Anthropic融资450亿美元却仍受算力约束；Thoma Bravo将Medallia交还债权人

Rory O’Driscoll

It’s entirely plausible that 10 super-big exits cover the entire nut from the LP perspective, such that it’s still a good business.

So, what’s on the agenda this week? $45 billion poured into Anthropic from the hyperscalers. Next, China blocks Meta’s $2 billion acquisition of Manus. And then, finally, Thoma Bravo hands over the keys to Medallia to creditors, with $5.1 billion of equity wiped out. What is the future of this stage of private equity?

Guest 2

It’s a whole new world where I think AI is even more competitive again.

Rory O’Driscoll

The dirty little secret of venture capital is how much of your money you make in that 1 year in 10 when everybody buys the dream.

Guest 2

More and more, the agent is going to choose which models and which vendors we use.

Ready to go? Boys, we are back. It is another week of Harry asks questions, Rory continuously puts them down as being terribly phrased and useless, and then Jason provides the actual wisdom and value.

But Rory’s turning on me. No, Rory, I love you, dude. Sorry. Feeling spicy. I just came from an LP meeting.

1. OpenAI Misses Growth Targets: Is This a Real Problem?

I wanted to start with OpenAI missing numbers, specifically across user growth and revenues. With the 2 obvious misses on numbers, it’s led to CoreWeave dropping and Oracle dropping, I think, 5% and 7%, respectively. Is this being made too big a deal of, or is this justified in terms of the response that we’re seeing?

Rory O’Driscoll

It feels a little overdone and a little late. In the sense that it accurately reflects what happened last year, if you zoom out and think of the 2 big-picture jobs here, you have 2 jobs when you run one: you have to build great models, and you have to buy enough compute to be able to run them, right?

There’s no doubt that, in the back half of last year, OpenAI failed at the first part of that job. They didn’t build great models and, as a result, their traction relative to Anthropic declined markedly. Their market share declined markedly, right? And that’s probably the shoe that’s dropping now.

If you look at the model that they shipped—I think it’s GPT-5.5—the reviews of the coding say it’s pretty damn good and arguably better than the current Anthropic model. So, to some extent, this is a late-dropping shoe on facts that were probably knowable 3 or 4 months ago if you were paying attention to the traction.

The funny thing is, in the super-connected Twitter AI universe, Anthropic is the one getting slammed right now. There’s a whole bunch of, “Oh, Claude can’t keep up, can’t support the users, and the current Codex model is better.” So, this feels a little like old news that maybe is news to The Wall Street Journal, but probably isn’t news to anyone paying more attention.

This is looking backwards, right? This is looking backwards through a lens into last year, and it confirms what we knew. Anthropic, obviously, the rate of growth was incredible, and some of that was market share. It wasn’t all like that. Some of it was market share, and it stole market share.

Elon was clear about this: Anthropic had, quote, “something special” in coding, which underestimates how much of the overall growth in the market coding represented. I think OpenAI acknowledged it by doubling down and getting Codex better.

Guest 2

I have just 2 thoughts. One, as crazy as it sounds, I think this is also yesterday’s war.

But I think going forward, more and more, the agent is going to choose which models and vendors we use. Do we use Canva, or do we use native AI-based generation tools like FAL? The agents are going to choose which LLM we use, okay?

Just like everyone from Dario down has said, there are going to be more and more agents doing coding. The agents are going to make the decision on everything.

As a consumer of LLMs—forget about coding, which is number 1 by dollars, right?—as workflows expand to do everything and agents do more, they will pick the LLM. I see no competitive advantage to Claude for most workflows.

OpenAI, whether it’s Codex 5.5 or the state-of-the-art models, is so good for my workflows that I think the advantage that humans get out of Claude and Claude Code—which is huge, right?—is ephemeral.

This was a story last year: humans shipping code, shipping products. We got an advantage. We got more superpowers from Claude and Claude Code. I’m not sure agents are going to get the same advantages. They may get just as many advantages from OpenAI.

I already see that with our agents—our AI VP of Marketing, AI VP of Customer Success, our applications. They love OpenAI. They love it.

So, I think this is another benefit that is ephemeral as agents take over more and more of the workflows of our lives. We’re going to look back at last year as a transition era, when most workflows were managed by humans, and in late 2026 into 2027, most workflows are going to be managed by AI agents.

Not AI agents working autonomously, not crazy OpenClaws blowing up our Mac minis, but running everything. And I think this is where OpenAI is very well positioned. The agents will pick what they want, and it’s not about what makes humans better.

Our agents like OpenAI, and that’s one of the many reasons I’ve come back to Team Sam and Team OpenAI. It’s not because I care; it’s because my agents like OpenAI. They like the API. They love it.

So, I have to follow my agents. Just like you had to back your team of humans in the old days, like 2024, today I have to back my team of agents. If they pick OpenAI, I’m on the team.

I’m not exaggerating. It’s a radical change that most folks who are still in the human-led AI world aren’t seeing yet.

We’re off piste already, but I’m going with it. I’m going to paraphrase, and then I’ll have 2 questions.

What you’re basically saying is, in a world where agents pick the models, you don’t have this human anchoring bias for your favorite agent, and thus it becomes more of an “every day’s a new day” kind of market, right?

Guest 2

Yeah. They have very different perspectives on what vendors to pick, and presumably a better perspective.

The interesting thing about that is, what does that mean for the large AI companies? If the choice is between OpenAI, Claude, and Gemini, then it’s still a nice, cozy little oligopoly. To get in the game where you can be chosen by an agent, do you still think it’s just the state-of-the-art foundation models that are going to be relevant here?

Guest 2

Well, listen, I think it’ll change. Over the weekend, my 996 project was that I built an agentic API grader. I had Claude, OpenAI, and Gemini take the top 120 APIs and grade which ones they thought were the best—tools from ElevenLabs on down.

Interestingly, Stripe got the highest grade. It got the only A+, which is a reason to go along with Stripe. I did not think Stripe would come out on top.

2. The Rise of AI Agents: Why Humans No Longer Pick Models

My Captain Obvious learning is that you’ll see the same thing if you just ask Claude what to use: it’s very biased toward the leaders. They’re very biased toward momentum. They’re not going to recommend Marketo for your agent to do marketing automation.

In fact, it mocked Marketo, Outreach, and Salesloft as tools useless to agents. It said there is no place in the age of agents for those products. An agent will never send an email through Outreach, Salesloft, or Marketo because it will just craft and send a better email itself and say these are worthless products in the age of agents.

But if you had to do a 2-by-2, they want market leaders that are innovative. That’s who the agents pick.

So, I think for now, a 3-way is Gemini, OpenAI, and Anthropic. In fact, the order is Anthropic, OpenAI, and then Gemini. The grader graded Anthropic just above OpenAI, and Gemini was just down here.

That’s what all of them wanted to pick, and I think that’s the world we’re going into. The old guard are going to be bypassed or useless.

So, to your point, I think this is an interesting story, but it’s a whole new story as the agents pick. It’s a whole new world where I think OpenAI is even more competitive again.

So, let’s go with that. My mental model remains at the 3-way oligopoly, just like cloud is a 3-way oligopoly with Google Cloud, Amazon, and Azure, right?

What you’re saying here is—which is fine, got it—the other question that I’d be curious to get your thoughts on is this: when OpenAI just announced that agent product, it seems to me that, if I were running one of the foundation-model companies, and if Jason’s world is the world you agree is going to happen, then you just make damn sure that you build the agent harness such that the device picking the agents is your device, or the agent picking the models is your device.

Guest 2

Under-discussed. The public markets have the right idea, but the wrong direction.

The public markets think vibe coding and Claude are their threat. No, the threat is what the agents pick.

Actually, if you look at it overall, the markets almost get it right. They’re worried about Atlassian and Monday because agents don’t need project-management tools. They have no use for them.

And the ones that are actually outperforming—the Twilios, the Cloudflares, and others—the agents still have use for them.

So, our whole narrative that public markets somehow saw the future that most podcasts couldn't see is that what matters is what the agents will pick. And to your point, this is why the agent wars—I mean, Marc Benioff gets it even more. This is why Sam Altman—they're all like, "You've got to win the agent wars," because if OpenAI wins the agent wars, then you have lock-in.

Then OpenAI will probably pick OpenAI as the API. Now, maybe they will evolve so that they're actually agnostic at some level, right? Where these agents are so successful that they have to pick the best of breed. One could imagine it. But you've got to own the agentic layer—not just the fabric; you've got to own the agents too, because they're going to make these decisions.

Do we place no value, then, on large, multi-year enterprise deals? A la ServiceNow, a la—we had Mike from Atlassian on, who talked about the increased rate of very large, multi-year enterprise deals. Do they just not have value because we're going to see the rapid acceleration?

Guest 2

Decay. Churn that is deferred still exists, and it is where the rent-a-CEO and the mediocre hide. Workday does 3-year contracts up front and 5-year renewals. So, the average Workday customer effectively signs up for an 8-year contract. Three and five is their standard term, okay?

Does that mean they're going to stay on Workday forever? No. It means they have 8 years to find better agentic solutions. Now, maybe the executives are all gone by the time that comes up. Rory's better at this than me, but if you believe that public stock prices are the sum of terminal values of cash flows and profits, then deferring churn or masking churn doesn't matter.

It doesn't help if you defer it 4 years, because if it dies—if the customer dies anyway at the end—you never have it, because it falls off your ARR rolls. I'm not sure it's extreme in the sense that I can envisage a world where, even 8 years from now, you don't churn off your SaaS system of record, but you're not growing.

Interestingly—and again, I didn't expect to be here—

Hang on, Jason. I'm going to agree with you on something. We're going to discuss ServiceNow at some point in time, right? It grew 20%, plus or minus, with a very negative market reaction. If you listen to the analyst call, this will make you very happy, Jason.

A lot of the really, really grindy questions were: "Is your AI agent revenue really real? Are you just bundling it? Is it growing fast enough?" In other words, basically saying, "I buy into the analogy, Jason, if you articulate it," which is: if all you are is a system of record for humans, you are a bounded cash flow, even if you're not a negative NPV. I think some companies will be—we'll talk about that later—but even if you're not a negative NPV, you're slow growth at best, with an NPV terminal value.

The only way to get the high price that you need to make the stocks compelling is to have agent-based activity on your platform. It was just super interesting that we had that talk a few weeks ago where you gave me clarity on that: you need to see agent acceleration. Then it was funny to look at the call. This is a company—I shouldn't know the numbers—doing $16 billion to $20 billion, and they're grinding the CEO about half a billion to $1 billion worth of agent revenue, because what they've recognized is that's the tell for the future.

I'm willing to bet that in a quarter or 2, someone's going to be asking Benioff, "How many calls to your agent, your headless API, did you get? How do you measure that? How do you measure value?" This is the way it's going.

3. The Collapse of Private Equity Exit Routes in VC

I think Canva's going to have a wildly successful IPO, and they just launched their agentic suite, okay? It's got a lot of great agentic products in it. You can vibe images, you can vibe—you can vibe everything. It's actually very, very good. This Canva 2, I think it's called Canva 2.0, is great.

Is it the best? It's definitely better than Make. No, I mean, it's great. But ask yourself a question: would an AI agent use it? No. An AI agent is not going to go in and move assets around, buy them in a—it's just going to create the assets. So, an agent doesn't need Canva.

This is the meta-threat that the stock prices reflect, but the narrative misses, right? Canva got the right 2025 product in 2026, but will agents buy it? I don't think there's any chance an agent is going to use Canva. I don't think there's any chance an agent is going to use Jira or Confluence unless it's forced to. It has no need for these products.

Can you help me out? I love Cliff. He's been a guest on the show. He's a really good friend of mine. If he's going out in 2028—which I think is a realistic timeline for when he would want to go out—and you just said he will have a successful IPO, but agents would never use it.

Guest 2

Yeah. I don't know when it crosses over at the low end, between consumer and enterprise. I actually think this is one area where the enterprise crosses over ahead of the consumer, because we want to automate these workflows as soon as we can, right?

I don't know the answer. I don't know if the average low-end B2C user who gets so much value from Canva is going to make themselves obsolete with an agent. They're still going to be designing. They're going to pay $18 a month and get incredible value out of Canva. So, it may take time, because none of us really want to replace ourselves with agents, right? It's our team.

The more people you have on your team, the more you're going to deploy agents to replace them. The more you're just yourself—a solopreneur—the more you're going to use AI tools, but not agents to replace you. Agents enhance you. Listen, I don't know the answer to your question, but I think this is going to harm enterprise workflows before it hits the prosumer market.

I think there's a lot in that, and I want to put a bookmark in it. I think that's actually very helpful. What it means is, if you take it to be 3 categories for software companies, a low terminal value, melting iceberg, you're in trouble. You have a low stock price. We'll talk about that later. And if you're leveraged, you're dead.

Then the middle category is system of record. They're going to keep you forever, but not have a ton of agentic activity on top. You're going to be worth something. There is a positive terminal value. It's calculable, and there's a price at which you should buy the stock.

Then the happy outcome is the agents are using you, and you're getting increasing returns from AI leveraging your technology. So, if you put those 3 buckets together, I think what you're right in saying, Jason, is that successful enterprise software companies can easily get to that top bucket, because you're right: companies want to automate, because it's called taking costs out, and it's called making yourself more efficient.

Successful SaaS companies in the enterprise that adopt this reality can probably reignite growth. Obviously, unsuccessful ones will fail. But that's what you're saying about Canva. I don't have a feel for it. I'm not a design person. I like the team, but I'm not a designer. I have zero creativity.

I think you could be right, which is that the individual user or small user wants to have AI tools, but they don't need to create a whole AI-automated workflow, because they're just not doing enough for it to matter. Intuitively, what that says is they end up in that middle bucket.

Bringing it back to the IPO, I think your point is that they have the scale and the profitability to be an IPO. The problem is, as we've discussed before, so much of venture is about the pixie-dust upside. Any IPO without pixie-dust upside just gets priced like a real company.

It's always a bummer for venture people when a company gets priced like a real company, because it's so much easier to make money when you get pixie-dust credit. The truth is, SaaS pixie-dust credit expired. We've talked about Rippling growing 70-some-odd percent at $1 billion, right? That, if this were an AI play, would be a jaw-dropper, right? AI—will it trade at a SaaS discount?

Guest 2

I sure hope not. No. It'll trade—see, I don't like the second half, because I actually think Rippling's a great story. It is, and I know you do too, but it's going to trade on a sensible adjusted P/E multiple based on growth and based on cash flows, entirely rationally, in a way that any value investor could buy it.

By definition, that means it won't trade like SpaceX, which is going to trade on hopes, dreams, and prayers, right? The dirty little secret of venture, again, is how much of your money you make in that 1 year in 10 when everybody buys the dream.

So, you're right. I think it's going to be a great outcome. It's not going to trade—I don't think of it as a SaaS discount as much as I think it's going to trade at fair value. Maybe that's the way to state it even more starkly: a lot of venture capital makes money when its assets don't trade at fair value. They trade at a narrative premium to fair value.

Rory O’Driscoll

4. 45B Floods into Anthropic from Google & Amazon

And in that 1 year in 10 when you make 30% to 40% of your total cash back, you get an unexpected gift, right? Yeah, good SaaS companies that aren’t AI-first are going to trade at fair value, which means if you’ve created value, you’ll get value. I think Canva and Rippling have both created enormous value. So they’ll get value, but what they won’t get is that stupid 30-times-revenue premium that, looking back, you might have gotten in 2021.

We started on OpenAI missing numbers, switching to Anthropic. You had Google committing up to $40 billion: $10 billion in cash now, $30 billion at—sorry, at a $350 billion valuation—and then $30 billion based on performance milestones. Then Amazon added another $5 billion to the round. This was the latest fundraising news from Anthropic. How did we analyze this? And is the ultimate loser here, when I read this, not Nvidia? You’re training on Trainium and TPUs and getting closer there, but no Nvidia.

Rory O’Driscoll

I mean, there’s just a lot to disentangle. Let’s put a pin in Nvidia for a second and go back to the big picture: What did the deals mean? I’ve been thinking a lot about this. Remember I said earlier, right? You have 2 jobs when you’re running an enterprise foundation model, leaving aside the consumer business. You have to build amazing models, and you have to buy enough compute to make sure they can run them at the demand you see. Both jobs are incredibly hard.

The funny thing is, right now, OpenAI got 1 job right: They have enough compute, and they got the model wrong. That’s why they’re in trouble. Anthropic did it exactly the opposite way, right? They got the model perfect. In fact, they may have over-succeeded. As a result, they’re light on compute, right? So that’s what’s going on at a big-picture level.

And Amazon—sorry, Anthropic—is massively constrained on compute, which is why they’re doing these big deals. Dario has articulated in the past, “I’m a little careful about this.” Let’s get real: No one had a business plan last year when they went from 1 to 9 that said they were going to go to 30 by the end of Q1, right? They were hit by their own success, right? So that’s what happened.

Going back to the 2 big jobs, I just internalized how incredibly hard and risky the second job—the buying-compute job—is, and how capital-intensive this is. I don’t think we internalize it, right? I’m thinking about it. If you’re at a $10 billion run rate right now, which is roughly Anthropic at the end of last year, and you’re looking forward 2 years and think you’re going to go 5x this year and maybe 4x next year—not crazy—which means you’re going to be 20 times bigger, that’s $200 billion 2 years from now, right? Let’s say it’s $100 billion 2 years from now, right?

Whatever capacity you have today to serve $10 billion, you run that model and say, “Now I need capacity to serve, 2 years from now, 10 times that amount, which is $100 billion. I need $90 billion of new capacity,” right? The capital intensity for every dollar of run-rate revenue is probably $4 or $5 of capex to support that, right? So if you’re going to add $90 billion in revenue capacity, someone between you and your partners has to find, plus or minus, $300 billion to buy chips, dig holes in the ground, build data centers, and make it all happen. It’s easy to lose sight of that. Think about how capital-intensive that is: You’re doing $10 billion in run rate, and you’re effectively saying, between you and your partners, that to be able to meet demand 2 years from now, you’ve got to invest $300 billion. Not all yourself—some of it through your partners—but think how—

By the way, if you get it wrong and end up doing $200 billion in run-rate revenue, you’re going to have only half the compute you need. You’re going to look like an idiot. And if you get it wrong and only get to $50 billion in revenue 2 years from now, you’re going to be left with $150 billion of stranded capacity.

I mean, we lose because in software land it was so easy. If you sold more, you made more money. You didn’t have to spend a lot to make that happen. At worst, you had to hire some reps. Microsoft had to hire no one when they exploded in revenue. They just shipped more PCs; more PCs shipped, and they got their $20 per PC. In this case, 2 years before you get the revenue, you have to bet 4 times that amount on capex, right? My big aha from this is, it’s obvious when you say it, but how incredibly risky this bet is.

It’s no accident that if you look at the 2 CEOs, who is going to take the risk to the upside and just spend the money and, you know, devil take the consequences? It’s going to be Sam. So he’s got lots of compute. And who’s the more careful guy and might underspend? It’s Dario, right? I don’t blame either of them, in the sense of the sums involved. It’s not just that the business is capital-intensive, that there’s $4 of capex for every $1 of revenue. It’s also 10x-again growth. The combination means you have to bet 4 or 5 times 8 times your current run-rate revenue in capex just to meet demand, and you’ve got to do that every year.

Guest 2

Yeah, I don’t think it’s a huge deal, but if OpenAI really missed last year—and I think some of it’s definitional, what the miss is, right? We’re reading a The Information report. Again, I don’t think this is a huge deal, but it’s possible you look back and see that as the first disconnect from “compute equals revenue,” right?

The risk mitigation, to Rory’s point, is, as stressful as this is, all the spend—if Sam’s right that compute really equals revenue 1-to-1, if there’s a perfect correlation—then it all kind of works out in the end, assuming that capital is available. If that breaks for any reason, then it just adds a level of risk to the model that’s even higher. I’m not saying that happened for sure, but superficially, it seemed to have happened, right?

Rory O’Driscoll

Yeah, Jason, you’re exactly right. I think it’s a fucking stupid statement by Sam, right? It’s correlation; it implies that correlation is causation. It’s just correlation. Let’s rephrase that statement, because Altman’s statement is, “Compute equals revenue.” Not true. I can tell you what is true: No compute equals no revenue. But compute and a shitty model also equals no revenue. See Grok for details, right?

The truth is, to succeed, you need to have enough compute to meet demand and a good enough model to generate demand, right? You’ve got to do both of them in sync. It’s hard. So I agree: Everyone was making that correlation argument as a causation argument—that compute equals revenue—only because, while they were making that argument, the demand seemed almost infinite. But the minute your model underperforms a little bit, it’s not quite infinite anymore.

The good news is, if aggregate demand is going up 5 to 10x per year, I think these air pockets are just going to be air pockets for both sides, right? Zooming out, Jason, the big-picture comment is agents. What do you think? How many more tokens does your agent use per day than you did, Jason?

Guest 2

Our Salesforce bill went up from $12,000 to $22,000 a year, and our seats went down from 10 to 2 plus 1. So there’s your math.

Rory O’Driscoll

Well, what about your tokens? I’m actually interested.

Guest 2

I don’t know the number, but it’s derivative of it. It’s like your cost-center number: Dramatically up. They’re using dramatically more tokens.

Rory O’Driscoll

That’s my point. I think they’re using literally—I saw a number like it’s 50 to 100 times more expensive, in terms of tokens, to serve an agent than a Jason. Actually, probably 10 times more than a Jason, 100 times more than a Rory, because, I mean, you’re pounding on it—

Guest 2

It runs constantly. Exactly.

Rory O’Driscoll

It does, if you let it, right? So the good news, and the reason these guys can all take these risks, is that in the short term, “compute equals revenue” is not always true if your model’s not there. But the big-picture trend is, as agents kick off, the demand for compute over the medium term will be there. So it makes sense to lean in, but you should also accept you’re leaning into something where there are going to be wide short-term swings.

There’s probably going to be a 6-month period where you’re like, “I’m an idiot. I don’t have enough demand.” And then 6 months later, “I’m an idiot. I don’t have enough compute.” It’s just going to be the journey.

5. "Compute ≠ Revenue": The First Crack in the AI Business Model

What’s easier to rectify? Is it easier to resell excess compute that you have, or is it easier to buy compute that you don’t have?

Rory O’Driscoll

Again, I hadn’t thought of it, but the problem is, if you’re one of the 2 big guys, you are so much—what are you going to do? Can you imagine it? OpenAI, you have a gigawatt of excess compute; Anthropic is desperate for compute. The hell, you sell it to them? You might say, “You want to buy it?” “Oh, sure.”

Guest 2

It sounds crazy, but Samsung would build phones and then sell its components to all its direct competitors, right?

Rory O’Driscoll

Totally.

Guest 2

That's fair. You get zen about it at some point. We're going to have 2 divisions: our compute division and our application divisions, and they've got their own P&Ls. What's more, remember, you don't have that. The truth is, you actually have the compute on a long-term contract, but Amazon, Google, Microsoft, CoreWeave, or Oracle will actually have the compute.

So maybe the way to phrase it is, if foundation model company A can't take their take-or-pay, the hyperscalers will probably take that compute to foundation model company B and say, “Hey, guys, I got some cheap short-term compute. It's like a sublet. Yeah, just a $10 billion sublet.” So, yeah, there will be some kind of market.

What we're seeing de facto is that happening right now at a macro level. Remember that whole CoreWeave—the one reallocating a data center from company A to company B? That's just going on in real time. People are trying to figure out, again, remember that forecasting problem I articulated.

On top of that, there's a 2-year lead time. It's not like you're forecasting next month's demand. You have to forecast 2 years out, bet 10 times your revenue on CapEx, and hope you're right. It makes running an airline look easy.

It benefits Google, too. Google's the big winner here. Why?

Rory O’Driscoll

Well, first of all, now Anthropic's deeply tied to them, right? So Google wins whether you use Gemini or whether you use Anthropic now, right?

Second, Google has infinite capacity because they're the largest provider of traditional web software. So they have all this capacity for themselves that they can allocate even better than Microsoft. Do I want to give it to my own compute? Do I want to give it to Anthropic? Do I want to give it to them?

They have the surplus—to Jason's point, to your point, Harry—that they can route between their customers and themselves and others. They win-win here. They have Gemini, they have Anthropic, and they have the capacity, and they have the ability to rotate it where and when they want. They also have the cash flow to manage it all. So, Google: win-win-win. They definitely have one.

And, to stick with the “more ways to win” comment, we forgot Harry's original question on Nvidia. The last shoe to drop here is that both Amazon and Google have chip products they can bundle into the equation. For context, GPU spend is roughly 50–55% of total CapEx on any buildout. So if you're building out a one-gig data center—and estimates range from $30 billion to $40 billion—$20 billion of that is compute.

Nvidia's gross margins are 70%, which means $14 billion of that per gig is raw profit to Nvidia. So if you're sitting there, that's one of the things Google and Amazon are trying to do: substitute that for their chips. Jensen will make the argument, as he did on the podcast, “Dude, it's a mistake. Our chips are better. They have more support.”

6. Why Google May Be the Biggest Winner in AI Infrastructure

You have to be in the weeds on that to know the exact answer, especially for specialized use. Google and Amazon would say that Nvidia's advantages aren't as good for specialized use, but I wonder myself. Nonetheless, that is what's happening: some attempt to bundle.

Neither of those 2 chips—the Google chip or the Amazon chip—is widely available on a standalone basis. So what both hyperscalers are doing is effectively bundling their chip with their capital and their equity investment to convince Anthropic to continue to run on their products and just take more of the gross margin, arguably with, as Nvidia would say, a substandard product. But tech has many examples of substandard bundled products succeeding. See Microsoft for details.

Mini quick-fire round. Google hit $4 trillion. Nvidia's a $5 trillion company. For maximum value gain on a per-dollar basis, which one would you invest in today?

Rory O’Driscoll

Okay, not the question I was expecting. For maximum dollar gain, it's a bad question. I'm not doing my thing again. I think, risk-adjusted, I would do Google, reluctantly, because if you just wanted the upside, you can paint Nvidia as a more single-threaded story around raw CapEx demand.

I think risk-adjusted you probably would do Google because, even to Jason's point, Nvidia's biggest advantage is if this thing happens—if this 1 thing happens, which is a CapEx explosion—they get it all. Google's biggest advantage is that it has multiple ways to win. It can win if AI adopts fast; it can win if AI adopts slow.

It's kicking off cash, so it's got a bunch of steady businesses. Only 1 thing can go wrong: ChatGPT eroding Google Search, which is the mother lode of cash. Provided that doesn't happen, they're golden. So, risk-adjusted, I'd probably reluctantly buy Google.

Guest 2

No, you gotta do Nvidia, okay? Because, despite the fact that it potentially has reached its market-share ceiling with Anthropic and others, it's the best pure play into the AI vector.

Rory O’Driscoll

I agree. I think we're saying the same thing.

Yeah, we are. So you don't want to minimize your risk. Just put it into VTI or bonds. If you want to bet on AI today, because we can't buy Anthropic or OpenAI, just buy Nvidia. That's how you buy AI today. Just buy Nvidia. Don't even think or spell it. Just buy it.

Rory O’Driscoll

For what it's worth, that's totally fair. I think if you're just going for max upside—yes, if you wanted to create your AI upside exposure, it's Nvidia and then a bunch of other weird things we can talk about another time.

Guest 2

Yeah, I don't even buy CoreWeave or these crazy things. Just back the truck up to Nvidia. If Nvidia loses—if AI stumbles—it's okay.

I want to be a long-only manager. Fuck it. Buy Nvidia, buy Google. Done. Go home for 3 years. Seriously, this game is great. I wish I were a long-only manager. It seems like the place to be.

Rory O’Driscoll

Yeah, just charge your fees and commissions and just—

Well, it looks good today, boys. Let's buy Nvidia. I heard good things about Google. My friends use it. Let's fucking buy some Googles.

Guest 2

Okay, fine. Yeah, our show was fucked. Yeah, we love Jensen. Go Jensen.

Rory O’Driscoll

Yeah, we love—look, the data says most managers—I mean, we all know that—underperform the index. Especially if you adjust for beta, they underperform the index. So no, it turns out to be remarkably hard, Harry, but keep telling yourself that.

I think it's because they don't do Google and Nvidia. I think it's because they try and have a diverse portfolio.

Rory O’Driscoll

I agree. Yes, and when you're not diversified, you're either right or wrong. I mean, yeah, survivor bias here, but yes.

Guest 2

Yeah. Yeah.

Well, look, ARK is down 3.88% this year. Is there anything else on Anthropic or OpenAI that you want us to discuss?

Guest 2

There's—I mean, there's a couple of things being missed: the ads.

7. China Blocks $2B Manus Deal

No, we're happy to move on. I don't want to be all Anthropic all the time. Right, let's do it. China blocks Meta's $2 billion acquisition of Manus. This was a surprise. Distributions have been made to investors. The company is a Singaporean company. The people aren't in China. This feels like a regulatory overreach.

Guest 2

Well, Benchmark has their money, all right? Who cares? If I owned 20% of Manus and got my $400 million out, I would love the boys. I'd want to help get the boys out of China, don't get me wrong, but I don't care if I got my money out. I ain't giving it back. I'm not accepting service of process. I'm hiding from the service-of-process provider.

I'm keeping my $400 million. I'm taking my $80 million in carry for myself, and I'm hiding. Dude, I don't know if you can hide in Woodside from the CCP.

Rory O’Driscoll

It is a real risk, but I ain't giving my money back. If I'm Benchmark in France, I ain't giving my money back. I don't want to trivialize it. There are humans at the heart of this who are at risk, stuck in China, right?

But I do agree with your assessment. The investors who've gotten their capital out—the chances of them having to, and being willing to, return that capital is 0. So when China says they want to unwind the transaction, I actually don't think they're talking about the money as much as I think the leverage point is over Meta, where they're really saying, “You have this technology; we'd like it back.”

And let me give you a clue: if that had happened to Tesla, where they have a massive car plant in China, they'd be coming to the table right now with the Chinese government and saying, “Maybe we shouldn't unwind this transaction, because you've got a lot of leverage over me,” right?

From memory, I think if you do a lot of business in China, this ruling is going to start a discussion. If you don't do a ton of business in China, no one's going to be pursuing the venture investors. I think that, to some extent, it's going to be pushing on Meta.

Then, obviously, the more human thing is that some of the team are still based in China, and they're not going to be able to get exit visas, right? I think this process will unwind. What it's really doing is less about getting this thing back than preventing it from ever happening again.

That's the first, last, and only one of these deals that anyone will do, right? Because I just think it's going to be really—unless literally, before you wire your money as a venture investor, you know, the night before you put everyone in a 737 in Beijing and say, “Dude, we'll wire the money when you hit Singapore and bring your family”—it's just not going to be a thing, right?

So who wins and who loses, then? Meta loses, then? Just because they've lost the money.

Guest 2

They've paid, and they're not getting it back. But they have the technology, and any of the team that's based in Singapore they have, right? I actually think what happens is there'll be some resolution. As I said, I go back—I don't remember how much business Meta does in China, right? But if they do a lot, they'll have to settle. If they don't do a lot, I can't even remember. I know Google didn't for the longest time.

I just don't care what Meta does in China. Neither subject interests me, and the combination interests me less. But I think that if they do, they're going to feel some pressure. As I said, just like if you were a big US manufacturing company or Tesla, and the Chinese government took this position, you'd have to take it seriously because they'd say, "Otherwise, we're just going to register a $4 billion judgment against you and exercise it against your local plant." Have a great day.

Human issues aside, to Rory's point—I don't mean to minimize them, right? I would just take my carry and hide. I don't think you can. The service providers will come from China. I don't want to spend too much time on it, but I think it will be a minor blip in some upcoming AI war between China and the US that is difficult to fully understand today: how this war goes, right?

Will NVIDIA supporting AI to China—let's do more; that's in their best interest? Others are against it. It's clearly a war at some level, but I'm not smart enough to fully predict where it will go. This will just be a start—not the start, but one of the first expressions beyond this NVIDIA chip drama—of where this war will go.

It's a war. I agree, and I don't love the word "war" because I think that implies actual violence, but I think you're right. It's funny: you often have to step into the other person's shoes. If you're looking at it from China's perspective, there was someone going to go to prison somewhere, I think in Singapore or the US, for selling NVIDIA chips to China in breach of the sanctions, right?

And they're probably sitting there going, "Well, if you won't give us your chips, I'll be damned if we're going to give you our research." It feels a lot more balanced from their perspective: "You evil Westerners are putting this dude in prison, and all they tried to do was sell us some Blackwell chips. Back off." The sanctions we're exerting on them probably feel problematic to them.

Now, I remain on Team USA. I live in Team USA. I'm with Team USA. But just put yourself in the shoes of the other side and think what they're probably sitting there going: "We'll show you with Manus like you showed us with NVIDIA."

Guest 2

No, it makes sense. It's at least slightly tied to DeepSeek raising outside financing to $20 billion, right? Maybe "war" is the wrong term. I think there's 2 great battles that will come before this pod ends, right, that are subtle, that we won't hit every week.

One is this: China versus the US in AI is a battle that's happening. The other is just the social dislocation from AI. It's already happening. I think there'll be more revolts and issues as layoffs happen. I think California will pass its billionaire tax and the exodus will continue. I think New York is already trying to pass its penthouse tax, which is already leading to wars with the Citadel founders and others.

So there's going to be this theme of social unrest and this war, this battle with China over AI, that won't bubble up each week. But I think at a meta, nonpolitical level, these are the 2 big themes that I think we can't ignore in our quest to get rich fast. And we're going to have $3 trillion IPOs. Who cares? Who cares about the little guys when we have $3 trillion IPOs? Who cares?

Yeah, but I think that bit at the end sounds mean, Jason. I think what you're saying in the rest of it is it turns out that the non-trillionaires, the non-billionaires, can see that the billionaires don't care. And you're right, I think the political climate has shifted.

This isn't the thing that preoccupies my day because I'm just trying to do my job. But you're right: if you were to zoom out and write a social history of the 2020s in 30 years' time, I think you're exactly right. Historians will talk about the revolt against inequality and AI, and they'll talk about the China-US battle over AI.

Guest 2

I think it's a very good framing. I think those are the 2 big social-political framing things here, provided we don't blow up the world.

Right. And I think it's interesting because I haven't seen the polling on the billionaire tax. My rule of thumb used to be that California—the electorate is quite sensible. They elect Democrats, but they're pretty profoundly right-wing at heart, which is what no one ever talks about.

Guest 2

Polymarket puts the odds of it passing in the mid-40s now.

Interesting, because normally they vote down any tax because they're like, "No, we've learned: just vote no to anything," right? Yeah, we're Democrats in our heart, but we're Republicans in our pocketbook. But if it's 40% already, that's interesting. I haven't paid attention because, unfortunately, I'm not a billionaire. I'm not in the price bracket, but duly noted.

8. Thoma Bravo Hands Medallia to Creditors: $5B Wiped Out

Okay. Now, in the venture game we have a lot of zeros. In the PE game, it's rare to have a zero. Thoma Bravo hands Medallia to creditors: $5.1 billion equity wipeout. It's the first total loss. There was $3 billion in debt, which seems to all be going.

It's just very significant because you never, or very rarely, see an asset of this scale being handed back to creditors, and it's the first of its kind. It might be second behind Pluralsight, depending on how you define it. It might be the second big one. We just weren't as focused on Pluralsight, but Pluralsight died under debt, too, under massive debt.

The size of that transaction, Jason? You're right, I'm wrong—I misspoke. It was a couple of billion, I think. It wasn't as big. No, you're absolutely right. I misspoke, so I'm sorry for that. Can we just confirm, though, on this? When I was reading it, I didn't quite get it. Are Thoma Bravo losing money here? Did they recoup that money?

Guest 2

Yeah, 100% they're losing money. A lot. From memory, I think the deal went down in 2021, and it was a $6 billion transaction or whatever, and $5 billion of it was equity. So it was not wildly overleveraged, right? Maybe $1.6 billion of debt, the rest equity. Not wildly overleveraged, right?

Fast-forward to today, they have more debt than that now. So it could be there was a minor dividend recap and they took some money out. Maybe they got 20 cents on the dollar. But the big picture here is this, and it's terrifying: this is a company, I believe, with a couple hundred million dollars in EBITDA, and despite—if you look at it from a capital-structure perspective—it was 4 or 5 times EBITDA. It was 80% equity, only 20% debt. And that should be pretty safe.

But when you way overpay for a company that now has way underperformed and, for reasons we'll talk about vis-à-vis AI, has very significant terminal-value questions, then even though you've only got a small amount of debt, the stunning thing is, with less than a couple of billion—what did you say it was, $3 billion of that?

Right. I thought it was closer to $2 billion, but that's okay.

Guest 2

They basically said the debt smothers the company, right? Even though it was fairly underlevered. What that means is, at $200 million, they basically realized that at 8 or 9 times adjusted EBITDA, it wasn't worth putting any more equity in. They've massively overpaid, and the deal's underperformed.

So it's a business that looked like nothing could go wrong in, which is enterprise software, and it turns out—and people would have said, "If something does go wrong, it would be, oh my God, you way overlevered it." They didn't way overlever it. They just way overpaid for it, right? That's the important insight that I think is missed, right?

Pluralsight was both, right? Vista apparently lost $2 billion, but it was very levered, right? This is not heavily levered, but they can't afford the $300 million of debt service, or it's not worth servicing the $300 million, right?

Actually, that's the thing, because I'd say relative to the—I wasn't precise here. In terms of the transaction size, most of the consideration was equity. So, in that sense, it wasn't overlevered. But relative to the size of the company—

Guest 2

Yeah. I think Medallia was doing $1 billion. You can't service—this is it—you can't service $2 billion-plus of debt on a $1 billion, low-growth company with a pre-AI story that has to transform to AI. You simply can't, and that's the big, scary aha across all these other companies, right?

It used to be that you'd be like, "Ah, you muddle along, you do 10% operating income, service the debt at low interest rates, and refinance it," right? You don't have a chance to do that now. There's nothing good about this because they don't have an AI story. They'd have to invest a lot to get one.

Stepping back, Medallia is kind of in the measuring customer engagement, customer happiness, kind of survey business. It's not a major system of record like ERP. It's fairly easy to transition to the next-generation product, and you can totally see a whole bunch of AI-first, very much better products in the space. We have an investment in Raptor, a small company that does customer analysis of customer sentiment.

And the point on product is that there's a whole bunch of way better AI-first products in this space. So they're looking at an asset that just doesn't have a relevant story. It's a full rewrite to change it, and it's just too hard. This is a full write-down, and that's not what this business is meant to be. I mean, that sales quota attainment was 21%, reportedly.

I think the other problem with Medallia—and I'm not sure it's true of all the ones that are at risk—is that there are some big ones at risk: Coupa, New Relic, Anaplan, even Zendesk, Avalara, and Smartsheet. They all look like they may not be able to fully repay their debt. But my limited understanding of the problem with Medallia is that it's just one of the ones that CIOs want to reduce. Correct.

Guest 2

It's just that simple. It's not even whether it's a system of record; that's an ultimate threat. But why it's already struggling to even retain 100% of its revenue is that, when you sit around the room, one under-discussed factor is the amount of vendor consolidation that's occurring at the same time as AI growth. Whether you look at Gartner's numbers, 30% to 50% of AI dollars are coming from consolidation.

Medallia's a top target. Do we really need that half-million-dollar-a-year dated survey product? Did we really learn that much from it, guys? No, so it gets cut before you cut your Workday or Salesforce, right?

Agreed. It's just prioritization.

Guest 2

I think for venture, the question is—and Rory would be the expert here; sorry, Harry, you're the boss—is: does it matter? And what I mean is, okay, so Thoma Bravo is going to take a $5 billion hit here on, like, I don't know, a $20 billion fund, right? That's not expected outside of the bounds, but it happened, right?

Even if all of these died—Medallia, Proofpoint, even Qualtrics, Alteryx, Cornerstone appears to be potentially going under, Coupa, New Relic, Anaplan—does it matter because we've got to just move on into the AI age? Does it really matter?

Guest

It matters on a bunch of different dimensions. First of all, and I'm sure I'll say it to save Cornerstone from ringing and yelling, don't say anyone's going under, because that pulls you into saying things that may or may not be correct.

Multiple term loans underperforming, apparently.

Guest

That's exactly right. They're already underperforming the loans. It's not a great sign, right? I mean, look, the horsemen of the apocalypse are, first of all, the debt starts trading well below par. And then the second thing is the debt starts doing payment-in-kind and activating the toggles that activate when you need more time. And then when the refinancing cliff happens, that's when you face the music.

So that's the movie, and I'm not commenting on any of those comments, but you're right, Jason: every one of them is in the category of highly leveraged 2021 deals, which means high absolute price. So again, back to my comment, even if the equity-versus-debt mix was fairly unaggressive, the debt as a percentage of current revenue—which is what you have to look at now, because the valuation you paid in 2021 is irrelevant—is probably pretty high.

And you're right: does it matter if half of these go bust? I think it matters in 3 ways. Well, in a bunch of different ways, actually. First, a lot of LPs are going to take a lot of losses if this happens.

I think we share LPs. You know, this looked like the other part of a balanced private portfolio. PE was always, "This is the safe part of the business," and venture we always said was risky, which is why you had to have the better return to justify the pain, right? And now, if the safe part of the business takes some significant hits, it's definitely going to reduce the appetite for risk.

But just to challenge that, is that true? And the reason I only ask the question is from ignorance. For example, most of the LPs I talked to pre-AI boom were like, "Well, we're expecting the 2021 funds are going to perform terribly. We've just got to move on." Okay, they were terrible investments. The LPs I talked to were like, "We've just got to give them a mulligan on the 2021 fund. It's done. It's time to move on, or we've got to quit the asset class."

I think a lot of LPs had internalized that the 2021 vintage was a tough venture vintage, right? Typically, smaller dollars at risk, right? I think the mental model was that the PE guys, in return for never giving that 4X, 5X upside, had been consistent 2X earners all the time.

And now it's kind of—it's one thing when your speculative early-stage seed fund blows up; it's quite another thing when your safest house's $500 million commitment to a mega-PE fund ABC ends up with subpar performance, right? And there's a lot of co-investments in there.

So I think if a bunch of these names that you articulated, Jason, do lose money, it'll be significant. It won't be fatal, but it will be significant. And in general, I've observed with people, including myself, that you can seem calm and phlegmatic about the prospect of loss, but when it actually happens, it hurts, right? So I do think there will be some element of loss there.

And then the other thing, just to put it out there, is there goes one of our exit routes. I mean, there's 3 ways to—

Guest 2

For sure. That's the biggest impact, right? There it goes, right? Yeah, I mean, you can wander around Thoma Bravo all you like and say, "Yeah, they'll say they're still doing deals," and they are, but the bar is going to be much higher because, as you know, you can't build a company big enough to go public; the strategics don't care.

So you can sell this thing for 3x revenues to a PE firm. That's not going to be true going forward, right? And that has significant consequences, in particular, for your older companies, your 2015 to 2022 companies, where if they don't have an AI story and they're tracking, they don't have a strategic outcome.

And if they don't have a strategic outcome or an IPO, what are you going to do with a $100 million revenue company going 10%? Even if it has no leverage, even if it's not blowing up from a performance perspective, the buyer of last resort is no longer in the market.

There's no exit. There's no exit. If the 3 traditional exits were selling to a strategic technology provider, one of the large incumbents, an IPO, or selling to PE, if the sell-to-PE route goes, and we all agree that smaller IPOs—aka non-massive IPOs, Andurils, or you name it—failed, do we only have 1 exit route left? Like, we have—

Guest 2

Secondaries to each other? What? I missed the route. What's the route?

I think there's no exit: selling to a strategic incumbent. Sell to Google, sell to Nvidia.

Guest 2

But here's the thing: they don't have the appetite. PE is a much better buyer for most, at least, B2B plays. The volume isn't there at these guys, and, more importantly, what they want is very specific. It's very specific. You can't count on anything.

I mean, I can tell you, when I was a VP at Adobe, you would say, "Oh, Adobe should buy these companies." It was the perfect fit. I'd be in the meetings; they never even heard of that company. And it didn't matter if you had a buddy. Unless your buddy was Shantanu, it didn't matter. They didn't care, right? It's narrower than you'd ever imagine.

What is the exit funnel of the future?

Guest 2

It's really straightforward. First of all, you're exactly right, Harry: the IPO's not gone away. They just have to be big. The strategics haven't gone away. They just have to be super-targeted. And the PE firms have gone away except at very low prices.

What it says to us is—and this is contrary to some of the received wisdom out there—at the stage all of us are investing at, which, even though it's slightly different between us, all to a rounding error is early. And I now define early as anything before you can squint and see an IPO, which is now $400 million minimum.

I mean, your portfolio construction has to reflect the reality that we call it internally fewer but bigger winners, right? Instead of having a bunch of companies exit early, you're going to have a bunch of companies taper out, maybe get so-so exits, and then the one that goes the distance and gets to $400 million in revenue could have an even bigger outcome than you've seen before.

It's kind of the corollary to the statement that we're having some of the biggest exits we've ever seen, and that's true. Both things are true together. The exits that you're going to have now are going to be huge. There's going to be a lot less of them, and therefore, from a portfolio construction perspective at the early stage—early, broadly defined—you just have to have a higher N count because your probability of getting one right is lower.

Now, at the late stage—and by late stage I now mean when you're investing in companies that could already be public, above $400 million—you don't have that risk, the risk that it won't make public scale, because you already have public scale. There are many things that can go wrong with a Stripe investment, but it's not going to fail to be big enough to go public, right? So therefore, at that stage, you see this massive concentration, because there's only a small number of companies big enough, right?

So that's why there are really 2 venture businesses now. There's the, as I say, early stage, which I think—pick a number—below $100 million ARR, where you have a pretty diversified spread, except it's fewer but bigger winners, and you have diversification.

And then there’s late, where Thrive puts $3 billion into company A and $2 billion into company B. But, as I think one of the guests on your show from Thrive said, partially it’s easy because there are only 40 names you even have to think about. It’s just a different business, right? The number of places where you can park a billion is few and far between. They’re both sides of the same coin.

I remember when I started the business in the ’90s, there were years where there were 300 IPOs a year. What we used to have was 50, 100, 250, 300. It used to be basically the Series C. That’s Harry’s average A round right there.

I know.

Guest 2

The point is this: the public markets had an appetite to be part of the IPO process through a process of regulation and a whole bunch of other reasons, and that’s no longer the case. The trend, which I thought would flatten out at the kind of 2015–2020 level, has even further accentuated. So, yeah, it’s a different game.

I literally had this discussion at a board meeting the other week with a company that just crossed $100 million, and I’m like, “Great. And you’re cash-flow positive. You’re in control of your destiny. Let’s be clear, though: to achieve your outcome in today’s market, you need to hit $1 billion in revenue, probably growing 40%.” And the room went silent, okay? Because $400 million growing 30% is not good enough, okay?

You’ll get it done. Yeah, but they’re all failed—the Navan, Figma, SailPoint, Netskope IPOs. They’re all broken or crappy IPOs. I’m not saying they’re crappy companies; they’re great companies, but the IPOs are crap. So the bar has gone up even further since the IPO, and there was just no answer.

So one of the things that I think’s going to happen is that Thoma Bravo decides these are all AI-enhanced winners it wants to buy, and Vista could do the same, which could happen. Actually, we could talk about it, but I don’t want to spend too much time. It could happen. They could come back into the market for a variety of reasons.

If they don’t, and the bar to IPO is $1 billion growing 40%, I think what’s going to happen more is they’re just going to give the company to their friends—CEOs, founders.

Let’s say I’m at $100 million in revenue, and my best friend, my peer—he’s my best CEO—is at $200 million. We’re both growing 40%. I’m done after 10 years. It’s not that I don’t care, but I don’t see any path to that IPO. I haven’t gotten an M&A offer from Google. Harry said it would come; I’ve never gotten an offer from Google. I used to get PE calls. I haven’t gotten a PE call in 3 years, and I don’t see it anymore.

So I’m giving the keys to Rory. I’m going to give a third of my company, right? Because I don’t see any exit, and the founder gets out, right? The emotional weight, the heaviness, the VCs, I guess, get to roll over this into a fake company where the valuations line up, but no one really gets anywhere, right? There are no distributions to the LPs. You haven’t achieved critical mass.

This is a micro-trend that I think is going to accelerate this year: founders giving the keys to their friends. Not completely quitting like 8 months after an accelerator that didn’t work out. I mean, I’m at $40 million, $50 million, $20 million, $100 million; I’m not going to get there, guys. So, Harry, here’s the thing: let’s merge our companies. I don’t know if Coda or Grammarly is taking any more mergers.

I’m giving my company to my buddy Harry. It sounds like I’m kidding, but I think we’re going to see this happen all the time: giving the keys to my friend who’s bigger and better than me. Just give the keys away.

I think it will be part of the overall process because, look, there’s a huge amount—I mean, there’s just a huge amount of rationalization that’s got to happen. Because, look, these numbers are big enough. I mean, if the total privately held FMV is plus or minus $6 trillion, and if the big 3 or 4 and the other guys who can comfortably get out are $3 to $4 trillion, and then the world of everyone else is $2 or $3 trillion bucks, right?

Let me tell you, no one’s going to just walk away from $2 or $3 trillion bucks. But at the same time, it’s not obvious what has to happen, and capitalism works. People are going to come up with solutions.

But, Jason, you’re right: it’s going to be some guy who’s a mid-career operator who’s willing to take the pain and is going to say, “I got this. I will take these 5 software companies, all broadly speaking in the systems-management space. We’ll put them together. I’ll run them like a hard-ass. We’ll get to 20% growth, 30% EBITDA, and just compound away because I’m a mid-market manager, and this is a chance for me to make $50 million as a CEO.

“We won’t have a ton of stock-based comp because only me and 5 other people are getting the stock, and there’ll be a whole bunch of tough, hard acts that will happen because people aren’t just going to say, ‘Okay, you caught me. It’s $2 trillion. I don’t want it.’”

Right? You know, I’m not going to walk away from my older companies. We have value there. My LPs have value, and frankly I have value. But you’re right, Jason, there’s going to be a fair amount of industrial, non-glamorous work involved in converting that stuff into free cash flow. All right, to distributable cash flow.

Before we move to venture, just one final thing on this. This is not exclusive to Thoma. You can go from Francisco Partners to Vista to EQT. Everyone’s got theirs. So, genuine question: what happens to this as an asset class, as a cohort of funds? Do they just raise the same-size funds and then carry on, and we move on?

Guest 2

My rule of thumb is this: whenever something looks incredibly easy, and it looks like it always works, and everyone who does it makes money, and everyone says that everyone who does it makes money, and it becomes the conventional wisdom that everyone’s going to make money, it’s going to blow up in your fucking face.

Right? And that’s what happened in PE. It’s like, well, you’re going to make 2x regardless, so whatever, and then let’s talk beyond that, right? And it’s going to happen in venture, too. Whenever someone says you can’t lose, you’re just about to lose money. By the way, the fact that you had 20 names all doing the same thing with exactly the same strategy, that was probably a clue.

And we’ve pointed this out in venture, too. Thoma Bravo and Vista in particular are like, “We’re all in on AI-enhanced B2B.”

I’ve only seen 1 soft offer in my own portfolio this year, okay? But it was from a PE firm that was exactly that: a startup at scale that is not growing at astronomic rates, but growing at really good rates, that is clearly AI-enhanced and in an AI category, got what I would say was a decent soft offer.

So those deals are happening, not at the rate they were in 2021 or even 2023. That’s the current playbook, as near as I can see it. They’ve been clear—Orlando Bravo’s been clear—that’s what he sees. That’s the playbook today.

The meta question is: Is the whole B2B thesis broken because it’s just not a stable category of software anymore, right?

Guest 2

And I think my sense is, no matter what, everyone’s talking their game. To use Rory’s language, I think they’re kicking the can on this issue because I don’t think most of these PE firms have a reason to exist if B2B software is stable.

Now, if it just means they need to evolve to a new category of B2B software, no problem. Raise another $10 billion, $20 billion, $30 billion. And if these AI-enhanced candidates exist that are affordable, you just buy them and do the same thing, right?

But if it’s not—not to use the trite term “durable”—there is an argument that the classic B2B market is just broken. There is an argument that even the high-flyers may not be durable—the ones, you know, the one that KKR just did for $1 billion for voice agents for plumbers, or Legora, Rillet. We may find they’re not durable.

I’m not saying I have the answers. If they’re not durable, then the whole classic PE model is broken, right? This massive amount of software. And that’s the crack in the debt market: it doesn’t appear durable. So, I don’t know, but there is a chance it’s all broken because AI has rendered it all non-durable.

That would be what the Yahoos that think Claude destroys everything would say: none of it’s durable anymore. It doesn’t matter if you’re great, or grinding, or struggling. It doesn’t matter if you’re Legora or Medallia; none of it’s durable.

It’s a great point, Jason, because in that world—and I’m not sure I believe in that world, but you’re right—if the AI-first, venture-backed startups that exist adjacent to the foundation models can’t make it with equity dollars only, then they sure as hell can’t make it with debt on top.

So what you’re saying is there would simply be no compelling investment opportunities for PE debt-type firms. It’s like the most depressing realization ever. Basically, exit markets have gone, B2B markets have gone. I do think that the exit narrowing is a little depressing, okay? And I think it will solve itself.

Guest 2

I will tell you, I beat myself up. Rory and I first met when I sold my last startup, and the post I wrote just a couple of months later had nothing to do with the timing. It was an okay decision at the time, okay? But I didn’t know about this PE market. I never would have sold at $1 million in revenue if I’d known PE would come to the rescue and buy me for 2 or 3 times more a couple of years later, when I had 140% NRR and was profitable.

It started just a couple of months later, and a friend of mine called me up.

And he said, “Hey, Jason, I just got an offer to buy my company for $100 million.” I'm like, “There’s just no way. I love you. Your little bootstrap company—who the hell’s going to buy you?” And it was the start of the PE wave. So it opened up this wonderful era, to Rory’s point, where we had Plan Bs. Everyone had a Plan B, right, for your investment.

I do think it is depressing. I think it’ll work itself out. The big exits will solve it, right? Wiz—I mean, we thought Wiz was big, and now we have Cursor. Now I’m going to win the bet of a $100 billion exit in the next year, right?

In the aggregate, it’ll work itself out, but I do think for the average person, it’s a little depressing. It’s a little depressing that there may be no exit for so many companies that there used to be exits for. I think it’s stressful as heck. It was stressful for me just before the PE wave came in. I was like, “God, I wish I hadn’t sold.” Just for this reason—only for PE—I wish I hadn’t sold.

It’s entirely plausible in a world of super-big exits that 10 super-big exits cover the entire nut from the LP perspective, such that it’s still a good business. And then, literally, nobody cares about the fact that the other 96 companies wither off on the vine, right? And the 96 other VCs wither off on the vine.

This is why many of the big firms are trying to get bigger, because they see this and go, “If there are only a small number of slots, and if you win those slots, you make $1 billion, and if you’re not in those slots, you make zero, then do what it takes to be in those slots,” right? I totally get the logic. This is what you do. It’s all Darwinian. It’s firms trying to adapt to that reality. I don’t think it’s quite as stark as that, but it is definitely on that trend line, and you have to adapt to it.

Okay, guys, we’re going to do privates. There’s a lot in privates. You guys choose. Maybe choose one with a positive slant. Sorry, my fault. No, you pick. What are the choices? There’s Thrive. There’s Chamath’s numbers. There’s Garry Tan on fake ARR. There’s SBF, the greatest investor of our generation. I think the Garry Tan one’s worth a quick discussion. We’ve hit it before, but I appreciated that he called out these issues.

Guest 2

Yes.

Can you provide some context, Jason, just for those that missed it?

Guest 2

Well, I think it was started by a guy at this legal-tech startup. What did they—I forgot. Spellbook. Spellbook pointed out—kind of made too much of it—how there’s a lot of fake ARR.

For example, I’ve got 1 investment I made that’s north of 9 figures in revenue. I get 3 different ARR numbers each month—3 different definitions. At least they’re trying to be honest, right? What’s core software ARR? What’s software plus variable usage? And what’s committed revenue? There’s a massive delta between these.

The point was, what startups are saying they’re doing in classic committed revenue—GAAP revenue, certainly—versus what a non-GAAP number has grown to is so great it borders on fraud. That was the initial point.

And rather than say, “No big deal, who cares at the pre-seed level, like YC? Who cares at the YC level—so early?” Garry’s like, “No, man. Be truthful and precise about your revenue. Be truthful.” And he laid out 5 points that hit most of the issues.

The ironic thing to me is, even I felt by the time I got through Garry’s whole memo, I didn’t even understand what revenue meant anymore. It was so correct, but also so confusing, the way we’ve rebooted revenue. I don’t know what you guys have seen, but everyone—I got burned once on this in the old days, but everyone who’s done a deal quickly has kind of been burned on this.

I personally found that if it’s been mostly disclosed, it’s been okay. If it’s been hidden, I ain’t going to make any money. I ain’t going to make any money when this is bullshit, which is to Garry’s point. And obviously, frankly, the fact that he had to say it probably suggested it is rampant at the seed stage, or he wouldn’t have to say it.

It’s rampant. That’s my experience as well. People radically lie. How can everybody get to $3 million in revenue by the end of Demo Day? Maybe everyone can’t. Maybe only a couple can.

Yeah. I think it was simultaneously really good and really shrewd. I’ll talk about the second, because the really good comment is pretty obvious. It’s necessary. You’re right, there’s all this ambiguity about the revenue. Young founders are overstating things and, at best, suckering people into making investments they shouldn’t make and, at worst, ending up in litigation and potential fraud allegations down the line if they misstate things.

Some guidance is really good and helpful. And I predict, if it sticks, the shorthand version of the seed stage will be, “Does this conform to the Y Combinator revenue guidelines?” Right? So that’s why it’s a good thing. It needed to be done.

Let me tell you why it’s a shrewd thing. If you own a market, you want to make sure that trust in the market remains. It’s a little like the way De Beers policed the diamond market for years. You want to know that people can transact in complete confidence, right?

Y Combinator has a dominant market share in the seed market—25%. It erodes the value of their product if a whole bunch of people start thinking the numbers are bullshit. So not only was it a good thing, it was a shrewd thing, because it’s now basically saying, “If you look at these deals at the margin, you want to say you’ve got the Y Combinator seal of approval. Here’s how things are calculated correctly.”

So I think, again, it was good and shrewd, and as such it’s going to stick. Some version of it’s going to stick, just like—

Guest 2

It’s a good point. If they’re a market maker, you want to have this level of transparency as a market maker, right?

If stocks lied about their revenue, at some point the NYSE would say, “We need to fix this thing here, people. Let’s get the auditors in the room.” And that’s just what happened here.

On the slightly other end of the bench is BetterHelp. Thrive Eternal—Josh just continuously bringing out new products and new packages for his ambassadors. Thrive Eternity, I didn’t want to say this, but it looks remarkably similar to Sequoia’s evergreen fund in terms of the hold periods.

Guest 2

But I think you misread it. I understand that the verbiage looked the same—“hold companies forever”—but—

And you were saying, “Is this an example of…” Again, for context, folks, in late 2021, Sequoia correctly said, “Over the long term, our very best companies continue compounding. If you had held all the companies, even the bad ones, the good ones would have swamped it, because you’d have Apple, you’d have Cisco.”

Guest 2

The analysis is entirely correct. It’s like the old analysis on any equity-return business: over any 20-year rolling return, it’s positive; over 10, most are positive; over 5, some are positive. And every once in a while, over 1 year, it blows up in your face.

Unfortunately, Sequoia opted to do the eternal hold—every stock forever—in that 1 year where it blew up in your face, right? So they felt a little foolish about that. Though I think over 10 and 20 years, their analysis will still be correct. If you build enduring companies, even in the public markets, the compounding will happen, right?

So that was the Sequoia comment that Harry was referencing, but I think the Thrive product is actually very different. If you read the prospectus, or at least the information on it, it’s much less about holding a public stock forever. It’s actually interesting—very marketing- and positioning-oriented around different kinds of assets that aren’t impacted by AI, that are going to be eternal.

It’s an entirely different form of investing, because I think their first investment is in one of the San Francisco teams. I can’t remember which one. Is it the Giants? I can’t remember. Was it the baseball team or whatever?

I think it’s the Giants.

Guest 2

Yeah. Again, my point is, what they’re actually doing is a totally different product line. They’re making the big-picture point that there are assets beyond the digital that are enduring and can’t be replaced in any way, shape, or form by digital.

They’re right about that. There’s no amount of automation—like those stupid people who say, “Oh my God, the robots can run faster than people in the half-marathon, therefore it’s over.” Well, as someone pointed out, a Toyota Corolla can drive faster than people, but we still watch the marathon, right?

What they’re saying is this group of assets is so different from AI that they’re enduring, long-run media assets. At that level, they’re correct. I don’t know if the average venture investor would be a really good buyer of sports assets, though history would say the Warriors has been a great deal.

It’s just a totally different bet than the Sequoia bet. It’s a different asset type, and if their LPs want to do it and they can pull it off, the guy’s showing great taste. Good luck to him.

This is totally off script, but we do The Business of Sport Show, where we interview the biggest owners of sports teams in the world. This business of sport is dictated largely, in Europe at least—I don’t want to speak for America—by media rights.

If you see the personalization of media, whereby everyone gets very independent media that they consume, whether it’s games or TV shows, they can customize and craft it to their own preferences, and it impacts—or, to Jason’s point, maims—the consumption of sports, then you have a significant impact on the digital-rights package that teams will get.

Guest 2

That is very, very significant. If you wanted to paint a world where AI changes content-consumption patterns, that has the ability to significantly maim digital rights for these sports teams, which would significantly impact their revenue-generating ability. That would be the bear case.

You're right. In the case of sport, you have the individual personal journey. You're seeing a bunch of that at the margin in sport, even at the high school and college level, where the athlete's personal journey is a large part of it and they can monetize that. In fact, the way Lionel Messi monetized being Lionel Messi when he came to America is an example of that. He extracted the value, which by definition means that value the sports-team owner didn't get because he was able to get it.

I do hear your point at the margin. I still think if you own the entity that's playing the game, you do have the marquee asset. The US, in fact, has been even more successful at creating sports money-printing machines than even in Europe.

I do hear you, Harry. Mind you, I will say something I said earlier. You do go back to that comment I made earlier, which is: when something is so obvious that everyone thinks it can't lose, that's just the time when you do lose. Sports has been a home-run win for 20 years, maybe 30 years, right? It's been the one irreplaceable asset.

I'll give you one fun example. When Ryan Smith sold Qualtrics, I think he made about $1 billion after 20 years or so. I believe that billion, most of it went back into the Utah Jazz, and it has quadrupled.

Yep. Absolutely. As sports teams go up, it's quadrupled.

Guest 3

Now, he needed the billion, of course, to lead that takeover. But he's up $3 billion on the Jazz or something like that, versus the 20 years to get there, on top of that.

This is a super US-centric perspective. Sports teams do not go up.

Guest

I just want to get back because I can comment on that. Actually, sports teams in Europe do go up, in the sense of—one key difference is, yeah, some of the best worldwide assets are some of the European sports teams.

One key difference, though, in England in particular, is that you have the concept of relegation, which our American friends might not understand. In the NFL, you're always in the NFL, and no matter what happens, you're in the NFL. Same thing in basketball. In English soccer, if you're one of the bottom 3 teams in the Premier League, you get kicked down 1, right? Your economics go to shit.

As Harry pointed out, Tottenham looks like it's going to be relegated. Leicester's been relegated twice. We're seeing this with Clearlake's ownership of Chelsea, whereby Clearlake's LPs—this is publicly reported—have been significantly concerned about the amount of time they're not being relegated, to the point of actually impacting enterprise value.

No, because they're not being relegated to the point of actually impacting enterprise value. I'm going to leave that for a second because I actually think the more interesting point, going back to the relegation comment, is that Europe, the alleged socialist capital of the world, has a far more performance-oriented sports culture than America, where it's a nasty little oligopoly.

The NFL and all the American sports leagues have been constructed partly because they're the only 3 businesses that have an exemption from antitrust. They're all constructed as nasty little oligopolies where there's no penalty for failure, which is the definition of socialism.

Europe, in general, from an American perspective—which is meant to be the home of mollycoddling socialist wimps—in fact has a brutally accountable soccer culture, whereby if you're at the bottom of your league, you go down and your revenue goes down 5 times. I actually think it's one of the best things about the English Premier League and the English league system in general. There's real penalties for failure. Wrexham could go up.

100% aligned, Rory. Of course, it's worth pointing out it's not the only part of Europe that has that accountability, and we have it everywhere else.

Guest 3

But okay. It shows what you think is important, Harry. Although I have to say, I don't think anywhere hates billionaires as much as the US right now. Maybe Norway does, but I wouldn't say you're exactly pro-capitalism, are you?

Yeah, okay, keep going. Jason, you can choose 1 more. Rory delegates decision-making to us on topics. Maybe a happier one.

Guest 3

I do wonder—I think the last one that would be interesting, and then the next show will be all happy, all good times. One that is mixed at the end, but maybe it is good times: I just think it's worth touching on Robinhood Ventures Fund I and the AngelList USV C Fund. Are these good, bad, or ugly? Should I put a couple hundred grand into each of them? Can I put them on the Disaster Fund website if I do? The underlying entities—are these good investments, crappy investments, or are these just play investments for a token amount of your portfolio, and it just doesn't matter?

Guest

I think it's catering to a need, which is that public investors have been denied access to these products and want to do it, right? It's a way to say, "I got an investment in SpaceX, Anthropic, and OpenAI."

At a symbolic level, I think they'll get some action, right? As proof of that, I put literally the lowest amount possible in the VC product this morning. I'm now an individual investor in Anthropic, SpaceX, and OpenAI. Even as we speak, I'm adding the logos to our website, right?

We'll have to have a disclosure at the start of each show: Rory is an investor in all of the show's companies discussed on today's 20VC.

Guest

My $500 works for me.

What? What's the minimum? $500? You cheapskate. You put in $500.

Guest

I just thought I genuinely wanted to process through the thing this morning because, in anticipation of that, I tried to use the products. By the way, wonderfully easy flow—took 10 seconds, done. It uses Plaid, which we can talk about in a second.

But the serious comment is: are they worth doing? I mean, in 2 of them, I think 30–40% of it is in those 3 investments. It boils down to whether you think those investments are good at $180 billion, for whatever it is, and $175 billion for SpaceX. I don't know what the stated value is because, look, for $500, I'm not doing the analysis.

Would you put 1% of your net worth in there, which is kind of the level of diversification? If you step back, I've been looking at this: if the big 3 go public at around 3 or 4 billion dollars, it's a little under 5% of the S&P 500. If you're 60% equities and 40% bonds, and you wanted to get that action a little earlier, putting plus or minus 1% of your net worth in a vehicle that offered those things privately would be logically correct, which is different from saying it is correct because I haven't looked at the valuations.

Before I put 1% of my net worth in there, I'd want to do a lot more analysis, but that's the product they're offering. If you think those valuations are correct, you could do it. It's a little like the logic for doing blockchain. Do you put 1% of your assets in Bitcoin? Do you put 1% of your assets in these high-market-cap companies?

I personally would be angsty about the valuations on aggregate before I'd put 1% of my net worth in there, but I get why the product exists and it's probably going to do reasonably well.

Guest 3

Let me ask a question that I'm ignorant on, Harry. Sorry, it's your—you're the boss—but there was some controversy on Twitter. AngelList charges 3.61% a year to manage this fund, right? I'm confused. On the one hand, for a mutual fund, that's going to destroy your returns, right?

If you charge me 3.6% a year to manage the S&P, not only is it expensive, but over 20 years it destroys your capital, right? Their point was, "Our cost to deliver this product, this complicated venture product, and manage these funds is actually as high as 3.61%. In fact, we're subsidizing that because it's not even 3.6%." Is this a high load on a mutual fund or a cheap way to get into the underlying managers and underlying funds?

Guest

I think what it proves is the argument for companies going public. First of all, you're right: if these companies were public, to look at the system as a whole, the companies would have to pay $5–$10 million a year more in compliance costs, but individual investors could buy into mutual funds that are paying 50 bips or less versus 361 bips. It would be a lot cheaper, right?

On the other hand, from the venture side, as a private asset, 3.61% is high. But let every venture investor—let he who is without sin cast the first stone. The average venture investor is charging 2% and then 20% of the profits, which typically turns into, if you're successful, a 4% or 5% drag between gross and net, right?

It would be hypocritical of me to say, "Oh, 3.61% is awful." If we're successful, our fee drag should be around 4%, including carry.

I guess the counterargument—you’re better than me—the counterargument might be: it's a fund of funds, so it's expensive for a fund of funds, right? I don't—

Guest

Yes, but the only reason you can pay, in the long term, 2% to VCs and 20% of the profit is because the gross returns have to be high enough—25% plus—that the net return is still 20%, which is so far above the Ibbotson small-cap return of 11% or 12% that it's worth doing, right?

If your gross return is only 10–15% and you put 4% fees on top of it, then you would have been far better off in the public markets.

So, the question is, do these companies still have 15% compounding returns from here? And look, the bigger you are and the closer you get to the public markets, the harder it gets. Now, it has to be said, the companies that have proved every sentence that I’ve just uttered to be incorrect have been Anthropic and OpenAI, where you’ve had 10x returns at $60 billion in the case of Anthropic. And the truth is, that’s why these products are taking off. There are some companies that, even at $60 billion, have demonstrated wildly great returns over an entire business cycle. Across all the investments of that size, will it return 10x? I doubt.

The lesson is, Rory, to your point: who made money from Medallia? Ultimately, Sequoia, baby. Who makes money from Anthropic with a 17.5% carry and a 1% upfront fee? Goldman. Be Goldman or be Sequoia—that’s the takeaway.

Guest

Yes. You know what? A related lesson from that: Sequoia owned like 40% of Medallia, right? It was basically bootstrapped, right? I think a reminder lesson is—and you don’t want this to be true—but when a top fund doesn’t go all in on an investment, it’s such a bad signal.

Not only is it bad if a tier-one fund—if Andreessen—does your seed and doesn’t lead your A, that’s the classic discussion we could have done on 20VC in 2015, right? But the subtle one is when you do the growth round, when you do the billion-dollar round, when you do whatever, and you don’t see the big fund lean in for the super pro rata, I just think it’s a terrible sign in today’s world. I know people are going to challenge it, but it’s my experience. If they’ve got the billions to deploy, they’re going to put it into your winners, and if they don’t stick you in the side of your chest with an elbow to get super pro rata, it’s a bad sign.

I’m going to be so honest: I just couldn’t take it. For the last few weeks, it’s been gnawing at me so much, my Figma and Duolingo positions. I was like, you know what? I’ve just had enough. I’ve had enough of this conversation. I’m selling them all. While you guys were doing my Skydio, I just sold Figma, 40% down.

Guest

The agents don’t need either of them, Harry. I in no way want to run your money, but okay. You do you.

The agents don’t need them. I do. I’m up 24%, Rory. I agree, but it just seems—yeah, okay.

Guest

Don’t worry. The big lesson I have is: don’t wait for the shit to come up. Just sell it and redeploy. I agree with that. I think that is very true.

Yeah, I think so. And that’s why I’ve spent so long waiting for Figma and Duolingo to come back. Don’t. Just sell it and put it back.

Guest

As a random comment, it is the big difference between public investing and private investing. As a private investor, you end up—especially when you’re on the board—mentally thinking, “We’re working this out together.” And the whole beauty of public companies is, no, dude, you’re working this out; I’m leaving because I don’t know how you’re going to work it out, right? It’s just a different mentality.

And it’s why I think it’s one of the things why I think venture investors can be mediocre public investors. And I talked to the best. I remember talking to Brad Gerstner from Altimeter. You can tell that’s a guy very dialed into every position and has an exit price. It’s a discipline that you need as a public investor.

So maybe I cancel my comment. You’re right. The way I’d version your thesis is: if you don’t have an active reason for holding the stock and a belief it will outperform the S&P 500, which you can get access to for 20 basis points, then why are you holding it? If you don’t know why you’re holding it, you shouldn’t be holding it.

Guest

Yeah. So, yeah, you’re probably right. Yeah.

Jason, sell me this pen on Figma, mate.

Guest 2

I do, honestly. I just don’t know that AI agents will work with Figma because they have to, but they don’t need it, right? They don’t need it forever, right? They definitely don’t need Duolingo. So I want to see the turnaround story for an agentic Figma. I do want to see it.

I’m just—you know, it’s getting—it’s May. I’m just going to leave on this: Jeff Bezos’s Project Prometheus establishes an AI lab in London. King’s Cross, baby. We’re back, boys.

Thank you, as always. Wonderfully uplifting episode. Every week, you have to have the feel-good story from 20VC. I’m voting for a new addition to the show. I’m voting for that every time.