Capital Inferno Heaven

Summary

Kyle on Down Round with James Hennessy and Raph Dixon — an Australian tech-business show, and a notably more combative, funnier room than his other appearances. Hennessy is a former colleague from Contrary Research, which gives the conversation a running licence to needle. The episode’s title comes from a line Kyle throws away mid-argument and the hosts immediately seize on: the three-party arrangement at the heart of modern venture is “a match made in heaven — capital inferno heaven. Or maybe capital inferno hell.”

The thesis: nothing has actually changed. Asked what’s different since the 2022 correction, Kyle’s answer is blunt — “the unfortunate reality is that I don’t think much of anything has changed.” He acknowledges upfront that his newsletter reads like “existential self-loathing” from a VC who won’t stop criticizing VC incentives, and defends it as the only intellectually honest posture: “it’s like having a grumpy mother-in-law in my head all the time that’s just telling me how terrible I am… helps me want to prove that voice wrong.”

The mechanism, built up carefully over the middle of the episode. Venture’s original bargain was a small pool of people willing to take venture-style risk in funds small enough that a 10x was the target — the Power Law logic from Sebastian Mallaby’s The Power Law, where most investments fail and one or two have to cover everything. Then much larger pools noticed the returns. But a sovereign wealth fund or a JP Morgan wealth-management arm needs to deploy $250–300M in a single position, and only needs roughly 1.8–2.3x over several years, an 8–13% yield rather than a 20% IRR. So the bar drops while the cheque size explodes. Firms that will take that money — Andreessen raising $6–9B fund families — then need somewhere to put it, which selects for capital-intensive businesses that can absorb $150–300M a pop. That’s the trinity: capital that must be parked, funds willing to agglomerate it, and companies built to absorb it. Benchmark is the counterexample he names, deliberately keeping fund size and return profile constant — what he calls cottage keepers in The Puritans of Venture Capital.

The Blackstone analogy is the sharpest part. From The Blackstone of Innovation: Stephen Schwarzman’s insight was refusing to be in the private-credit business or the real-estate business — “I’m in the asset management business” — which is how you reach a trillion in AUM. Venture used to be a specific kind of investing in a specific kind of company; the agglomerators have made it style-agnostic. And the tell is in the valuation: public analysts write off the carry entirely. Blackstone’s stock is priced on the 2% management fee as an annuity, “like a freaking treasury bond.” Kyle is careful that this isn’t an accusation of greed — a16z is “doing it the best,” the people are “most of the best firms in the world anyway” — it’s an ambition game, where hundreds of millions in operating budget buys hundreds of employees and 100+ investing partners against Benchmark’s five.

Flow as the worked example. Saudi money into a fund proven to deploy at scale, into a founder proven to absorb it — Adam Neumann, who raised ~$12–13B for WeWork and got out with a billion. a16z’s largest single cheque ever at $300M, justified in the It’s Time to Build housing-crisis language, into luxury residential with “saltwater pools and dog valets.” Kyle’s read is consistently structural rather than moral: “don’t hate the player, hate the game… I don’t think the people at Andreessen are bad people. I don’t think that Adam Neumann is a bad guy. I think that they are a product of their incentives.” He does draw a line at the Saudi-sponsored conference where Andreessen, Horowitz and Neumann shared a stage praising MBS as the kind of leader the world needs — “that feels super weird.”

Storytelling, again, and earlier than the Sourcery version. Pressed that Flow is really just a real-estate roll-up with an app, Kyle gives the formulation he’d repeat three days later on Sourcery: “I’m a firm believer that storytelling is very powerful, but at the end of the day reality is what matters and it will catch up. And I’ve increasingly become convinced that storytelling shapes reality. It doesn’t matter what is real. It matters what people believe.” Applied to Neumann, the story has to bridge belief until either it comes true or you exit with a billion. Applied to Sam Altman, he flags the resemblance to how SBF talked about FTX — our TAM is the global economy — while conceding Altman is a dramatically better founder and OpenAI a dramatically better business, because the point is the mechanism, not the comparison.

Hopin as the cautionary tale nobody learns from. A few hundred events a month to tens of thousands; $3–4M of revenue to $75M to $200–300M in a year or two; a $7B valuation; the founder taking ~$200M off the table. Kyle’s verdict is that it was never a real business — “who attended one of them and was like, this rocks, this is the future? Most people go to conferences to get drunk and not think about their work.” The pattern generalizes: Blue Apron fell to a ~$20M market cap and got filed as an isolated incident, and now dozens of post-SPAC companies sit in the $300–400M range and get filed the same way. “People are largely incapable of learning the lessons of cautionary tales.”

Why VCs behave this way, in one line. The episode’s most quotable original: “VCs talk about themselves as long-term investors… but the reality is that venture capital firms are just collections of short-term careerists who are trying to associate themselves with as much success as possible, as quickly as possible, and distance themselves from as much failure as possible for as long as possible.” That’s also his answer to why Contrary’s model is hard to copy — spending five or six years building relationships before anyone starts a company is “a crazy long-term game” with real risk of backing the wrong horse, and it isn’t what careerists optimize for. He contrasts it with the “hurry up and wait” game at Index and Coatue, where until someone raises, “all I can do is invite you to a dinner every once in a while.” On concentration he gives a genuinely two-sided answer: pattern-matching on schools and logos is a real problem, and that filtered pool still contains hundreds of thousands of people, so the question is who finds them first.

On AI. The Y Combinator batch where 60%+ pitched AI and 30+ claimed their own models. Why ChatGPT was “an awareness revolution” rather than a technical breakthrough — transformer architecture was nearly a decade old — and how OpenAI inherited an IBM effect: “nobody gets fired for buying OpenAI, because that’s the thing that everybody’s boomer boss has heard of.” On training data, he reads the licensing deals as territorial and performative rather than necessary (“Google’s pretty good at getting access to data”), and reads the AGI-risk messaging as regulatory capture, citing Amjad Masad at Cerebral Valley on Microsoft’s sparks of AGI framing: “who benefits if everybody loses their mind and regulates the crap out of AI? The people who benefit are the largest companies spending hundreds of millions on lobbying.” His summary position: “I don’t know that I’m scared of anything other than highly incentivized capital-driven systems that can lobby for their best interests.”

Asked to pick bubble or revolution, he takes neither. Citing his Cohere CEO interview for Contrary Research — the next ten years are indiscernible, and if we’re still leaning on transformer architecture in five years he’d be disappointed — Kyle argues it’s a story of inequality rather than a single verdict: a subset of companies will genuinely change music, art and content, and most of the rest are coasting on the tidal wave someone else’s storytelling created. Inflection getting hollowed out by Microsoft and Stability AI’s unravelling are the leading edge of an extinction event he expects to arrive in AI the way it arrived elsewhere in 2023–24 — with the twist that big-tech participation, because it’s really an extension of the cloud wars over compute, both accelerates the failures and cushions them. And the constraint nobody escapes: “you can be building a computer god in a Microsoft Azure server, but you cannot liberate yourself from the fact that you must at some point become an advertising business.”

Transcript

Transcribed locally with Whisper (scripts/transcribe-audio.py, mlx large-v3) from the published episode audio — the show’s feed carries no publisher transcript. Whisper does not separate speakers, so attribution is reconstructed from context: Kyle’s answers are his, verbatim and cleaned; the hosts are labelled JR Hennessy or Raph Dixon only where the audio makes it unambiguous and Host otherwise, and their questions are given in condensed form as the prompts that set up each answer rather than word-for-word — the episode is Down Round’s own work and lives at downround.net. ASR errors and names corrected (Mistral, Amjad Masad, Sebastian Mallaby, Andreessen Horowitz, Stephen Schwarzman, Adam Neumann, Hopin, Peter Thiel, Coatue, ChatGPT, Reid Hoffman). Nothing reordered or summarized within an answer; [?] marks anything still uncertain.

Crypto, and why Contrary has almost none

Host: Crypto’s pumping again — what tokens should our listeners be buying?

Kyle Harrison: Oh man. I’m not a crypto guy — which means I’ve probably made a lot of money on crypto, because the dumber you are, the more money you make. I can’t point to the coins. I may be a Bitcoin purist, maybe.

Host: Why no investment into crypto?

Kyle Harrison: Contrary’s portfolio doesn’t have a lot of crypto in it. Our firm is a little bit unique — it’s very people-centric. We spend a lot of time and effort identifying super smart people, getting to know them and what they’re working on. But occasionally that is not too dissimilar from having an uncle you have to talk down from a rant at family dinners: sometimes really smart people get excited about things they maybe ought not to be excited about, and we try and bring them back.

There are certainly some interesting things in crypto, but for the vast majority I’d subscribe to the belief that many of them are solutions looking for problems, and I’ve not yet seen compelling problems to solve. That doesn’t mean there aren’t people smarter than me who understand it more deeply and are going to make a ton of money solving actual problems that do exist. I’ve just not heard them yet.

AI has reached the parent level

Host: What are the big themes you’re looking at — ten-year horizon?

Kyle Harrison: Obviously the thing everybody and their mother wants to talk about is AI — including my dad, who was texting me about it today. It’s touching everything.

I saw something interesting about the Y Combinator demo day. It’s over a hundred companies pitching now, and somebody was counting how many mentioned AI. It’s well into 60-plus percent of the batch being AI companies, and more than 30 companies that had built their own models — which is usually quite expensive, especially for 30 seed-stage companies.

Host: What does it actually cost to build your own model?

Kyle Harrison: This is a very interesting and hot topic. There’s effectively a bifurcation between what you can think of as open and closed — or open and proprietary — models. There are ways to do this more affordably: if you’re just trying to leverage a model to do something specific, a lot of people have built models you can take and fine-tune for a specific use case. Facebook has Llama, which is very popular. Mistral is a French, very open AI company. There’s a bunch on Hugging Face — tons of people who’ve built models specifically trying to be open. And then, ironically, leading the pack among the not open models is OpenAI. They’re not open in the sense that their models are proprietary, so you have to leverage their APIs, and you’re pretty tied into their ecosystem.

If you’re building a model from scratch it’s typically fairly expensive, for a few reasons. You have to have a lot of data — and for people who aren’t already large platforms sitting on data, they have to go buy it. Then you have to store and compute that data, which is honestly just an AWS or Azure charge, and it can be pretty significant depending on the size of what you’re trying to accomplish. So there are a lot of getting-started costs before you even deploy. And then to deploy you have to think about how many people are using it and how often, so scale matters too. At the peak — even in the early days of ChatGPT, when it shot up to a million users very quickly — it was costing OpenAI something like hundreds of thousands of dollars a day just to keep it running.

What’s funniest about ChatGPT as a moment in time is that it was not a technological breakthrough per se. GPT as a model had been around a long time; large language models are based on transformer architecture, which at this point is almost a decade old. None of it was necessarily new. It was a way of packaging a very specific tool in a way most people had not been exposed to. So really it was an awareness revolution — which fascinates me, because they didn’t necessarily build anything magical. But overnight they inherited this IBM effect: nobody gets fired for buying OpenAI, because that’s the thing everybody’s boomer boss has heard of. Even though in many cases it’s probably not the best thing for what you’re trying to accomplish — it’s just the easiest thing to convince people they should be paying money for.

Training data, and who benefits from the fear

Host: What’s your take on training on copyrighted material — video models trained on YouTube?

Kyle Harrison: It’s funny, because I don’t know that these companies have positions on what they should be doing. When you have the CTO of OpenAI saying “I’m not sure if we are training on YouTube”… she equivocated. I reckon that was for legal reasons — I don’t think she’s stupid, I think she knows what’s going on.

And somebody senior at YouTube was talking about how if they are using YouTube to train, that’s obviously in clear violation of our terms of service — we have very clear terms of service that state — and then basically faded off and changed the line of talking. So candidly, I don’t think they know what they want to happen, or think should happen. They’re constantly trying to navigate the legal territory of what should and shouldn’t be happening, and it’s a pretty complicated thing.

Then there’s the Reddit stuff. A couple of days before their IPO, Reddit signed an agreement with Google to license their data. And Google doesn’t necessarily need anybody’s data. I think it’s more a territorial siphoning-off of who is the legitimate user of that data, as opposed to “do you really need it, can you not get access other ways?” Google’s pretty good at getting access to data.

Raph Dixon: There’s a performative aspect — establishing partnerships with the Associated Press, egos that need massaging, a sense of legal compliance. Plus regulatory capture: it suits Google and OpenAI if the new rule is that nobody’s allowed to train on copyrighted material — but we’ve already done it.

Kyle Harrison: It’s not even just boring legal posturing. It’s literally the incitement of existential dread that they’re leveraging in this legal arena. There’s a talk I’ve referenced a bunch, at a conference in San Francisco a year or two ago called Cerebral Valley — Amjad Masad, the CEO of Replit. He tells this story about GPT-4, before it had been released, being teased in Microsoft academic papers and research. They literally had this narrative of this was the first contact with AGI, and then they changed it, and in future versions it said sparks of AGI.

Microsoft is trying to stoke the flames of people’s fear — this is AGI, we’re going to have to deal with this, it’s going to take jobs, this could be an extinction event. And it’s like: well, who benefits if everybody loses their mind and regulates the crap out of AI? The people who benefit are the largest companies spending hundreds of millions of dollars on lobbying efforts.

When you go back to the closed-versus-open debate, what’s most interesting is watching these drooling incentives get bandied back and forth. I don’t know that I’m scared of anything other than highly incentivized, capital-driven systems that can lobby for their best interests. I think Microsoft’s going to do a really good job at that.

Nothing has changed

JR Hennessy: Your newsletter covered the poor incentives of the late boom closely — the high-burn, high-valuation startup nobody was building into an efficient business. Looking at AI, those incentives feel like they’re bubbling back up. What’s changed since the markets came down?

Kyle Harrison: The unfortunate reality is that I don’t think much of anything has changed.

For context — I often get confused responses to my blog, because it feels like it’s filled with existential self-loathing. I’m a VC who loves to crap on how poorly structured the incentives are in venture capital. But I feel like that’s the way to be intellectually honest: to constantly remind yourself. It’s like having a grumpy mother-in-law in my head all the time telling me how terrible I am. It helps me want to prove that voice wrong.

The way venture capital as an industry has gotten off track is that it’s demonstrated it’s a really valuable way to park a ton of money. When you have organizations with exorbitant, ungodly, Saudi-oil levels of money, you’re not actually looking for the type of yield you’d typically expect from venture capital as an idea.

Host: What is the typical venture yield people expect?

Kyle Harrison: When you think about typical early-stage venture, there are a few ways to think about what return you’re hoping to generate. One easy standard is a 10x. One of the reasons is the power law — there’s a really good book by Sebastian Mallaby that talks all about how this dynamic works. It’s the 80/20 idea: the vast majority of my investments will fail, maybe a few get me my money back, and so I need the one or two that actually work to do so well that they not only cover all my losses but generate enough return to have made it worth anybody giving me money.

The way Contrary thinks about it, we want every investment to be capable of being a fund returner. If we own X amount of this business now, and maintain close to that ownership, this company needs to be a certain size outcome for that to return our entire fund.

Ten-x returns are not easy. It’s not easy to 10x your money on things. But if you’re a sovereign wealth fund… a typical seed round is call it two to three million bucks. If you’re a sovereign wealth fund with $500 billion to deploy, a 10x on two million bucks is — you pay more for paperclips.

JR Hennessy: You distinguish cottage-level funds from the capital agglomerators — the Andreessens — who are playing a different game, where their LPs don’t expect venture-level returns.

Kyle Harrison: The way a fund structure works, there are a ton of people out there with cash — from a wealthy person who had a business do well and is sitting on tens of millions, all the way up to, literally, when SoftBank was at its height raising $100 billion funds and the Saudi Public Investment Fund was something like 40 of that Vision Fund. They’re investing tens of billions at a time.

But think about people putting out — JP Morgan has a big wealth-management business that wants to put out $250, $300 million at a time. They want to put that into one fund. And for some funds that’s more than their entire fund times three or four.

When you have these people with a ton of capital, all they really need to generate on that size pool is something like 1.8 to 2.3x their money over the course of several years. It doesn’t need to be a 20% IRR over eight to ten years. It needs to be 8 to 13% or whatever.

So this equation has changed. It used to be you had a select group of people willing to take venture-style risk, investing into funds small enough that they wanted to generate 10x returns. That was typical venture. That’s one of the reasons folks like Benchmark have kept their funds relatively small — trying to raise a consistent size of fund and generate a consistent return profile. In my writing I call those cottage keepers.

Over the last ten years these big pools have recognized that in venture the size of the returns can be quite good — in some cases you’re generating three or four hundred times your money. So being exposed to it at all is good. The tricky part is they said: I want to put $300 million to work, I don’t want to put $50 or $20 million to work. And firms like Andreessen came along and said, you know what, we could take that money. They raise — I think they just announced a new set of funds, and it’s like $6 billion or $9 billion. That means deploying that over the next two or three years. That’s a huge amount of money. You’re just not going to find returns that generate 10x at that price.

The thing that’s enabled it even more — which is why I think things have not changed — is that you have three groups of people who are a match made in heaven. Capital inferno heaven. Or maybe capital inferno hell. You have people with massive amounts of money, like the Saudis. You have funds like the Andreessens of the world willing to raise billions and billions, just mass-agglomerating capital. And then the third: the capital-intensive businesses. The OpenAIs of the world, who not only want billions to go do this stuff but potentially raise a trillion, or have hundreds of billions to go build chip factories. For the capital agglomerator, that level of capital intensity makes them drool.

The Blackstone analogy

Kyle Harrison: One of the most fascinating analogies — I wrote a piece called The Blackstone of Innovation. Blackstone is one of the largest asset managers in the world; BlackRock was an offshoot from it. Two of the largest asset managers in the world managing trillion-dollar-plus of assets.

Stephen Schwarzman, the guy who started it, has this great quote where he basically says: we were different from everybody else, because everybody else would say oh, we’re in the private credit business, or we’re in the real estate business. And he said, I’m in the asset management business. So I don’t care what it is — if I can find a person who has a strategy, I’m going to do that thing. That’s how you get to a trillion dollars of assets, because you will literally do anything.

In venture it had always been a very specific type of investing in a specific type of company. You’re typically looking for capital-efficient businesses — that’s why people love software. And even though venture capital was born out of semiconductor manufacturing in 1950s Silicon Valley, it got away from hardware for the most part. But that’s all changed. Firms like Andreessen are these Blackstone-esque businesses saying we will invest in all kinds of different types of businesses — defense, aerospace, new-age manufacturing, robotics. And in part there are a ton of valid macroeconomic drivers behind that: global conflict is coming back, manufacturing is more at risk in a rapidly de-globalizing world. Those themes are legitimate. But you also have people agglomerating capital asking where can I put it, and looking for places that can take a ton of capital. If I as Andreessen can go park $150 million a pop — it’s the same reason the Saudis love Andreessen, because they can put $300 million a pop. Andreessen loves capital-intensive businesses because they can put more money to work.

And one of the reasons the size of what you’re managing matters so much goes back to the Blackstone example. Blackstone is publicly traded. When most people value Blackstone, they don’t care at all about the 20%. My business model as a capital allocator is: I get 2% of the capital I’m managing every year to keep the lights on, plus 20% of the upside. If I raise a billion-dollar fund and make $10 billion, I give the billion back first, then take 20% of the $9 billion increase. That’s pretty good. But analysts completely write off the upside. All they care about is that Blackstone is a trusted enough asset that it can continue to collect on that 2% — like a freaking treasury bond. Just an annuity. That is all Blackstone stock is valued on.

Host: And you think that’s playing out with big VC too — at the end of the day it becomes about the recurring fee?

Kyle Harrison: They would never say that. I think it’s an unavoidable incentive — the more assets you have under management drives more fees. Andreessen is certainly doing it the best, but a lot of firms have gotten much, much bigger than ever before.

Most of these firms are doing totally fine. They’re making a ton of money anyway, they’re most of the best firms in the world anyway. So I don’t know that it’s we just want more fees in our pocket. I think it’s largely an ambition game — if I as Andreessen have hundreds of millions of dollars of operating budget, I can go hire hundreds of people. I think they’re now past a hundred investing partners, which is insane when Benchmark has five. The scale is totally different.

Flow, and “don’t hate the player”

JR Hennessy: You used this to explain the $300M into Adam Neumann’s business — a guy with a demonstrated track record of deploying huge amounts of money in a very capital-intensive business.

Kyle Harrison: This keeps coming back. The Puritans of Venture Capital is the piece comparing cottage keepers and capital agglomerators, and there’s another called The Rise of the Cash Man where I talked about the Adam Neumann thing. In all of those pieces, where I try to unveil what the incentives at work probably are, I try to illustrate that I’m a believer, for the most part, in don’t hate the player, hate the game. I don’t think the people at Andreessen are bad people. I don’t think Adam Neumann is a bad guy. The Saudis — maybe that’s a different discussion. But for the most part I don’t know that these people are bad. I think they are a product of their incentives.

On Flow specifically: he went from commercial real estate, made a billion dollars failing at commercial real estate, and segued that into residential — buying luxury apartment complexes in places like Nashville. Not New York and San Francisco and LA, but bougier areas in different cities. The idea is basically take the WeWork vision and apply it; even in WeWork they had WeLive.

When Andreessen wrote their blog post about why they invested, it was the largest single investment they had ever made — $300 million in one pop. And in it Marc Andreessen talks about the piece he wrote about it’s time to build, and building housing, and the housing crisis in the United States. Then people dig into what Flow actually is and what properties they own, and these are not solving the housing crisis. These are the luxurious of the luxury. They have saltwater pools and dog valets.

So: Saudi wealth money, into a fund that has proven it can deploy a ton of money, into a guy who has proven he can raise a shiz-ton of money — I think he raised twelve or thirteen billion for WeWork over several years. And there was a conference, very matter-of-factly sponsored by the Saudi crown, where on stage you had Marc Andreessen and Ben Horowitz and Adam Neumann talking about how MBS is the type of leader the world needs more of. That feels super weird. I don’t know that I agree with that. It’s just so quiet-part-out-loud: people with capital giving it to people who can deploy capital into companies that have proven they’re going to be capital-intensive.

Raph Dixon: Buying up residential real estate and renting it for recurring revenue — that just sounds like private equity with extra steps. It’s called VC because there’s a crazy founder.

Kyle Harrison: I’m a firm believer that storytelling is very powerful, but at the end of the day reality is what matters and it will catch up. And I’ve increasingly become convinced that storytelling shapes reality — it doesn’t matter what is real, it matters what people believe, and largely those beliefs create reality.

So to your point: on paper, yes, it is effectively just a real-estate roll-up. The difference is the story you tell and how long you can keep people believing it. You have to bridge that belief until either it becomes true, or you make a billion dollars and your commercial real estate company crumbles into bankruptcy. One of those two things happens when you’re a really effective storyteller: either he was going to build WeWork into a hundred-billion-dollar behemoth that controls the way every single person works in an office, or he was going to get out with a billion dollars.

Hopin, and the unlearnable lesson

JR Hennessy: Has anything actually changed in how companies are run, beyond the surface froth of “we’re all hardcore operators now, no table tennis tables”?

Kyle Harrison: Picture any chart you’ve seen of the broadening wealth gap — income inequality over time. You can apply that chart to what’s happening among both VCs and startups. It used to be that most venture funds had some similarities: they saw similar things, paid similar prices. Same on the company side — even though no company had a super high likelihood of surviving, they all had a similar likelihood. It was all kind of in an innocent place. Then over the last couple of years you’ve had this increasing inequality of outcomes.

Has anything changed? I think it maybe shook up who was in which bucket. Some crypto firms took big hits and had egg on their face, and that’s hard to come back from. On the company side there were a lot of companies that were the hottest thing on the planet and now very few people will talk to them.

Hopin is a perfect example — a virtual events platform. In the heat of COVID, everything’s locked down, conferences represent big revenue opportunities, and here’s a platform where you can do it. They went from maybe a couple hundred events a month to tens of thousands. From three or four million of revenue to $75 million to two or three hundred million, over the course of a year or two. And they ended up raising at a seven billion dollar valuation.

There are these quotes that are baffling to me, of people saying virtual events just went away way faster than we expected. Who thought that was going to stay? Who attended one of them and was like, this rocks, this is the future? Most people go to conferences to get drunk and not think about their work. Who wants to be sitting at home going, this is a bummer? Of course it was not going to stay the same once things got back to normal. But people convinced themselves — it goes back to the storytelling. You tell a big enough story that this is going to be the professional metaverse, and you look at their revenue, and if this continues even if it slows down a little, this is going to be an incredible business.

The same pattern: I remember Blue Apron, the meal-kit company, went public and traded down so hard it had like a $20 million market cap. A legitimate penny stock. It was bonkers. And people looked at that and said, that’s just an isolated incident. Now there are dozens of public companies that went public during the SPAC boom sitting in the three-, four-hundred-million-dollar range, and people continue to ignore those. They say that sucks, but we’ll grow faster than that, we’ll be a better company, that won’t happen to us. People seem largely incapable of learning the lessons of cautionary tales.

Short-term careerists

JR Hennessy: Canva did a big secondary sale. As time-to-IPO stretches out, is that making VCs think differently about exits?

Kyle Harrison: A hundred percent. It changes the math. This is another one of the tricky incentives.

I have a line I’m very proud of that I wrote: VCs talk about themselves as being long-term investors — we invest for ten-plus years, that’s the life of our fund. But the reality is that venture capital firms are just collections of short-term careerists who are trying to associate themselves with as much success as possible, as quickly as possible, and distance themselves from as much failure as possible for as long as possible.

The justification people made during COVID was: if Zoom is a hundred-billion-dollar company, and if Hopin can be even 10% of that, it can be a ten-billion-dollar company — so why wouldn’t we invest at three or five or seven billion? And then Zoom tanks to a fraction of that and you go, oh crap, the whole math falls apart. There was actually no math to begin with.

Host: So what does all this mean for firms still doing traditional venture, when the big funds push valuations up?

Kyle Harrison: It goes back to inequality, because it’s not that every company is super highly valued — it’s not universal, it’s increasingly centralized. The same VCs chase the same small subset of companies, and those companies see incredible explosions in valuation.

Peter Thiel has this quote about how any competitive round is by definition undervalued. If you think about it as a pure supply-and-demand calculation: Stripe at $50 billion — could you go find somebody willing to invest at $60 billion? Yes. So where does supply and demand hit equilibrium, where there’s no longer any more demand at a higher price? It’s almost never. If you ask anybody about OpenAI at a hundred billion, there are for sure people who would do it at more. So by definition a hundred billion is undervalued, because you’ve not reached maximum equilibrium. That creates the FOMO fervor of everybody tripping over themselves to give a higher price to the select group of things that appear to be hot.

Host: Is the “top 1% founder” always the same type — same schools, same companies?

Kyle Harrison: Yes and no. Broad strokes, that is true — there’s a crazy pattern-recognition concentration problem in VC, where people look for people who look and sound the same, with similar backgrounds, whether it’s the schools they came from or the companies they worked at. That’s definitely broadly a problem.

But within that select universe of people who’ve gone to a certain number of schools or had certain experiences, there are still hundreds of thousands, probably millions of people who fit that profile. If you count everyone who’s ever worked at Uber, Canva, Stripe, Airbnb — there’s still a ton of people in that pool. So the question becomes how do you identify those people?

The way Contrary approaches it differently is we want to identify people just one clip earlier than anybody else. At Index and Coatue I joke that I frequently played the hurry-up-and-wait game: if I found somebody super sharp, I was kind of just waiting for them to start a company. Because until they start a company I don’t have a product for them. I can maybe invite you to a dinner every once in a while, and that’s it.

Contrary spends, in many cases, five or six-plus years doing things for the people we’re close to. We meet people in undergrad, grad school, PhD programs; we help them get their first job, join a startup, network, find mentors. There are a bunch of different product SKUs that make us relevant in that person’s life long before they ever start a company — in the hopes that when they do start thinking about one, we’re their first call.

The problem is that it’s a crazy long-term game, and you have the risk of backing the wrong horse. We spend five years getting to know somebody and they say, actually I just want to go work at Google and make good money for a long time. Bummer.

Host: What stops other VCs doing that?

Kyle Harrison: It just takes a really long time, and it is not a normal part of the business. Going back to venture firms being collections of careerists — they want to be able to say I was an early investor in Uber, and then move on to hunting the next hot thing.

Bubble or revolution — and the answer is C

Raph Dixon: Next five years for AI. A: it’s a bubble, generative AI has no real uses and gets things wrong. B: the whole world is revolutionized and 80% of people lose their jobs. Which is it?

Kyle Harrison: I’ll tell you a story first, because it’s late my time so I get to decide when I hang up.

I interviewed the CEO of Cohere for Contrary Research — which James and I worked on together — and I asked him the same question: what happens over the next ten years? He said the next ten years are indiscernible. It’s impossible to know what’s going to happen over the next year; maybe I’ll give you the next five, and even then I’ll probably be wrong.

He focused specifically on AI infrastructure, which is what he’s focused on. He talked about how transformer architecture is what unlocked this large language model revolution, and his main takeaway was that if in five years we are still leaning heavily on transformer architecture, he would be very surprised and disappointed. So here is an architectural change that unlocked a massive wave — if we keep making those unlocks it’s pretty powerful, but it also creates orders of magnitude of difficulty in forecasting, because you don’t know what that’s going to look like.

To your question of whether it’s a bubble or a massive unlock that completely revolutionizes the global economy: again, I think it is a story of inequality. There are certain companies that will create certain things with massive implications. Not every company will benefit from that as much as they will try — they will coast.

You’ve already seen this. There have been two high-profile hiccups: Inflection and Stability. Inflection is the one Microsoft invested about a billion into, whose co-founder was the former co-founder of DeepMind. Microsoft said, hey, you should come run this Microsoft AI research lab, and he bounced. Really, Microsoft just hollowed out the talent of the company — it wasn’t just him, it was several people who got Microsoft AI offers. They took a bunch of the talent and left it a hollow shell that allegedly will at least repay investors.

And with Stability, a lot of people saw the writing on the wall, because there were sus claims about how much ownership they had over Stable Diffusion — Runway and a few others were largely responsible for a lot of that, and it mostly wasn’t actually Stability.

So there are going to be way more of those. There are a ton of companies that are not going to live up to the hype, that are going to burn a shiz-ton of money and not deliver anything. Even something like Perplexity, which has gotten a lot of hype because it’s grown so fast and people really like it — some have called it the first legitimate threat to Google’s dominance in search — they literally took the position that search deserves to be free of the ads-based model, and that went away sometime in the last month, around the same window they announced intentions to start selling ads. You can be building a computer god in a Microsoft Azure server, but you cannot liberate yourself from the fact that you must at some point become an advertising business.

Host: Is there more scrutiny on P&Ls this cycle — do companies get less time before someone asks about revenue?

Kyle Harrison: I think that’s wrong. There is just as little scrutiny in most cases. The few failures we’ve seen are unique outliers.

One thing that stirs up the dust more than in the 2021 crypto craziness is the intrusion of the big tech companies. Number one because they’re participants — Google with Gemini, Facebook with Llama. But number two, they’re massively incentivized from a capital perspective to own this world, because it’s basically an extension of the cloud wars — it’s so dependent on compute. Inflection is the example: if this was normal and they were just living on hype, they could have lived a lot longer on hype. But Microsoft said, I’m done with this nonsense, I want to build Microsoft AI, and I have a billion reasons why you should come do it over here.

The vast majority of AI hype is still in full force. It’s still very healthy. In 2022, when things changed, people said there was a tidal wave of companies going out of business — and that proved true through 2023 and 2024, lots more shutdowns, small acquisitions, acqui-hires, people failing to find product-market fit and getting consolidated. That mass extinction event for startups that people called when the economy changed — I think that will happen in AI too.

So back to your original question of whether it’s a generational economic revolution or a hype cycle: I think it is both. For a lot of companies it’s just people riding the hype. But I am absolutely convinced there will be a subset of companies that dramatically change the way we think about music, art, content — everything these things touch. It’s going to be massively powerful, and people are going to figure out how to use this stuff in production, and it’s going to be really valuable. But it is not going to be the vast majority of companies.

Raph Dixon: We said A or B and you chose C, but that’s fine.

Connections

The show

Down Round (Podcast) — James Hennessy and Raph Dixon’s Australian tech-business show. Hennessy is a former Contrary Research colleague of Kyle’s, which is why the episode has more needling than a standard guest slot.
The title is Kyle’s own phrase, coined mid-answer describing the LP → agglomerator → capital-absorber loop: “a match made in heaven — capital inferno heaven. Or maybe capital inferno hell.”

Essays this episode is arguing

The Puritans of Venture Capital — the cottage keepers vs. capital agglomerators frame, and the piece Down Round linked in the show notes.
The Blackstone of Innovation — Stephen Schwarzman’s “I’m in the asset management business,” and the observation that public analysts value Blackstone on the 2% annuity and write off the carry entirely.
The Rise of The Cash Man — the Adam Neumann piece he names by title here (published 2022-08-18, so it predates the episode by well over a year).
Institutionalized Belief In The Greater Fool — invoked by name on the Hopin discussion (“passing the bag… who gets left holding this giant pile of crap”).
Investing 101 — the newsletter, and his defence of why it reads like a VC criticizing VC.

The argument

Incentives — the load-bearing concept, and the episode’s most consistent move: every apparent villain is re-read as a product of incentives. “Don’t hate the player, hate the game.”
Power Law · Sebastian Mallaby — the 10x logic the agglomerator model breaks, and where the 1.8–2.3x / 8–13% alternative comes from.
Storytelling — the reality-vs-story reversal, stated here three days before he says it again on Sourcery and six days after publishing The Glass-Half War — Empty or Full. April 2024 is the week the idea locks in.
Secondaries — the Canva sale as the prompt for the “short-term careerists” line.
Peter Thiel on competitive rounds being definitionally undervalued — the supply-and-demand argument behind FOMO pricing.
Talent Vortex — the “hurry up and wait” contrast with Index Ventures and Coatue, and why Contrary’s model resists copying.

Companies and people