You mean an X?

Martin Casado: “The idea that non-consensus investing is where the alpha is is actually quite dangerous in the early stage. Follow-on capital tends to be more and more consensus-aligned.”

I thought it was a better tweet than he got credit for in the Twitterverse. I saw that tweet, and he also did a really good piece on gross margins and the way people are misunderstanding that, which, if we had more time, we'd talk about.

I thought that tweet wasn't crazy. People then cited the cons. And yes, there are always outliers that are not consensus. In 2016, the non-consensus bet would've been to do OpenAI, true, but it's also probably true that 90% of non-consensus bets would've failed entirely. At that stage, SaaS was probably consensus, and only about 50% of SaaS bets would have failed entirely.

When you're on this megatrend of an architectural replatforming, a goodly amount of the correct investments to do are fairly consensus in terms of the broad macro themes. I remember, I think it was IVP years ago—20 years ago—they had this concept of 70% of the bets being very much on track: faster, better, cheaper. Then I remember 30% being brave-new-world bets.

I don't think you could build your entire business on waiting for OpenAI. It's like the explore thing. You are betting on the megatrend that's probably gonna last 20 years. It could be AI. Twenty years ago it was SaaS. Fifteen years ago it was public cloud. That's a consensus bet that paid off for 15 years.

I'm rambling a little, but I think his comment was more correct than the 140- or 280-character comments made it out to be. You don't wanna just be consensus, but consensus is a bad word for being on point with where the industry is going.

My reaction—he responded to my reaction, too—was that I thought one of the implicit points, and we've talked about this the entire series of the show, has been putting money into consensus bets, right? Half this AI stuff is.

Today, 10 deals are consuming 40% of venture capital. Everyone we knew who used to do B2B deals only does AI. My point back, which he agreed with, was that if you're gonna do bets outside of that, you better not count on much follow-on capital.

Yep.

Because they're not interested. They're not interested. I've done several B2B-plus-AI deals in the last 18 months that I love, that will do great, and the advice I give to all those founders is, “Don't expect any money.”

Yeah.

Eighty percent of the folks I can refer you to are not gonna take your meeting, and it's a reality. I don't know—I mean, he was like, “That's exactly part of the issue,” right? So there may be several layers, but if the whole industry is consensus, the capital's concentrated.

It's not just your buddy that put 100 million in the SPV. Everything's concentrated here, right?

We had an IC today for a fintech business, and they scaled to 5 million ARR in a year, and the founder—

Yeah.

—was great, and I said, “Guys, why is this not moving fast? What's wrong with it?”

Yeah.

One of my team was like, “Oh, it's not AI.” Yeah.

And that's an example where I think, Jason, you were spot on. It's not that you shouldn't do non-consensus bets. There are a couple of different things, but that's a classic example where you should do it. You should buy it at the right price because you're not going to get the magic pixie dust next round, and you should run it capital efficiently because you're not going to get people throwing 4 billion dollars at you.

So, Rory, what you're saying is the price should reflect that it's not AI.

It will and should be valued on fundamentals.

But that's different from what it was in the last years.

Yeah. If it's non-consensus only because it's doing something different, then by all means do it, provided you understand what's different and you understand what you're getting into. I think the really true thing, for example, that didn't quite come out is that we talk about this when we think about our megatrends.

It's one thing to say, “I'm going to do a deal that's not the ultimate consensus bet, AI,” but you really have to question if you're doing something that effectively is a bet against the megatrend. Knowing what the consensus is has quite a lot of value because it also speaks to where the industry as a whole is going technically. Let's call it the technical consensus, as distinct from the financial valuation consensus.

Going back to what Benioff said, the technical consensus is that most software is going to be agentic for the next 20 years. Do you really want to take a bet against that? Because that's probably where the industry is going, and that's, again, where I think Martin was right.

But valuation aside, I think the bigger issue for venture is that, when times are good, we take follow-on capital for granted. No one's worried about the follow-on round for Anthropic that they're throwing 10 billion in. There's not a single investor that's worried about the next round, is there? Greed. It's just greed. Okay?

But most of our careers, we've worried about follow-on capital. I worried as a founder. Capital in B2B was scarce until as late as 2018. It was very scarce. So that's just—

Agreed.

Doing Harry's bet might be great, but not burning a million bucks a month. Then it's like—

When I—

—who the hell's going to? Harry's fund isn't big enough. He doesn't have billions yet. Yet. And he doesn't like to carry his investments through 3 or 4 rounds. So you have to pass on that one unless the burn rate's zero. Then I would do it.

Because I'm having this experience right now, when I look back at my mistakes in the last 3 or 4 years in terms of investing, I have actually made both kinds. I wish I'd made more consensus bets, because “consensus” is such a negative word. I wish I made more on-trend AI bets. We made a lot. I wish we'd made more, because the megatrend was bigger and more dominant.

But equally, I have 3 or 4 utterly non-consensus deals that I looked at, was intrigued by, should have pulled the trigger on, and regret not doing. I just saw one of them today where I'm like, “Wow, I really missed that one.”

But I'll say it: what you don't remember is the 90 non-consensus bets that you didn't do that just haven't worked out. Both statements are true. It's a lot more forgiving in the consensus marketplace because, as you say, you get buoyed up by other people's capital, right? And it's easier in the short term to, A, survive long enough to get the feedback.

Two of my best investments today required me to create a round out of nothing when there was no capital. I had to create a round. I didn't have enough money. I had to create it. I don't want to do that too many times.

It's true.

This isn't as hard as creating Snowflake from scratch, man, but it's hard. Okay? It's hard.

Going back to the consensus comment, maybe it's okay to do the consensus bet, but you don't want to do the consensus bet where the odds on the consensus are lower than the accuracy of the consensus. In other words, you want to be in AI because that's what—because we've wrestled with a lot of these “consensus AI” bets, and we're not doing them, and we can't make the prices work.

You still have to assess the risk accurately, and all the consensus statement says is it's more likely than not that this is the direction the technology is moving. Therefore, you probably don't have that “Oh my God, are you totally wrong?” dimension to your business, which is why you can lean in a little into this AI consensus bet versus some of the others.

But you still have to get all the other shit right, to your point. On top of that, if you overpay beyond the dreams of man, then there's nothing you can do to save yourself. So, everything in investing ends up being way more nuanced than consensus versus non-consensus. The consensus-bet risk is that you're probably right on direction. You might ludicrously overpay.

What—

With the non-consensus bet, you could be way wrong about whether it's even going to work. You probably won't have any follow-on capital. But if you get it right, you will have a beautiful thing. You'll have a high-ownership, low-capital, N-of-1 outcome.

Again, as always, it turns out investing is hard, and you can't just paint the numbers and collect 100 million bucks.

Final one. What consensus shit do you wish you'd done more of, Rory?

I think I underestimated the impact of, A, the scaling laws in AI, and B, the ability of primarily Altman and some other folks to inspire belief in those scaling laws and unlock 600 billion of CapEx spending a year.

Anything that was attached to that AI trend has just had a wall of money for the last 5 years. It includes the foundation models. It includes Nvidia and the public markets. It includes the inference companies.

My mental model is that we have 600 billion being spent making AI, and right now we have 28 or so—whatever it is—in the recent survey of apps using AI, most of which are OpenAI and Anthropic. I did not think that we would be able to find 600 billion a year to spend in this space.

And if you knew that was going to happen, I think you'd have looked at the inference companies. I think you'd have looked at the model companies at prices you thought were super high. I think you'd have broken glass on your financial model to try and get some of what is now the scaling-law consensus.

So I suppose you could argue that, at the time, it wasn't consensus, which may be the actual counterargument as I process it in real time. But that's the trend that you almost could not have had too much exposure to in the last year.

Boys. Jason, anything to add, my man?

No. We can edit in Marc's AI and be tougher on him if you'd like. We can build one together. We'll build this clone for him, and we'll be tougher. Sorry if we weren't tough enough.

Tell me, Marc, why are you so brilliant? How were you so prescient to think about this agentic change?

Okay, listen, let me be clear.

I think Rory was a suck-up. I don't think I was. I think you're going to look back at mine and you're going to say I had some pretty good stuff. I honestly think this. I think Rory was a suck-up, but he doesn't know Marc. I barely know him, but he doesn't know him, so he—

And Rory was a little tough on the growth. He was just nice about it, but—

You know, I think you're going to like me better.

I think you're going to like me better.

Do you know what I find so funny, guys? It's like, “Who the fuck am I? I'm a kid from London,” who's done a podcast and stuff.

Marc, you've got to try harder, young sir.

Yeah, exactly.

Most of the companies I've advised at 41 billion in revenue have committed a little earlier to the AI trends, Marc.

I'm embracing your advice, Harry. We're not trying to make people feel—you want your guests to come back. And actually, I'm going to say it again: I actually thought he was more on point and balanced than the other AI gurus who are saying it's AGI. He was just like, “We're going to sell some of this shit to our customers. And they're going to buy it and it'll be good.”

Please provide the podcast transcript you’d like corrected.