Kush Bavaria, the 23-year-old co-founder and CEO of Ornn (a company building the financial market for AI compute), joins host Harry Stebbings to explain how GPU pricing, infrastructure investing, and the future of compute markets really work. Drawing on his unique vantage point building price indices and futures for GPU capacity, Bavaria breaks down why neo-clouds are great businesses, what everyone misses about chip depreciation, and why he believes compute will become a larger market than oil.
> "I think compute is a larger market than oil will be or any of the other commodities have been for the past 100 years, 150 years." — Kush Bavaria
> "I think infrastructure investors actually end up winning because they will always keep collecting rent." — Kush Bavaria
> "I think dying will be a thing that maybe isn't like real in like a few years from now. I think like humans will be like a biological vessel for like AI." — Kush Bavaria
Kush Bavaria is the Co-founder and CEO of Ornn, building the financial market for AI compute. Ornn helps companies price, finance and hedge GPU capacity—bringing the tools of commodity markets to AI infrastructure. Kush is just 23 years old and has raised funding from some of the best in the world including Andreessen Horowitz, Thrive Capital and BoxGroup.
AGENDA:
05:00 Building a Viral Sports Betting Platform from a Dormroom at MIT 09:00 Can AI Win While Infrastructure Investors Lose? 13:00 What Is Everyone Missing About GPU Prices? 14:00 Why Is Everyone Betting Against CoreWeave and Nebius? 19:00 Are Older GPUs Really Becoming Obsolete? 24:00 Where Are AI's Biggest Untapped Opportunities? 29:00 What Would It Take for Compute to Trade Like Oil? 43:00 What Wins the AI Race: Better Talent or More Compute? 48:00 Why Raise Money When You Don't Need It? 59:00 How Do You Build a Team That Hates Losing? 01:05:00 Would You Buy OpenAI or Anthropic?
Harry Stebbings
I think infrastructure investors actually end up winning because they will always keep collecting rent. I would not be surprised if their Rubin costs are like $20, $30 per GPU hour. The easy categories are all done. I think New York's much better. We have this whole theory that people in the Valley are coin-operated. I think compute is a larger market than oil will be or any of the other commodities have been for the past 100 years, 150 years. I think dying will be a thing that maybe isn't like real in like a few years from now. I think like humans will be like a biological vessel for like AI. I think the AI will be like so much smarter than we are. This is 20VC with me, Harry Stebbings.
Now, I got a message a month or so ago from Myles Grimshaw, who I think is one of the best investors in venture capital at Thrive. And he said, you've got to meet Kush. He's an incredible talent. And I said, introduce me, Myles. I really rate his talent assessment bar. And I met Kush and I was like, dude, we have to do a show. And so joining me in the hot seat is Kush. He's the co-founder and CEO of Orn. If you don't know, Orn is the compute market building price indices and futures for GPU compute. This is an incredible analysis and breakdown on the cost of intelligence today, the future cost of chips, the future cost of compute, everything that we don't know that we need to know as we look forward. This and so much more in what is an incredible discussion with Kush.
But before we dive into the show today, founders and investors sleep on 8sleep. It is the intelligent sleep system trusted by Mr. Mark Zuckerberg, Dr. Andrew Huberman, Charles Leclerc, and the Aston Martin F1 team. The pod is a smart mattress cover that fits on any bed. It cools each side to 55 degrees, heats to 110 degrees. It even tracks your heart rate, HRV, and sleep stages without a wearable. Thank God. The new Pod 6 has a hub 50% smaller than the last one and starts at just $1,999. Visit 8sleep.com slash 20VC. That's E-I-G-H-T-S-L-E-E-P dot com slash 20VC.
While 8sleep helps you sleep better, MongoDB helps you build better. MongoDB has always been the database developers love. Well, now it's the data platform AI agents need. Agents need accurate context, fast. MongoDB stores, searches, and reasons over your data in real time. JSON-native database, vector search, and Voyage AI embeddings all in one place. One system instead of 10. No data pipelines to maintain. Ugh, this sounds too good to be true. Build and scale from your first user to billions of vectors. Run on any cloud, on-prem, or your laptop even. That's why 75% of the Fortune 100 run their most critical apps on MongoDB, moving trillions of dollars every single day. And it's why AI-native companies like Eleven Labs run 40 million agents on MongoDB. So if you're building an AI, MongoDB for startups helps you move faster with Atlas and Voyage AI credits and dedicated support. Don't build agents that answer once and forget. Build agents that remember and learn from your real-time data. Go to mongodb.com slash agents. That's m-o-n-g-o-d-b dot com slash agents.
While MongoDB scales the product, Framer scales the story. If your team wants a website that looks and feels handcrafted, but is still fast to ship, Framer is built for exactly that. Framer is the AI website builder that helps creators, teams, and businesses ship production-ready sites faster than ever while getting every detail right. Prompt, inspect, edit, and publish in one place at a whole new pace. Agents and humans work in tandem. Agents bring speed and scale. You bring taste, judgment, and control. The work lands on the canvas and stays editable. Build custom code components, manage CMS content, optimize SEO, and audit for issues all in one place. Enterprise-grade hosting, security, and 99.99% uptime SLAs trusted by leading brands like Perplexity and Miro. Learn how you can get more out of your site from a Framer specialist, or get started building for free today at Framer.com slash 20VC for 30% off a Framer Pro annual plan. That's Framer.com slash 20VC. Rules and restrictions may apply.
You have now arrived at your destination. Kush, dude, I'm so excited for this. So people don't know, but I got a text from Miles at Thrive being like, you have to meet Kush. He is one of the most exceptional founders. And I think Miles is probably one of the smartest guys in venture alongside Pat Grady. So when he said that, I was like, introduce me. And then we spoke and I was like, wow, he is as good as Miles said. And so I'm so thrilled that you would join me in the studio today, dude.
Kush
No, thank you for having me. It's exciting to be here.
Harry Stebbings
Now, I want to start with a little bit of texture on you before we just get into the meat. We've got many kind of weird and quirky things about you, with greatest respect. One that I want to start with, you created a viral sports betting platform in college.
Kush
So Wayne and I really wanted to intern at this company called Novig.
Harry Stebbings
Oh, yeah. Jacob.
Kush
Yeah. We were freshmen or sophomores at MIT, and they didn't let us basically intern there. And so we were basically like, all right, fuck you. We're just going to go do this ourselves. Which at the time, we were only good at coding. We were really bad at everything else. We were really bad at commercials, legal. We had no idea how hard it was. We were just super good at coding. So I ended up building, it was just an app that you could buy and sell. It was just like sport. It's what Novig and Kalshi and Polymark and all the apps are today, where you can buy and sell sports things. And it was like you can bet on player stats. I ended up just working with all the fraternities at MIT. So you could literally go and buy and sell different players or odds. But it was cool. It never really took off past MIT, but within the school, it was pretty sick.
Harry Stebbings
Did you tell Jacob at Novig this?
Kush
No. I have not told him that. Spoiler alert. I don't think he even heard about it because it was so like, we were so bad at go to market and all the other stuff besides building the app that it just never worked out.
Harry Stebbings
That is hilarious. So I'm going to send him this clip right afterwards. Another kind of inspirational person that you very much look up to is Travis. I heard this from some of your investors and actually some of your teammates. Can you talk to me about what would TK do?
Kush
I'm a huge Travis Kalanick fan. And the funniest thing is like a Fortune reporter asked me like, what do you think about Travis? And our media team was really mad because I just said, oh, he's a Steve Jobs in my generation. But I really think there's so much to learn from older entrepreneurs that have done this before. And I'm always like, oh, if we're in this situation, what is Travis going to do here?
Harry Stebbings
What is the result of that? Is it be more aggressive? Is it be more ambitious?
Kush
Way, way more aggressive and ambitious. I think a part of me that is really bad is not being ambitious and not being aggressive enough. I'm a very nice person. I will not buy back or push back on things when I need to. And so I think Travis is the exact opposite of that. And so it's a good thing for me to learn.
Harry Stebbings
What were you not aggressive enough on that with hindsight you're like, damn, I should have?
Kush
So many of our early commercial agreements. We sold a lot of our data early on to large hedge funds and trading shops, marketing. Those guys are like sharks. Their whole job is like, I'm just going to make as much money as possible. And we're pretty bad. So we ended up selling it for like $5,000 and $10,000 a month. And I was like, are we losing here? Because they would immediately say yes. Then I realized they're willing to pay $100,000 a month, like $500,000 a month. And I'm like, oh, wow. We are underpricing this crazily.
Harry Stebbings
You almost sound European with some of you. Like, I'm not pushing hard enough. I'm not aggressive enough. Welcome to London, baby. You are quite nuts, though, in the way that you did a marathon with no training. And several of your investors told me about this.
Kush
Yeah. So it's funny. So I used to like, I did a bunch of running in my college. I was never that good, but I ran a half marathon because my friends told me I could never run more than a mile, so I ran a half marathon. That ended up being, I ran a marathon while still in school, and then I signed up for another marathon, and I didn't know I was gonna start the company. So I signed up for it in July, and we started the company in September. And then, as you probably realize, when you start a startup, this is your whole life. There's no time to do other stuff. In marathon training, it takes two hours a day, go run for an hour, and then you have to shower and put on your shoes and all these things. It takes a whole time off your day and I just didn't have two hours. So I literally didn't have time to train for the marathon. And day comes, I'm like, all right, we have to do it. There's no like, I'm not a quitter. We're not gonna not do this. And that was the time when Jensen released a podcast where he was like, oh, we are not a car.
Harry Stebbings
Oh yeah.
Kush
So I woke him in the morning like, we are not a car. We're gonna go run this marathon. And I ended up just like, I mean, I did it. It was very bad. It was painful. But I think my time was not that bad either. It was like three hours and 30 minutes or something. So it was decent, but it was a terrible experience.
Harry Stebbings
I trained for a long time and I still did it in like three hours, 45. Dude, I absolutely love that. Well, I think that's sufficient context. When we were chatting before, there were elements that I picked up on that I think I really want to dive into straight away, which is number one statement, AI can win while infrastructure investors lose. Can you unpack like, what do you mean by that?
Kush
Yeah, I don't think infrastructure investors will ever lose. The way we think about it is infrastructure investors are like landlords for an apartment or rent. So an easy analogy is to use data centers or just apartment buildings where someone buys the apartment building and then the actual people that take the ownership of the GPUs are renting the GPUs from the apartment building or they're renting units within the apartment building. So that's companies like Fireworks, Modal, FAL, Base 10, all the inference providers, along with OpenAI, Anthropic, depending on how the deals are structured. And so you could argue that I think infrastructure investors actually end up winning because they will always keep collecting rent.
As long as the tenants, there's tenants to live in the apartment building. Now, if it turns out the city is doing bad, which is in this case, the city is AI, then that means not a lot of people will want to be renting an apartment. Or in the case that there's not enough demand for apartments and everyone's kind of looking to buy a house, quote unquote, as in I'm going to go build my own data center. And so there's all these different sorts of ways. But I don't think at the end of the day, infrastructure investors end up losing.
Harry Stebbings
I think there would be a lot of large enterprises who say there is still a remaining ROI question. For sure. And there is still a cost question. And if there are cracks in enterprise demand, that would trickle down to a lot of the providers and that would trickle down to the infrastructure investors.
Kush
Yeah, I think so for sure. I think the infrastructure investors, again, it's the same way as rent works, right? If the economy is doing bad, obviously prices for rent are going to fall down. And that means the cost of GPU hour will fall down because there's not enough demand for the GPUs. And it's purely just like, it is at the end of the day, a supply and demand game where demand, if enterprise demand continues, if it goes down, obviously supply of the compute will go up because there's just not enough demand for the tokens or the training or whatever that may be.
Harry Stebbings
How undersupplied are we given the lead time for these projects? It takes like, I mean, it's very undersupplied. Like if you want to go to CoreWeave and buy something, it's probably like you're probably quoted like Q4, 2027, Q1, 2028 at this point. If you want to do like a build your own and get chips from the OEM, the OEMs usually lie to you and will tell you that chips are going to come earlier than they are. But it's probably like 20, 22 weeks time period to actually get the chips and then install takes like another two to three weeks. And this is on a small scale. I'm talking half a year to build your own.
Are you concerned by the debt cycle that we're seeing to fuel this infrastructure?
Kush
I don't think really, because today we've seen that actual demand for tokens and the output of the AI producers actually ends up being useful. I think the public markets are way too scared about, oh, AI is not useful. I'm like, have you used ChatGPT? Have you used any of the products? Clearly, there's a use case for this technology. And which means if there is a use case for this technology, the thing that powers this is the data centers, the chips, all these things. There will be uses of these chips for a long time past the period. And prices have also gone up over time. The prices for literally H100s and even the B200s or B300s are way higher than when they started.
Harry Stebbings
I think Gavin Baker said that every single day for the last 60 days, they have increased in price.
Kush
Yes. Yeah. I think literally chip prices have been increasing. And that's because the demand for AI is just continuing to increase. That demand is exponential. You've probably seen all the revenue curves or all the companies are just like straight up. And that's literally because what do these companies do? They sell tokens. That means in order to sell the tokens, you'd either need to get way more efficient at generating the tokens or just get way more compute. And you're going to have to do a mix of both things. And part of it is, if I want to get way more compute, I'm just going to have to start paying more than everyone else.
Harry Stebbings
What does no one see about chip pricing that you see because of your very unique position?
Kush
I think the biggest thing is everyone's, I don't know, maybe everyone does see that it's just going up over for a long time. But when we started the company, a lot of people thought like, oh, the forward curve for chip prices is actually like backward dated in the sense that three years out from now or five years out from now, the price per GP hour is going to be less. And so many agreements you would do early on was like people would offer free call options to people that would buy the chips where they'd say, okay, three years from now, or I want to renew the contract for another two years, I'll just put in, I can renew it for $2.50. No one's going to renew that. And now you look at it and someone asks you, oh, can I put this call option in? So three years from now, I want to renew for the same price that it is today. No one gives that for free. Because I think people in the market see that, oh, maybe three years from now, the chip prices are actually higher than they are today. I would not be surprised if their Rubin costs are like $20, $30 per GPU hour.
Harry Stebbings
What are they today?
Kush
I think it's like seven or eight for a five year.
Harry Stebbings
Wow.
Kush
Yeah. I genuinely think the demand is insane. It's a bad thing and sort of a good thing that this is pure just supply and demand, just like markets.
Harry Stebbings
So why does everyone shit on these NeoCloud providers like Nebius and CoreWeave? And why is that debt, like bonds being through the roof?
Kush
I mean, I think it's because they're levered to the point that they believe that the residual value of the chips and the residual value of AI will continue to progress over time. And people just think, oh, this is a bubble. This is not real. And that kind of stacks on and on and on to each other. But if you actually look at the data, the price per GPU hour is still increasing. So I don't understand why people are like, oh, we need to go short CoreWeave or Nebius. Part of it's also a credit question. And the tenants of CoreWeave, will they be able to pay their rent for the GPU hour? And that in itself is like, it's an AI question because the tenants are AI companies.
Harry Stebbings
Sure, this is your files and your base.
Kush
Yeah, it's like, will they be able to continue to pay the rent for CoreWeave over time?
Harry Stebbings
Totally get that. So you would buy CoreWeave and Nebius?
Kush
I would go long CoreWeave and Nebius, yeah. I'm not allowed to publicly invest, but if I could, I would go very long.
Harry Stebbings
So I love it when you have a very clear, we believe that AI demand will continue. We believe that there'll be a series of knock on fire. Yeah, what's the bad case? We've got the bull case there. What's the bad case?
Kush
I think the bear case is that AI demand doesn't continue. The other option is like, okay, people stop buying tokens. Enterprises don't see that AI is actually impacting their business or helping their business, which I think is also partly true today. If you look at all these studies from the consulting firms, I used to work at BCG. I saw, like, quote their study.
Harry Stebbings
You used to work at BCG?
Kush
I hated it, yeah. That's why I started the company, partly because I didn't want to work a corporate job. But I think partly is like, when you look at all the studies, it's like, oh, how did the enterprise actually implement AI? Are they being useful with AI? And I think the answer so far has been like, not yet. I don't think it's been very useful for many of the large companies so far. And maybe that's because it's an implementation problem and all these things. And I think demand is still increasing. It's not a problem. But let's say it doesn't work out. Maybe Coca-Cola or McDonald's or many of the large Fortune 500 companies realize like, oh, AI doesn't really change my business for the most part. And in that case, it's like, okay, then all these chips and all this compute I bought, and you're kind of overbuying. You have to buy before you know there's demand for it. It doesn't work out. And then the people can't pay the rent. And then you start defaulting on these chips. And then creditors basically have these chips that are worth what? They're worth the metals in them, or they're worth something in them that is not the actual value that you can plug it in and use it for.
Harry Stebbings
How do you advise founders on forecasting the buy of compute? I asked this from Eric at Modal.
Kush
Yeah, I watched it. So what Eric said, he buys like 18 to 20 weeks in advance. I think he should start, he's probably going to start buying like 22 to 26 weeks now, just lead times increase for everything. But that's really the case. So part of the company for us is to help solve this problem. It's like, okay, people need to buy before they know what's going to happen. So that needs, you need to hedge your risk somehow, right? There needs to be a similar way. If I'm buying today, I don't actually know what prices are 22, 26 weeks from now. Maybe they're higher, maybe they're lower. Who knows?
It's a long time. But I would genuinely say if people have the cash, and many of these Fortune 500 companies especially, if they're like, oh, I'm going to invest in AI and I have the cash, just go buy the chips, go build the data centers, or go do something, because I think prices will continue to increase and you're just going to be left paying more and more every single time. Let's say you don't use them, then you can just do exactly what Elon does. We allow you to rent the chips through us, but even if we don't want to go through us, do exactly what Elon does. He goes to SpaceX and he says, oh, I have extra chips.
Anthropic, pay this much. Hey, pay this much. And so you can just start becoming like a NeoCloud and start charging as much as you want for the chips because you have the money and the credit for them.
Harry Stebbings
Elon has such a good business being in NeoCloud. It amazes me he spends so much time building GrokBot and everything else.
Kush
I mean, look, I think their models are pretty good too. We are big users of Cursor and GrokBot and all the.
Harry Stebbings
Are you locations?
Kush
Yeah, our team, I a lot like. So I'll walk you through our stack.
Harry Stebbings
Yeah, we want to.
Kush
Okay, so we use GrokBot for all the daily tasks. I don't have an assistant or any of these things, just pure GrokBot. And it works well. It works really well.
Harry Stebbings
Have you tried it?
Kush
No.
Harry Stebbings
Okay, you need to. I use Instinct every day religious.
Kush
Oh, Instinct's great, too. So my co-founder uses Instinct.
Harry Stebbings
Fucking great. Yeah. So good. It's all the same sort of similar things.
Kush
Yeah. Yeah, actually, maybe I should switch to Instinct. I pay for GrokBot. It's like whatever. But anyways, I use GrokBot. Then we have Cursor. And we all started using Factory for coding. I think the Factory part that's really good is that you can bring your own key. So we have a bunch of credits because we're a startup from Claude and OpenAI. And we just keep getting credits from every VC. And so you're just using bring your own key for that, which has been very helpful. And they're like $200 per month plan is basically unlimited tokens. Matan and Ina are amazing.
Harry Stebbings
Yeah, and I also think the app is better.
Kush
We used to be on Devin for a while and Cognition. And the problem with Cognition was it was just so expensive. I don't know how, but the tokens used on Factory end up going 10x higher than they do on Devin. And it's kind of like this look. They're like competitors in the same product. But my feedback to Devin is be like, why does it cost me so much to use the product itself? Factory is just so much cheaper. And I think also the underlying harness of Factory somehow is better than the Devin one.
Harry Stebbings
I'm thrilled to hear that. As a Factory investor, I completely agree with you. I just give them an ad. Going to another one of your statements, which you kind of touched on, but you said older GPUs aren't necessarily obsolete. What do we not know about chip depreciation?
Kush
I think it's very simple that CoreWeave has also said their A100s are still being rented out for a three-year period, and the H100s are also being rented out for a very long time. People thought that they depreciate. If you asked this question last year, people would have said, oh, these chips are getting depreciated tomorrow. It's like a next day sort of thing. But now people have realized, oh, wait, the H100s are still renting out for three years and these chips are five years old. The A100 is, I think, actually, H100 is four years old and the A100 is six years old. And they're still being rented out. So it's like, oh, OK, maybe it's longer than we actually thought. I also believe that the chip depreciation, the older chips will have a use besides being frontier model training. They'll be used for lower scale inference, smaller model inference. And if you believe inference as a market is going to split up between all these different categories, like fast inference, slow inference, specialized.
Harry Stebbings
Fast inference, slow inference, specialized. What do these mean?
Kush
So I think fast inference is that people want their answers really quickly. And so people just want highest tokens per second on their inference. And so there will be a company that specializes in fastest tokens per second. There'll be a company that specializes in cheapest but slowest tokens per second. And I think there will also be a company that specializes in biological inference. Or like FAL, in a sense, is inference for video or image generation specialized inference.
Harry Stebbings
That's so interesting. And so you don't think that it'll be a horizontal provider, a Fireworks, a Modal that win?
Kush
I think it'll be specialized.
Harry Stebbings
Really?
Kush
Yeah, I think it'll be specialized. I mean, maybe Modal gets really good at fast inference and image generation inference. And they'll own those two verticals. But I think it's like you own one part of it.
Harry Stebbings
You said there with FAL, image and video. What other mode requires specialized?
Kush
I mean, I think biological like people in physics as well. Like these inputs are not just text to text, right? It's DNA or proteins or some chemical formula to some other chemical formula.
Harry Stebbings
Can I ask what market composition do you think the inference market takes over time? Will there be three or four big providers? Would it be an Uber Lyft? Would it be an AWS GCP?
Kush
So I think there will be a horizontal provider that is very good at providing base level inference. That will be the everyday person's inference company. But then there will be a hundred different specialized inference companies that exist. There's this joke on Twitter that there's a thousand companies that do inference today. I believe all of them will survive. They'll just specialize in one specific thing. And no one's kind of found it. So everyone's like, oh, we're doing this type of inference or this type of inference.
Harry Stebbings
Do you not think it's a moment in time, part of the cycle, where there's a lot of new people that come up and you can grow very quickly, but they'll consolidate as markets do and the actual number of players shrinks?
Kush
I think the actual number will shrink, but at the same time, it will be verticalized. Maybe I'm wrong about that. Maybe it doesn't verticalize, but I believe that it will verticalize.
Harry Stebbings
Can I ask you, late cycle momentum investing in AI, for me as a venture investor, it has never been more binary. I tweeted this last night. You either get a round done in two hours or not at all. What are your thoughts on late cycle momentum investing in AI?
Kush
I think it's a lot of hype. If you can get a round done in two hours, clearly besides a lean investor, no one else is actually understanding or looking at what they're doing. And so I think it's a lot of fear-based investing. A lot of it, I believe, and maybe this is not true about all investors, is people didn't get into OpenAI or Anthropic early enough, and now they're like, okay, it's a $1.5 trillion IPO or $2 trillion, whatever the number they're like, OpenAI is going to get it to. Anthropic's like $1.5 trillion. And I missed out on that company. And so people are now scared that, oh, I'm not going to miss out on the next OpenAI or the next Anthropic. And so everyone's kind of just putting as much money as possible because no one wants to miss out on the next large company. It's a fear of missing out. FOMO is really what it is. It's just FOMO investing. But it's not FOMO investing in the sense that VCs are like, oh, a tier one VC led it. Everyone, all the other VCs try to go after it. It's FOMO investing in that they see the public markets like, oh, wow, like I need to start investing in this.
Harry Stebbings
Do you actually think VCs add value?
Kush
I think some do. Some don't, but some definitely do.
Harry Stebbings
Eloquently done. What is overheated then? If you look at a space where you're like, oh my God.
Kush
It's like for me customer support. I'm like, what the fuck? 15 providers.
Harry Stebbings
Yeah, I mean customer support. It's probably maybe one of them.
Kush
Part of it's like the specialized things. A lot of people are like, oh, we're Roga for this or we're Harvey for this. I think those categories are definitely taken at this point. I've talked to Gabe a few times. I mean, Roga's killing it and Harvey's killing it and Lagora and those like the easy categories are all done. I don't think there's, you're not going to compete against Roga. You're not going to beat Harvey at legal. If that is a category that exists, you're not going to beat them at it. And maybe it is that OpenAI and Anthropic beat them at it. But if you believe these things will be specialized over time, there's no room for those. I think that's part of it.
Harry Stebbings
Okay. Where is underheated? You're a GP at 20VC today. Well done. Where do you want to go and invest as a space?
Kush
As much of the energy and sort of physical AI stack as possible. That's not even a crazy take. I think that's a normal take now.
Harry Stebbings
But let's go deeper into that. What is that in particular?
Kush
The data center space. That's probably because I'm in the industry, but I've realized over time that there are not enough people that are building out products for the neoclouds or the people building infrastructure. Crusoe is such a good company because they build out all of the actual electricity infrastructure and they build out the transformers and they vertically integrate very well. So if there's a company that can build transformers tomorrow, the startup, like, I'm going to build better transformers, I would invest in that company.
Harry Stebbings
Do you see that often?
Kush
No. It's a kind of boring thing. That's why. Or maybe it's like people are interested enough or they don't realize the market for it. But I think if there's companies that are very specialized in building out specific products or tools, that as a market doesn't look as big. If you look at the transformer market, you're like, oh, how big can this be? But if you think data centers are $7 trillion, a lot of that has to go into the physical building blocks of this.
Harry Stebbings
Do you miss out by not being in the Valley?
Kush
No, I think New York's much better.
Harry Stebbings
Why?
We have this whole theory that people in the Valley are like coin operated. And this is true. If you look at it, if you pay a researcher 100K more, they're joining your company. If you give them 50 bips more equity, they're joining your company. In New York, there's loyalty in people. There's like this culture of like, we are so loyal to like the people that we work with and the family. And like, there is much more of a commodity between like the people there compared to SF. Do you think that's true?
Kush
Do you know what? I think there's just a lot more group thinking in Silicon Valley than there is New York. It's so insular. It's so incestuous. And it's like, oh, my God, I'm going to FF and then I'm going to Bessemer. And I'm not saying I wouldn't take a tier two because I want the price. But it's just like, fuck off.
Harry Stebbings
Yeah, no, no. I'm a real... That was a great accent. I just can't. I can't stand it. But I'm going to be totally honest. I'm going to get in so much trouble for this. That's why I don't particularly love the American way of doing business, which is like everything is a transaction.
Kush
Yes. And actually, that is not how I hate that.
Harry Stebbings
Yeah, I just want to do amazing work with people I respect. And sometimes there's no actual money to be had.
Kush
Yeah, I think part of like when we started the company, it's part of the reasons like, oh, we're on this like market or all these things. But I get to work with like my four best friends from school every single day. We all went to college together. We're all in like fraternities together. It's like there's no better life that I would like live besides like working with these people. And that's like it's like a very like common reason like i think people start companies for all different reasons but there's no other place in the world where i get to work with my four best friends.
Harry Stebbings
I totally agree with you. I think it's really romantic and sweet. So why did you start the company?
Kush
Part of it is like I think throughout college if you ask like my friends or like investors or other people they'll say like oh he was like definitely gonna start a company and they're like i did all these jobs and like it was like whatever but it was really like I spent a summer at VCG, and I was on the private equity team, and I actually learned a ton throughout it, but I was coding, and ChatGPT had just come out. And so what I would do is I'd figured out ChatGPT was very good at making Excel formulas and all these things. So you could literally, the MD would give me a task, like, oh, go organize these filings.
Great, I'm going to take ChatGPT, give it to that. It would come out with the Excel formula. You paste it into Excel, done. I would do it before the VP or any of the senior people could do it, and people were like, wow, this guy is good. But it was not that I was good. It was just that I was just so good at using ChatGPT to figure out the coding and the formulas. And then I was writing Python code. So I was a CS major in school. So I could code, and I was writing Python code for them. And they were like, oh, wow. I've never seen someone that could do this. Because they're all consultants. And at the time, I was like, yeah, I just can't do this.
It's a full-time job. So I did a bunch of things.
Harry Stebbings
They must have been really sad to lose you.
Kush
Yeah. And I actually told them no. And it was funny. The recruiter called me in like March of this year, like April of this year after a16z did our round. I was going home for like a few days and they called me and they're like, hey, like why aren't you in the office? It was a Monday. I was like, what do you mean? Like I quit like two years ago. I was never coming back. And they're like, oh, like apparently it says you're coming back and all these things. And I was like, yeah, sorry, like I am doing like I have a whole different startup now. So it was a very funny experience. Yeah, it was a very funny experience.
Harry Stebbings
That is hilarious. So if we get into it, why does compute actually need a market? It is purely just supply and demand. That's how prices work for compute.
Kush
In the sense that there are many suppliers that sell basically the same good, which is an NVIDIA GPU. And there are many demanders who are demanding all the same good with different parameters associated with it. Some people need the location to be this. Some people need like three terabytes of RAM instead of two terabytes of RAM. But it's all kind of in this like same sort of bucket. And if that's true, that means there needs to be a market for this because there's many suppliers supplying the same thing as many demanders demanding the same thing.
Harry Stebbings
And that's just purely a market. If it's simply supply and demand, is there no moat in this market?
Kush
I don't think neoclouds essentially have a moat. And people are trying to find a moat today.
Harry Stebbings
Why do you think like CoreWeave is like, oh, we're doing post training, oh, we're doing like tokens? Well, they have to expand product to gain moat.
Kush
Yes, that is what everyone's doing. They realize like early on there is no moat, and so now everyone's trying to expand to find some moat.
Harry Stebbings
Is that paradoxical to what you say earlier though, which is like, I would buy them, I would be long, and then they have no moat.
Kush
Yeah, but that's, I think that's fine. I think they're still, like, they're all good businesses. I want to be like the landlord. Like, that's a great business to be in. But yeah, I mean, I don't think not having a moat is a... I think the moat for the NeoClouds is like, I own the infrastructure.
Harry Stebbings
I completely understand you. We have demand, we have supply. Why have we not had a market for them before?
Kush
Partly because the suppliers, there weren't that many. I think early on in CPU land, there's only three suppliers, AWS, GCP, Azure, Oracle maybe. Now there's 300 NeoClouds.
Harry Stebbings
Okay. And so we have massively fragmented supply side, and that means that we don't have a market?
Kush
No, the fragmentation means that there is a market for it because there's so many different suppliers for it. I also think on the demand side now, there's 100x more demanders than there were for CPU products. And they're all kind of asking for the same thing, all trying to go after the same goal division, like, oh, I'm going to build this, I'm going to build that. And so that creates literally like a way for people to bid on the price. GPUs today, every all the neoclouds have now started what do they do now? Auctioning for the chips, which is like it's a market.
Harry Stebbings
What has to happen then if we think about it compared to a traditional oil market? What has to happen then for compute to trade like oil does?
Kush
I think it today it will trade. A lot of people, like we build index products for compute itself, and then we do that by running like a physical market where you use the data to actually build the index products. And similar to oil, where if you look at Brent in the Nordics, people always see how much of Brent is actually of the global oil volume is Brent. And it turns out, I think it's like 0.8%, 0.7% of the actual global oil volume, which is such a small fraction of the total oil that exists in the world. And if you think of compute, it's very similar to that. Because compute, there's all these different flavors, these different gradings of quote-unquote crude oil of a B300.
But what you really need to figure out, you need to track a very specific type of that compute. And that becomes the index, so that becomes the default of which you track. It doesn't have to be all of them, but it just has to be a very specific type. And then that becomes a reference price that everyone sort of mentions, and that becomes the sort of standard, the gold standard of what compute is. And I agree, I think Eric said probably last week or a few days ago, he's like, oh, compute is like, oh, there's all these different flavors. I totally agree with them. There's all these different flavors.
It doesn't mean that we have to somehow figure out a way to standardize all of them. Saudi oil is not the same as Venezuelan oil. And they all trade still, like everyone still quotes off of like the Brent price or the WTI price or whatever that may be.
Harry Stebbings
So what is it then that is required to make it trade? It's like a standardization.
Kush
It's picking a very specific type of the compute and working like just getting that compute standardized.
Harry Stebbings
How much of it is trust and relationships?
Kush
I think a lot for the trading part of it, it's fully trust and relationships.
Harry Stebbings
What if it's as numerical as demand and supply and auction?
Kush
It is so it's partly that, but what is the moat that S&P has or what is the moat that like a Moody's has? It is like trust and long-standing business.
Harry Stebbings
I totally get you. What is different about your business compared to an S&P and Moody's?
Kush
Building in a different market, more earlier, and we're faster moving and hungrier. That's what we say. But I think reality is like we say like we want to be the S&P for compute. That is like the part of like the like that's the scale we want to get to an even larger scale than that. And part of that's because I think compute is a larger market than oil will be or any of the other commodities have been for the past like 100 years, 150 years. And we really want to build the business in the same way where they run the physical exchange and they use that data to build out the actual index and then you build a bunch of data products on top of that and then you do ratings and all these other things on top of it.
Harry Stebbings
Or pause, pause. You meant that you're expanding product suite there. How do you know what to start with? Why do you choose that and why do you choose what's coming?
Kush
Part of it's we started with index product because we think compute needs to be traded on a public market and there needs to be futures, options, all these sort of financial derivatives that sit on top of a GPU.
Harry Stebbings
Okay, so it's indexed on top of GPUs, product one.
Kush
Yes. And in order to get that data, you need to run like a physical exchange, right? A spot, like you need to run Brent in the Nordics for compute. So we built that out and that requires like figuring out, oh, how am I going to get chips? So we obviously don't have a ton of cash and instead of building a neocloud, it's like how do I get other people to give me cash to like, can they invest in chips? And then I monetize their chips for them, sort of running like an Airbnb style model. And so we started doing that where we find other people to do it and we'll do a sale lease back or all these sort of different financial ways. And then a day, we will control the chips, but someone else ends up investing in the chips or investing in the data center project and all these different ways to do it.
Harry Stebbings
What did you not expect that you have come up against and found challenging?
Kush
Going like this deep. I never thought we would have to like go all the way to the chip level. I thought we were just going to be like a software data company.
Harry Stebbings
Why do you have to go to the chip level?
Kush
There's no one that's going to give you their data or standardize it for you. So in order to get the data for the index, you have to either go to like a bunch of NeoClouds and say, hey, please give me your like transaction data of what you sold this for.
Harry Stebbings
You think CoreWeave was going to give you that?
Kush
Hell no. Like they're going to be like, this is the most like valuable thing in the world for me. I'm not going to hand this over to you. We tried for like six months. Like, oh, can someone please like give me their data? And we had some, to be honest, we had some people actually give us their data. But at the end of the day, we want to build a very robust market. And part of that's like, we need to have billions of dollars of data. And having billions of dollars to compute, you have to either go to a neocloud or go do it yourself. And we have this concept of like, no one's going to come save us. We got to save ourselves. And part of that's, we have to go do this ourselves. And I didn't realize, like, you asked me a year ago, would you have to do this? I didn't realize we'd have to, but now we kind of do.
Harry Stebbings
Does that inhibit the speed of scaling?
Kush
No, because you can just get other people to put their money. I think there's a ton of people that want to invest in data center projects. They just don't know how to monetize it. They don't know how to do all these things. And so you kind of just teach them, like, okay, if you put this much money here, this is what the IRR looks like. This is where you can return it. I will control the chips. I will manage it. I'll run it for you. You just put the money here. They're like, great. It's kind of the model.
Harry Stebbings
Yeah. It sounds very simple. Obviously, there's so much complexity to it, but that is like the simplest way to explain it. No, I totally get that. Who will be the first mega traders in compute futures?
Kush
A lot of it I think it is the people that are exposed to the price of compute. And the way they're exposed is they're exposed to the credit risk of the people buying compute. If I lend out my compute to a startup that has just raised, a Neolab, raised like $100 million, and the compute they're buying is really like $250 million, my bet as a lender is that, okay, this team is going to keep raising venture dollars, hopefully, and that venture dollars will be spent on this. Today, many of the Neo Labs don't make money. So it's not like they're going to generate revenue or profit to hold this. Or my bet is that the NeoCloud somehow, the price for compute goes up and now they bought it for very cheap and they can sell it for higher and then they're going to pay me back.
So you're really exposed to the company itself that you're lending to. And if they default, that means I have to go in the market and now sell that compute for a certain amount of price, which could be lower, which could be higher than what it was before. That's like reason one. Reason two is that no one knows what the cost for a GPU hour is on the depreciation cycle. And a depreciation cycle is like, okay, three years from now, what is my cost per GPU hour going to be? I need to basically get to four years to pay back my loan, or I need to get to five years, but I only have a three-year contract.
So you're kind of exposed at the end of it. And maybe your contract is all the way to three years, and it covers everything, and you get the lenders, and you have the chips. I still need to make my money three years to the end of life, and I can lock down my money now. Way farmers used to do this back in the day, if you look 100 years ago, what did farmers do? They needed to plant corn. And to plant corn, or whatever that product was, they needed to sell futures back in the market because I was at a plant today and I only realized my revenue a while back after this. And I don't know what the price is.
And if I make a loss, I've just wasted like six months of time and money and all these things.
Harry Stebbings
You're only as good as your underlying tenants, as you kind of said that. With that in mind, is Meta's bond too expensive? The health of their business is phenomenal. Anthropic's bond price, which I think Blackstone is kind of the regulator or the provider of, is actually not amazing. These are both fundamentally good businesses. Are they super expensive?
Kush
I think Meta's is like getting, I mean, CoreWeave's is way higher than Meta's is, to be fair, but I think...
Harry Stebbings
But it doesn't have such an underlying good business.
Kush
Yeah, yeah, yeah. Meta's is partly like, I think it's getting more expensive just because people are just a little scared in the market. And also the amount of growth that money or people are putting into the infrastructure. If you look at all the hyperscalers, they're literally taking every single dollar they make on their regular business lines and just putting it straight into data center capacity. And so people are like, wow, I've never seen this much money being put into a project before and a lot of public market investors. And you're moving from positive free cash to negative free cash at Google and Facebook, which is alarming. And it's partly like people in, I mean, I live in New York, but people in New York don't realize how much AI, what really is this thing? Is there really demand for this? And people are just getting slightly concerned about that.
Harry Stebbings
Can I ask you, and this was a brilliant one I think from Marc actually, but he said when you look at the index today, does an index drive standardization of compute clusters and how do you think about that?
Kush
I think the index does drive standardization of the compute clusters and it's a good thing because the way standard... I mean NVIDIA themselves will say NVIDIA clusters are fungible. What they mean by fungible is that if one tenant backs out, someone else will take the same cluster. It's not that you're doing a very specialized build for every single person. The index itself will drive standardization, but I think it's a good thing for the credit markets or it's a good thing for any of the lenders because if the product is standardized, that means like if one customer defaults, someone else can move it. If I have an apartment that all looks the same for every single building, that means it doesn't matter if like I lose it, someone else can just move in because you're just going to get the same thing.
Harry Stebbings
Yeah, and I guess it would remove the discount rate.
Kush
Yes.
Harry Stebbings
That's interesting. Okay, I didn't see that before. So given how much time you spend with people who buy and sell compute on a daily basis, what do you see that you don't think other people see enough?
Kush
The prices people are willing to pay. Now it's just like caught up way higher.
Harry Stebbings
Are the buyers changing?
Kush
The buyers are definitely changing. It's going way more to, as to your point earlier, a Modal, Fireworks, a FAL. It is going to people that actually have the cash. Val just raised like $1.6 billion. Modal also raised, what is it, $800 billion? Or $800 million, something like that?
Harry Stebbings
I think they raised $750, yeah.
Kush
Yeah, $750, all that money is just going to compute. And that's way more than all startup has. So part of it for us, our job is making sure that the small startup still has access to compute while feeding these larger customers.
Harry Stebbings
I totally get you. Are you able to do both at the same time? Does it make sense to do both at the same time if Modal can just pay you $100 million?
Kush
It does, because without that, there is no market. You have to somehow seed liquidity in the market. If Modal is the only one that buys all the compute, it's going to consolidate. No one else gets to access it.
Harry Stebbings
What's the hardest thing about seeding liquidity?
Kush
You have to say no to a Modal. You have to say no to a Fireworks. You have to say no to a FAL, which is willing to pay you gazillions of dollars in revenue and profit for your underlying business to kind of look in the long term. It's like short-term game versus long-term profitability.
Harry Stebbings
How much is a market ultimately worth? If you think about ICE today, and when you look at Ornn in five years time, being the kind of market for compute, how much is that actually worth?
Kush
So I think partly it's like we want to be like we're a data and index company. And I think the market is part of it. And like a lot of the team today focuses on the market of it because that's how we actually build the data and index products. It's very like, it's funny. If you look online, all these things, like you're like, oh, it's like a data and index company. That's true. And that's like five people on the team, six people on the team. But in order to get that data, that is from the other 30 people on the team to actually make this team look really good. I mean, ICE is a $100 billion business. CME is a $120 billion business. S&P is worth more than both of those. I think they're like $150, $130 billion business. And so these are all $100 billion businesses plus. And they built their businesses off of electricity and oil and traditional commodities. But if you think compute is going to be the next largest commodity in the world, it's going to be much larger than all of these. So the market opportunity could be in trillions. Especially as we move forward to an inference led market.
Harry Stebbings
When does inference, not training, set the forward price to you?
Kush
I think it does today. I would argue many of our customers today use a lot of their compute for inference itself, that they're just like buying compute to serve tokens off of it. But the good thing about the inference demand is it's not speculative. It is real. The tokens have to be used by someone. Maybe it's a startup that uses them. Maybe it's an enterprise that uses them, but it's real. Someone is paying on the other side. Training is much more speculative, right? It is kind of like biotech, where you spend all this money developing this drug, and you don't actually know if it works out or doesn't work out.
Harry Stebbings
So if you think about misprice risk in the market today, where's the misprice risk in the compute market?
Kush
I would argue that it is a lot of these neoclouds that are not actually looking at the tenants that they're lending to and just saying, okay, great, this company's raised like 100 million, have the compute. And they don't actually do the enough diligence before lending out the compute. And inevitably, the way the world works is many of these neolabs, many of these companies will end up failing. Startup success rate is very low. And today, all boats rise with a tide is the saying. And that's kind of true today. But at some point, I think a lot of these companies will not succeed. And these neoclouds will be left exposed without a way to hedge.
Harry Stebbings
I'm really sorry. Given the demand access, will they not just fill it instantly?
Kush
They will, but part of it's like, which model is going to win, right? There's a lot of Neolabs that are all going after the same thing. And you kind of have to believe some will do well, some won't do well.
Harry Stebbings
What do you think will determine the winners in Neolabs versus not?
Kush
Talent. I think everyone has the same access to the tools. Also, I think it's talent and compute. If you have more compute, that means the models will get better. It's a very known thing. More compute, better models. Better talent means also better models. So it's a combination of both things. And you can maybe even argue that having more compute is better than having better talent in the company itself. It's such a big moat. If you have the compute, you just kind of like...
Harry Stebbings
Why do you think like SpaceX balkers? If you could have more compute or better talent, what would you choose?
Kush
More compute.
Harry Stebbings
Wow.
Kush
I think that the models are good enough to do like RSI. So you just kind of just need compute. Or they will be good enough in like three months from now.
Harry Stebbings
You know, I think everyone is so challenged and the markets, Shopify dropped 15% after a tiny miss in earnings, like tiny miss. And everyone is just like unsure of the five year. Like for the first time in a long time, everyone's like, I'm not sure where we're going to be in five years across so many different markets. And that makes pricing anything just incredibly fucking difficult.
Kush
It just increases volatility.
Harry Stebbings
Yeah. That's what it is. So the options are honestly like you could argue are underpriced. And my question to you is, do you believe that things will be as significantly changed as people think they could be? Or do you actually think that things take longer to adjust?
Kush
I think the enterprises will take longer than people think to adjust. And then the world will change faster underneath them. I think the San Francisco bubble will change way faster than the rest of the world will. The rate of progress, the San Francisco bubble is like the AI world, is what I call it that. And is that the model, the capability of the models would be so much better and so much like it'll be exponentially better while the world is still like stuck in like a very linear growth curve because the enterprise is like, oh, I have this technology. Like how do I put it in my business? Like Coca-Cola is probably wondering like, oh, how do I put like AI in my business? Like I have to use this somehow. And people haven't really figured that out.
Harry Stebbings
What do you believe today that everyone thinks is pretty fucking kooky that will be very common in five years?
Kush
I think like humans will be like a biological vessel for like AI. I think the AI will be like so much smarter than we are that the role of like humans is to like basically not even control the AI, but like teach it like biological like feelings and emotions and all these like things besides like being the sort of parent of the AI. I think also the longevity stuff. I really believe like I think dying will be a thing that maybe isn't like real in like a few years from now. I believe like the drugs and the technology of like biotechnology will be so much better than it is today. Which ends up meaning like the average lifespan for humans will go from wherever it is, like 80 something years today, all the way to like 150, 200 years. And the quality of life will also increase. It's not that when you get to like 80, you can't do anything. You can't like stand up and we're like 100, you're just like sleeping all day. I don't think it'll be that. It'll be like you're living the same life I'm living at like 23 today that you will when you're like 80 years old.
Harry Stebbings
Well, Brian Johnson would be very proud. Can I ask you, what revenue line do you not have today that you think will be very prominent in five years' time?
Kush
I wouldn't say that we'd not have, but I think debt ratings for data centers will be large for us going forward.
Harry Stebbings
Sorry, this is where you are the certified provider of a debt rating? Moody's for data center debt. And the inherent value of that is for all lenders, then you become this trusted...
Kush
Yeah, and part of that's, do you think Moody's understands how data center debt works? Who knows? Do they really understand the credit risk that Modal is having on their people that they're buying the infrastructure from? I don't know. And it's funny that I say it's like Moody's for data center debt ratings because I just think it's just like data center debt ratings. And Moody's has made such a good name for themselves in like ratings in general. So has S&P and many of the other businesses. But I think for us, like we want to get into ratings soon. Not today, but soon.
Harry Stebbings
Do you find it difficult to retain focus?
Kush
Yes, there's so many things to work on, which I think a lot of companies just like, just keep hiring more people and just go into all these different verticals.
Harry Stebbings
If you were unconstrained by resource, what would you do that you're not doing?
Kush
Everything. I would just try to do as much as possible. Honestly, we would start building out literally physical data centers. We would just go full, tell Crusoe we're competing with you.
Harry Stebbings
You're seeing the inference providers do that, your Fireworks of the world do that. Do you think that will be a successful play as they eat more and more of the vertical chain?
Kush
I think they will, but they're going to be way slower to do it than the other neoclouds are. The NeoClouds have gone through this cycle, I think two cycles now, deploying chips, learning things, deploying chips, learning things, deploying chips, learning things. As a company, we're basically halfway through a cycle of this. I think the more cycles you get through, the more you learn and you figure out things. I think they will be successful, but it's saying that, okay, are you going to be able to beat CoreWeave in their own inference game? Are they going to be able to get to inference faster than are you going to be able to get chips? And I think I would take the side of like a CoreWeave or an EBS and say, they're going to be able to get to the inference before you are going to be able to get to the chips.
Harry Stebbings
I said to you about if you're resource unconstrained, you kind of are resource unconstrained in cash. I mean, you've raised when you didn't need to. Why raise?
Kush
Part of it's getting the right partners around the table. So the way we build a business is like we want to build this as like a family business. We believe it is truly like we are a group of friends building a business together that will last for generations. It's a sort of idea. There's different ideas to build a business. It's not like I'm going to build a business for five years, then I'm going to go sell and make a ton of money. This is supposed to be a long-term play for us, and we want to build it a very long-standing business.
Harry Stebbings
Why the name Ornn?
Kush
I used to play League of Legends, and it's just a character. And I think it was a very cheap four-letter domain.
Harry Stebbings
Yeah, that's the honest answer. No, dude, that's fantastic. I, you know, sometimes you think like it's that Greek mythology and it's kind of very thoughtful.
Kush
Well, League of Legends works better. And yeah, and we literally so what I did is I realized like Andrew and Palantir all from like Lord of the Rings. So I went through the list of like all characters or like people from all those sort of like cool like games. I was a huge video gamer so I went through all the game things and League of Legends, you can go through all the players. And I found out what was the cheapest four-letter domain. And there's a study that startups with shorter names are five times more likely to succeed. And so I was like, okay, so we need to go five letters or below. And so I just went through all of them and I found that, oh, this works. And so it was just a cheap four-letter domain.
Harry Stebbings
You chose Miles, one of my favorites. Can you just take me to why did you choose Miles?
When we met Miles, the first like 45 minutes, he literally didn't care about the business. He just talked about us as people. And so I know when he invested, he cares about the business and the idea and all these things. But he cares more about us as people. How do you think that's so interesting? Because I also spoke to the Vine guys. I think it was Alex. That was wonderful. Yeah. Who said that he went for dinner with you with Adam Valkin, and Adam was trying to talk about tennis with you because Adam plays tennis and you play tennis, and you were kind of honestly quite uninterested and you were like glazed over, and then you were like, "Okay, enough of the small talk, let's get to business." And so I'm just like, that's intriguing because that's seemingly different.
Kush
Yeah, I think. Were you just not feeling it that day? No, no, no. So the story is my co-founder Wayne is like a pro tennis player, and Adam also played. Wayne played at MIT, Adam played at Harvard, and so they started talking about it. Alex and I were just sitting there, and this was a 10-minute conversation. We're like, we don't even know where to jump in. And at the time, I had another VC meeting right afterwards. And I genuinely, the way we met Vine was Alex LinkedIn DM'd me. "Hey, wanna chat?" Called him. I literally didn't take a Zoom call because I was like, this is probably not worth my time.
I called him on the phone. I was like, "Oh, this guy's pretty good. Maybe I should talk to him." He's like, "Oh, you wanna go for dinner at this place?" And I was like, "Oh, wonderful." It was like a one-minute walk from my apartment. And so I had gone, and I had another big VC meeting, think large fund, that also gave us a term sheet. And I was like, Wayne and I were just negotiating the term sheet at the time. We're like, we'll probably take it and not go with these guys. And so I was like, "All right, let's just get to business, see what these guys offer us." It turns out, Vine offered us a term sheet.
It was like same price. And I was talking to Wayne, and I was like, "Okay, how do we want to do this? If we go to this large fund, we were just going to be 0.01% of their fund, 0.001%, some small fraction. Are they really going to care about us? Or as we go to Vine, it's like, we're going to be like 5% of the fund, or like three, whatever it may be, like some significant amount." And at the time we realized, I was like, "Okay, Adam and Alex really care about us as people. And they're way more interested in who we are rather than the idea. And for them, this will matter way more than it does for other people."
Harry Stebbings
That's so interesting, but because it's different to every other young founder that I meet who just is magnetically attracted to the brand.
Kush
Yes. I think that it is a short-term gain versus long-term gain. And you just have to be disciplined enough to choose the long-term gain. We're sitting there and we're both founders. Dude, I've got a term sheet from a big multi-stage and I've got a boutique. Would you advise me? I'd go for the boutique. And that's because it cares more. I do all these reference calls for these investors. The boutique investors want to care about you a lot more because their success is tied inherently to your success. I've also realized a lot of the partners at the boutique firms are there for more of the mission. They're also in a startup, right? A boutique firm is kind of like a startup. They're also there to make a name for my VC firm. And part of that is I got to work extremely hard to make sure these startups succeed. So I'd really go with the boutique firm. It's very contrary, right? I think to the world today, it's hard to resist, but I think it is the right decision to make for everybody.
Harry Stebbings
And then you raised from Andreessen.
Kush
Yes.
Harry Stebbings
And then we did the big firm. How did that go?
Kush
Honestly, we were talking to another large multi-stage firm at the time, and they were just taking a very long time. And so it wasn't even me. It was Wayne. Wayne's like, "All right, we got to start moving this." And so we went to Chris Dixon's team and talked to them for a week, got the term sheet, and we ended up going with them.
Harry Stebbings
I bet the other firm were like, "Oh, no, no, wait, wait, we've got one. We've got one." Afterwards, yeah. Brilliant way it works. And then Thrive, you chose Thrive over others. That's because of Miles or the Thrive brand?
Kush
I think Miles. I also think, I mean, look, the Thrive brand is great. There are other brands that are also great, but I think the partner matters way more. We were a people company, and I think it was Miles, and we also met Nithin on their team, and I had lunch with Nithin. It was the best lunch I ever had. I learned so much. For me as a founder, it is so much about being so young that I'm going to make a ton of mistakes. And I know that if you talk to other founders that are older than me, like if you talk to Eric at Modal, he's already figured out that if I do this, I'm going to fail. I don't know that yet. I don't know all these things. I was never in industry really before this. So I have so much more room for mistakes. I need to minimize that. And part of that is just talk to as many older people as possible and learn as much as possible that you can't find in books or texts, all these things that you learn as people skills.
Harry Stebbings
What do you know now, being the short while that you are, that you wish you'd known at the start about being a founder?
Kush
The mental stress and the level it is. There's a saying that it's very difficult. I think it is very difficult and it is mentally consuming and taxing because every day you're just in this struggle like, "Oh, what's happening next? What's going to go on?" I don't think people realize when getting into it how difficult it's going to be. But that is why we got into it. We were so naive that we thought this was so easy. We're like, "Why is no one done this? This is going to be the easiest thing we've ever done." And then you go in, you're like, "Okay, this is why no one else has done this before, because of how hard it is." But I think it's the level of grit it takes, the volatility that you'll feel in 24 hours of "I'm the best in the world and this is going to be amazing" and then "This is not going to work and we're screwed" in the same day.
Harry Stebbings
Yes, that is what you have to get used to. You have to get used to emotional volatility, which is so hard for people. There's this thing like, what do startup founders need? I also think if everyone got a therapist, it'd be the best thing.
Kush
Do you have a therapist?
Harry Stebbings
No, I don't. But honestly, I wish I did. If our investors are watching, someone go find me one, please.
Kush
Dude, I'll find you a therapist. That'll be my gift to you. She'll be like Wendy Rhoades from Billions. Have you seen Billions?
Harry Stebbings
I have not.
Kush
Oh, dude, come on. You seriously?
Harry Stebbings
Oh, it's a brilliant show. Do you watch TV?
Kush
Not really, no.
Harry Stebbings
I'm going here. You don't date.
Kush
No, so that's maybe the... So there's this huge joke that the reason I say I don't date... The Vine guys got these hats that say "Single Till Series B," and Series B is like next round. So I'm like, "Okay, this is soon. Like we can't be single after Series B." So I'm getting them new ones that say "Single to IPO." And so my joke is until I return investor funds, I'm not going to date. It's not actually the reason why, but we'll go with that.
Harry Stebbings
Do you think that dating does actually just distract?
Kush
It definitely does distract you. You lose the grit, the anger inside of you, or the "I got to go do this. I got to do this." But at the same time, it also helps emotionally regulate you, right? Humans are emotional people. And so part of it is being able to regulate all the volatility with stability and decreasing the volatility every time.
Harry Stebbings
Do you care about money?
Kush
I think when we started the company, I used to a lot. Now at this point, it's like no. If it was like, "Oh, you make no money in the startup," I think I would still do the startup. I think it's the people that I work with. There's just so much fun.
Harry Stebbings
What's it about for you? So like, yeah, I could be working with my friends. Is that it?
Kush
That's literally it. There's this concept of, okay, if you believe that when you're living your whole life, as long as I'm comfortable and I'm at the age right now where I have no one else to take care of, right? It's just me. And I live in a six-floor walk-up.
Harry Stebbings
What is a six-floor walk-up?
Kush
So in New York, the legal amount of stairs, there's no elevator in the building. So the legal amount of stairs you can get is six floors before it's illegal in New York to make people walk up. You have six floors of stairs.
Harry Stebbings
Yeah, six floors walk up. Yeah.
Kush
Me and one of the people we started the company with, Drew, were having a six-floor walk-up.
Harry Stebbings
What? Yeah. Why?
Kush
I mean, it was just an apartment we found that's close to the office, and I was just like, why not?
Harry Stebbings
Is it nice?
Kush
It's fine. Yeah. It keeps us gritty. That's what it is. It keeps us hungry.
Harry Stebbings
Well, you're not going to have burglars.
Kush
No, we're not going to have burglars. No, no. It keeps us hungry for more.
Harry Stebbings
That's bizarre. You live with someone?
Kush
Yeah, we have two beds.
Harry Stebbings
Dude, wow. I think a lot of people think like, "Oh, your company's worth this much, you must have so much money." Are you missing no, not really. I never worked a job before too. I don't have savings. I completely understand that now. Would I, if I was you and your co-founders, take a million bucks in secondaries when there are so many people who want to give you money? Yeah, I would take a million bucks off.
Kush
And maybe we should. And again, I'm not advocating for private jets or bullshit, but like a great gym, healthy food, and we do all that. We have a gym membership and we all like, but like a nice apartment?
Harry Stebbings
Yeah, but it doesn't really change things. I spend most of the day at the office. On the weekends, I'll go for a run in the mornings and I'll just go back to the office. There's nothing else to really do.
Kush
Do you think you have a good culture?
Harry Stebbings
Yes, I think our culture is by far the best out of any startup. In the company, when we started, I really didn't think work culture was important at all. I thought it was like you find the right people that are smart and you work. I've realized now over time, we made mistakes, we've done all these things. It's like work culture is so important because you end up spending 60 hours, 70 hours, 80 hours a week with these people. With our engineers, sometimes it's a guy named Will on our team. We spend like 100 hours a week together, just me and him, figuring out issues and going through so many things together.
You're going to spend this much time with this many people. You're spending more time with them than you're spending with your friends, your family, everyone else in your life. You spend more with the people you work with. That means that the people you work with have to be so enjoyable that you never want to do anything else. We care so deeply about the culture of the team that we've hired people. And the people that we've hired are all ex-college athletes or have played some sort of sport growing up. I heard this, I don't know who said this, but someone said this: when you grow a team from 20 to 50 to 100 people, what gets lost is communication between the team.
You get these middle management layers and all these things. And to solve that problem, to keep, if I go from today we have like 28 people, 29 people, if I want to go from that many people to 100 people in the company, I need to figure out how to 3X my team while 3Xing my revenue numbers or 3Xing the productivity of the team. To do that, I need to find people that are really good at communicating with each other. It turns out the best communicators are college athletes because what do you do when you play a sport? You're just communicating every single time. And the second thing is you are so competitive that you hate losing.
You want to find people that are competitive enough that if you tell them, "Hey, competitors launched this feature. Why don't we have it?" They're like, "We have it tomorrow." That type of level where it's like, I just want to win.
Kush
Do you know why I like athletes? I like athletes because you have to consistently do an activity that has no immediate benefit. And this is specifically like gym. I like people who really are very fit because you have to consistently do an activity that has no immediate benefits. You don't suddenly get jacked off five sessions, often on your own with no inherent motivation, for a delayed payoff that may come later. It's like short-term game versus long-term game. You got to take the long-term. And when you're going to the gym to get jacked, you're basically either doing it because you're insecure in yourself and you want to look better or you desperately seek someone else's approval. It's one or the other. I don't give a shit, but I like both.
Harry Stebbings
Or you just want to be a healthy person and live a long life.
Kush
No, because you're not getting jacked. Sure, sure. Yeah, yeah. You're running a lot.
Harry Stebbings
Yeah, you're running a lot, whatever. You're getting outside. You're doing the Brian Johnson workout.
Kush
Yeah, exactly. That's so funny. Anything else on culture and team building that you've learned that you do that's weird or different?
Harry Stebbings
The interviews, we take people to dinners and weird, uncomfortable situations to see how they interact or act.
Kush
What's an uncomfortable situation?
Harry Stebbings
We'll find people on the team that have no... We'll take an engineer with go-to-market people and have a dinner with them. And then we'll just see, do they vibe with the team? I think we have two rounds of interviews where it's technical and we'll see, can they do these things? But with AI tools, I think a lot of people are technical enough, and we're not doing frontier AI research. This is technical in the sense that it's technical and very specific domain, but really it's got to find people that are willing to be mobile and can navigate well. So can they handle uncertainty? Bit of a hard one.
Kush
That's Axios reports about you. There's... Yeah, I see you on Twitter a lot yesterday specifically. And I'm not actually going into the material, but I'm sorry if this sounds patronizing, but you are young and the numbers are big. I've seen this before where people believe the hype. How do you kind of check yourself?
Harry Stebbings
Revenue numbers, profitability, look under the books. Tell me if my business is more profitable than any of the companies that exist today. And going back to what I said before, we are a long-term family business, and part of that is our numbers have to check out with what the valuation of the company is. And today, I think our numbers do check out. Our multiples are much lower than any of the other companies that exist today, and that's because we've gone through this process of fundraising that is very disciplined and structured. We went with Miles even though we had term sheets that were way higher than that. We went with our investors that gave us term sheets that people gave us term sheets of like two billion and one and a half billion and crazy numbers, and we just said, "No, I think we're going to be disciplined at this point and just go with that."
Kush
Also, it says a lot about the investor if they're willing to slingshot a two billion dollar term sheet.
Harry Stebbings
Yeah, I think so. Part of it. Okay, we're gonna do a quick fire.
Kush
Yes.
Harry Stebbings
What was the single best VC meeting you ever had? Can I exclude our current investor list? All right, well, we'll exclude the current investor list so they don't all get mad that I don't pick one of them.
Kush
I met Nick Kranz at AIX when I was 17 years old. They ended up investing in the last round or this round that we just did. But the reason that I've known Nick for a very long time is he's mentored me since I was 17 years old at MIT. I was a freshman when I met him at a coffee shop. It's called Flower Bakery on that street's avenue, right next to my team. And I met him at the coffee shop. He gave me advice for like 20 minutes. He used to work at Great Groth at the time. He's moved to two different firms at the time. And I've talked to him, probably when I was still in school, every few months. Now I call him like once a day. It's crazy. And the really thing I enjoy is that he cares about me as a person rather than the company. That's the most important part.
Harry Stebbings
I love that. And AIX is Rich Socher's fund.
Kush
Yes.
Harry Stebbings
Yeah. I like him. He's good. You can add one person to your board who's not...
Kush
Travis Kalanick.
Harry Stebbings
Wow, you really thought about that one, didn't you? Have you met Travis?
Kush
I have not.
Harry Stebbings
No, I really want to meet Travis. We've got to make that happen somehow.
Kush
Why don't you?
Harry Stebbings
Well, I haven't asked anyone. Maybe I should. I'm like a fanboy, so it would be really awkward if I meet him. I think with a clip after this show, you might be able to make it happen. What would you ask him? One question.
Kush
So how do you handle emotional volatility?
Harry Stebbings
That's a good one. That's a really good one. Okay, you've got Anthropic at 1.5. You've got OpenAI at 1.2. Which one would you rather buy?
Kush
OpenAI. They have more compute. It's that simple.
Harry Stebbings
It's that simple? Yeah, but Anthropic have the cash. They can just buy it. You can't. I don't think you can spend like... How much are you going to spend on it? Wow. So Anthropic will be bottlenecked because...
Kush
Yes.
Harry Stebbings
What does that mean?
Kush
I mean, you can spend as much money as you want. You can't make the fabrication line at TSMC move faster. So what happens then? They are not able to provide the products they want to.
Harry Stebbings
Yeah, it's that and the models will be slower. There will be less growth than OpenAI. I think the models for OpenAI and Anthropic, one will... What is it? Is ChatGPT Sol 6 better than Fable 5.1? I don't know what the benchmarks are, but maybe Fable 5.1 is better than 6, or Astra is better than Fable 5.1. But that changes every two weeks when a new model comes out. Where is Jev in five years' time?
Kush
Oh, so I haven't used Jev. Maybe I should. Diego might be mad when he sees this. Maybe I'll go home and use it.
Harry Stebbings
So I don't know what the right answer is. I've got him coming on the show in a couple of weeks, and I need to sound thoughtful and insightful with him.
Kush
He's cool. I like him.
Harry Stebbings
But my team has. Honestly, I'll give you an answer soon. What have you changed your mind on in the last 12 months?
Kush
I used to think that coding harnesses were all kind of bullshit, that they don't really work. And like, "Oh, you're just going to use OpenAI. I'm just going to use Codex or Claude." And I'm now very pro-harness. I see the value in Factory or Cognition.
Harry Stebbings
And the reason for that mindset change is?
Kush
Switching models very quickly and being able to find the right model for the right task. And so fundamentally, it has to be an independent provider.
Harry Stebbings
Yeah. Independent provider, but also the tools that they build on top of it is a better experience. So fundamentally, the independent harness provider relies on a multimodal world. Yes. As a coder and myself. And I use it for all sorts of other things too. But yeah. You can put all of your money into only one company. Which company do you put it in? Today? Public, private. You name it.
Kush
Honestly, I'll so this is a wonderful question. Thank you. So I used to tell investors when we started the company, and you can ask our current investors this, it's true that I would put... I used Factory, and I would tell our investors put all your money in Factory. And I stopped using Factory for like three months or four months because Cognition came out and they gave us some tokens, and I used it for a while. And now I recently switched back to Factory. Better product. I would put all, if I had personal cash, I'd put it all in Factory.
Harry Stebbings
You got to get Metana. Let me invest. You're not invested already? No, I want to invest. Tell him I want it. I think the product is so good. I walked around the park with him, and I have to admit, it was one of those... Sometimes you meet founders like I did with you where you know so quickly that you're so good, but you have to pretend like you're still discussing, and you're just like, "Yeah, I just so want to invest." Do you know what I mean?
Kush
Yeah.
Harry Stebbings
And you have to play the game. You have to play it cool because if you're like, "Okay, great, I'm in," it's too quick. But I want it. I think the product is so good. Maybe the business is not like... I actually don't know the underlying fundamental business is amazing. Okay. The business is doing well, and I'm sure they'll sell to enterprises, but I think the product is so good that I know a thousand people like me that are probably not using it because they're all using all the other tools. If they used it, they're just going to switch over. Yeah. I'm giving them a free ad too. They should definitely let me in. They need to give me it. Okay. Final one. No, actually penultimate one. Which of your beliefs about 2030 is most likely to be wrong?
Kush
Humans will be a biological vessel for AI. I don't think that's going to be true.
Harry Stebbings
It's quite bold.
Kush
Yeah. For 2030. Or that people will live to 150 years. I want to believe that it'll be true. I don't think a lot of other people agree that it'll be true. For 2030, it's a push.
Harry Stebbings
Yeah, I agree with that. Final one. What are you most excited for?
Kush
I like Optimus, but I think generally, yeah. The goal of humanity is fundamentally we want to live as long as possible. That's our biological goal in life. So it's live as long as possible, live the best life as possible, the happiest life. I think AI helps us get there. And so I am very pro all of the biotech AI possibilities that exist in longevity tech, especially. I don't work in that field at all. But I like to think that the work we do by helping...
Harry Stebbings
You kind of do. You enable people to do amazing things.
Kush
Yes, I help. I like to think that the work I do by helping people access compute and hedge their exposure means that there will be more compute in the world. And that by having more compute in the world, that compute will go to good causes and will help humanity flourish.
Harry Stebbings
Listen, I think you're just an enabler of great things like bio to happen.
Kush
Yes.
Harry Stebbings
Listen, Kush, I've so enjoyed doing this. Thank you so much for doing it with me. And it's so nice to do it in person.
Kush
Thank you for having me.
Harry Stebbings
But before we leave you today, founders and investors sleep on 8sleep. It is the intelligent sleep system trusted by Mr. Mark Zuckerberg, Dr. Andrew Huberman, Charles Leclerc, and the Aston Martin F1 team. The pod is a smart mattress cover that fits on any bed. It cools each side to 55 degrees, heats to 110 degrees. It even tracks your heart rate, HRV, and sleep stages without a wearable. Thank God. The new pod 6 has a hub 50% smaller than the last one and starts at just $1,999. Visit 8sleep.com slash 20vc. That's e-i-g-h-t-s-l-e-e-p.com slash 20vc. While 8sleep helps you sleep better, MongoDB helps you build better.
Every serious AI company right now has the same problem: agents need accurate context fast. Most infrastructure though, it wasn't built for it. Well, MongoDB is. 11 Labs runs 40 million agents on MongoDB. 75% of the Fortune 100 run their most critical apps on MongoDB. Trillions of dollars moving through MongoDB every single day. Instead of stitching together 10 different tools, you get one JSON-native database, Vector Search, and Voyage AI embeddings in the same system. And it runs anywhere, no lock-in. If you're building an AI, MongoDB for startups helps you move faster with Atlas and Voyage AI credits and technical support to help you ship.
Don't build agents that answer once and forget. Build agents that remember and learn from your real-time data. Go to mongodb.com slash agents. That's m-o-n-g-o-d-b dot com slash agents. While MongoDB scales the product, Framer scales the story. If your team wants a website that looks and feels handcrafted, but is still fast to ship, Framer is built for exactly that. Framer is the AI website builder that helps creators, teams, and businesses ship production-ready sites faster than ever while getting every detail right. Prompt, inspect, edit, and publish in one place at a whole new pace. Agents and humans work in tandem.
Agents bring speed and scale. You bring taste, judgment, and control. The work lands on the canvas and stays editable. Build custom code components, manage CMS content, optimize SEO, and audit for issues all in one place. Enterprise-grade hosting, security, and 99.99% uptime SLAs, trusted by leading brands like Perplexity and Miro. Learn how you can get more out of your site from a Framer specialist, or get started building for free today at framer.com slash 20VC for 30% off a Framer Pro annual plan. That's framer.com slash 20VC. Rules and restrictions may apply.