E396: The Future of Compute, Data Centers, and AI

What if AI's most valuable commodity isn't software—but the computing power that makes intelligence possible? In this episode, I sit down with Kush Bavaria, Co-Founder and CEO of Ornn, to discuss why AI compute is becoming the next global commodity and how financial markets are evolving to support it. Kush explains why GPU capacity should trade like oil or electricity, how derivatives and futures markets could reshape AI infrastructure, and why access to compute may become one of the defining competitive advantages of the next decade. We also explore data centers, energy constraints, AI capital markets, and what it takes to build a venture-backed company at just 22 years old.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  You look at Venezuela, for example, they have the largest oil reserves. You take that away, the government starts collapsing. And I think the same way is true for compute going into forward. It's a national strategic commodity that needs to be treated that way and needs to have a market built around it.

• 0:13David Weisburd?

  You're trying to size the market. How do you go about sizing market like compute?

• 0:18Speaker 00

  The projection, seven trillion. If you take a look at the world today, where is most of the money being poured into? It's all in data centers. Every single sovereign wealth fund, every single private equity firm, any person with capital has an investment somehow into a data center project. And that just shows the sheer of construction that we are going to be doing in the next few years.

• 0:47David Weisburd?

  Kush, you're founder and CEO of Orin, and you created an exchange for data compute. Why create that?

• 0:54Speaker 00

  We basically create an exchange for compute. The reason is we think compute is going to become the next commodity of the future. Every enterprise today, and if you look back 100 years from now, everything was powered by oil, right? The clothes you wear, most of the enterprises, the largest ones, were powered by some sort of form of oil, whether that's jet fuel or crude oil or any refined form of that. And I think today, the largest economic input for our society is going to be compute. And whether that's as a usage of like AI or different tools off of that. So we think there needs to be a liquid market that exists for compute the same way exists for any other commodity.

• 1:29David Weisburd?

  Commodity markets in general exist mostly for people to hedge away the risk of the commodity. That's the base case. What's the base case for compute?

• 1:41Speaker 00

  It's probably the same reason, right? Today, enterprise's largest cost, people talk about, oh, we're spending this many dollars on tokens. We're spending this many dollars on compute. Compute will become the largest cost for any enterprise in the next five, 10 years in the future. And there needs to be the same way to sort of hedge away that risk.

• 1:56David Weisburd?

  Maybe you could explain to a layman who your early customers are and why are they using Ornn?

• 2:03Speaker 00

  Take any inference provider, for example, right? Their demand is very spiky. You don't know when someone's going to click generate on your application that you're building. And so the inference providers need to serve their customers in a very spiky way, right? People come in and buy tokens a lot for one hour and they won't buy any for the next hour. So that demand inherently needs to be hedged in a way that it's a futures market because you have to be buying on demand or like a month to month sense or a week to week sense. Where you have to buy in short term chunks because you don't know what the long term outlook is compared to many different companies like OpenAI or Anthropic where they have like a base level of compute. I think that we work primarily with a lot of the inference providers today.

• 2:41David Weisburd?

  Maybe you could double click on why compute is such a volatile market today and whether you expect that to continue?

• 2:47Speaker 00

  Today, look, demand outpaces supply by a lot. And so that creates a huge gap and it's just a simple supply and demand economics question about what has happening in the market. Volatility in the market just depends based off of what's happening in a day to day sort of sense is like, hey, this cluster just came online, then maybe supply increases a bit and so demand decreases. And then also like if a new model comes out, usually demand outgrows supply by a lot. So there's a lot of factors like that that take place in the market. The reason we think compute is probably more a commodity in the world sort of thinks of it this way. If you look at the war in Iran, the one thing that they're also targeting now is oil fields and data centers. So clearly, I think the world starts to realize that compute is a very strategic asset as well.

• 3:30David Weisburd?

  If you take a step back and you think about why is compute a strategic asset? You think, well, if all the country's applications and all their data infrastructure is built on this compute, if you knock out that compute layer, you now have applications that are not working, businesses that are not being productive.

• 3:51David Weisburd?

  And the country starts to grow at a slower pace than a country that has access to compute.

• 3:56Speaker 00

  Exactly. It's just like a commodity market, right? You look at Venezuela, for example, they have the largest oil reserves. You take that away, the government sort of starts collapsing. And I think the same way is true for compute going in the forward. It's a national strategic commodity that needs to be treated that way and needs to have a market built around it.

• 4:13David Weisburd?

  Maybe you could talk about the stack in AI, who the different players are and where you plan out.

• 4:19Speaker 00

  You can kind of think of it in like five different verticals. I think is the easiest example that people use. So there's the whole application layer companies. So think of like Harvey, Lagora, Replit, Cursor, all these companies that build applications on top of the models. Then there's the model layer companies. So you can think of these as like OpenAI, Anthropic, XAI or SpaceX now, and those sort of companies that build these models. Below them, we say that there's the infrastructure companies, right? And so you take a look at these and these are like CoreWeave, Nebius, Crusoe, AWS, GCP, Azure, all the cloud sort of platforms.

• 4:58Speaker 00

  Below that, there's usually construction or colocation facilities to help turn that energy from a greenfield into a colocation data center. And so think of those sort of as like Brookfield, for example, is a really good.

• 5:12David Weisburd?

  Data center.

• 5:12Speaker 00

  Yeah, exactly. They build the the actual physical shelves of the data centers. It's just construction at the end of the day. And then below that is all the energy companies. So you look at like Bloom Energy, for example, is one of the the largest ones. There's SB Energy, GE is a huge one that they've been building turbines now for data centers. And so that's kind of like the stock that we look at today. A lot of the money today is being poured into all these like application layer companies, right? And obviously, below the stock, there's the chip companies. I forgot to mention those, like the Nvidia, AMD.

• 5:40David Weisburd?

  That's at the.

• 5:41Speaker 00

  Yeah, at the very like core.

• 5:42David Weisburd?

  And who needs to hedge their compute money?

• 5:44Speaker 00

  It's the model companies and the sort of data center companies. That's the stock that we play in, right between both of those. And the main reason is because the model companies are the ones that buy a lot of the compute and they sell tokens off of that. And then the data center companies are the ones that are selling. The compute itself, they're the farmers in the, if you think about commodity market, like who's buying corn and then who's selling corn, the farmers are the data centers and the buyers, the corn are the model companies.

• 6:11David Weisburd?

  Expert calls have always been one of the most powerful ways to build conviction, but today investors are asked to cover more companies, move faster and do it with leaner teams. With AlphaSense AI-led expert calls, their TGS call service team sources experts based on your research criteria and lets the AI interviewer get to work. The magic is in the AI interviewer, purpose-built and knowledgeable-based information to conduct high-quality, context-rich conversations on your behalf, acting as a trusted extension of your team. Then they take it one step further. Your call transcripts flow natively into your AlphaSense experience and become queryable, searchable and comparable, so your primary insights plug directly into earnings prep, digital work streams and pitch books with zero tool switching.

• 6:54David Weisburd?

  And with AlphaSense expert call services, the AI-led expert calls are just one option because we know the importance of a hybrid expert research approach. AI for coverage and efficiency, humans for complexity and conviction. It's the institutional edge that scales research without scaling headcount. For hedge funds, that means validating thesis assumptions across dozens of experts before earnings instead of a handful. For private equity, it means faster pre-IOI scans and deeper commercial diligence. For investment banks and asset managers, that means pulling real operator perspectives straight into models and sector positioning without disconnected tools or manual handoffs. All of it lives inside the AlphaSense platform, trusted by 75% of the world's top hedge funds

• 7:37David Weisburd?

  alongside filings, broker research, news and more than 240,000 expert call transcripts, turning raw conversations into comparable, auditable insight. Take advantage of AlphaSense AI-led expert calls now. The first to see wins, the rest follow. Learn more at alpha-sense.com/HowIInvest.

• 7:57Speaker 00

  It's interesting because these markets exist in relationship with each other and they influence each other.

• 8:05David Weisburd?

  The recent example is Colossus sold a lot of its offtake to Anthropic in this famous deal between Anthropic and SpaceX. And because of that invariably SpaceX could build more data centers because now they've de-risked that asset. Do you see that happening in terms of compute?

• 8:21Speaker 00

  There's going to be a lot of bilateral deals done between large corporations and there's also going to be a market that exists for compute. And so there's like sort of two, today you would say a lot of the compute being sold is between these like large bilateral deals. But that takes away many startups that exist today that, hey, we need capacity, like we're not the size of Anthropic. And so there's a need to be a market that exists for them to get that compute.

• 8:43David Weisburd?

  I've been thinking a lot about how to think about compute innovation and a post-AI future. And I oftentimes come back to this David Deutsch, beginning of infinity. In the beginning of infinity, David Deutsch argues that innovation is infinite. Why? Because as you have more and more innovation, it starts to combine with each other and create products. Now you have 10 new innovations that combine with other 10 new innovations that create new innovations and those innovations could combine with every other innovation. Do you believe that to be true?

• 9:14Speaker 00

  I definitely think so. And I think all these new things that are going to be created have to be powered by some sort of compute, right? I think compute is the fundamental resource that is powering all the innovation today. And so that's what we're focused on.

• 9:25David Weisburd?

  A lot of people see what SpaceX is trying to do in the future with orbital data centers and data centers in space. But what about in the present? What's the timeline in the next couple of years and how does more compute go online given so much demand?

• 9:39Speaker 00

  The next couple of years, you see the headlines and that's one thing, but there's a lot of things that are happening on the ground. And so there's a bunch of data centers being built. Exact number might be like three to five gigawatts, but a lot of compute passing coming online in the next year and the next year after that. And that number has been growing, I think, double every single year. So there's going to be a lot more data centers on Earth too before we get to space. And I kind of have the Sam Altman opinion of this where he's like, data centers in space will exist, but it's going to take around 10 years to do so. Data centers on Earth are going to be the first ones that begin to come up and exist.

• 10:12David Weisburd?

  A lot of investors are weary of data centers. They see these enormous pools of capital going online from Blackstone, Apollo, name your large private equity firms, and they think that there's a bubble. What do you think about that?

• 10:27Speaker 00

  If you just look at prices of chips that are six years old or even four years old for the Hopper series, the price have been going up over time just because of how much demand there is. I think the fundamental question is, you just look at the data. Like, for the past year, AWS has kept their A100 Ampere series chips in their data centers the last six years, and they've been all rented out fully all the time. There's literally just no availability in the market for chips that are this old. So clearly, there's a lot of demand that exists, and investors should just, like, look at the data.

• 10:57David Weisburd?

  You're talking about the depreciation schedule, so a lot of companies depreciate these chips over five years, meaning that, in theory, in year five, they should be worth zero, but they're actually going up in value.

• 11:08Speaker 00

  Yes. Today, you can buy a H100 server for 180K, and that's basically the same price as you would have got it for when you bought it new. I think the new ones go for 220K.

• 11:21David Weisburd?

  I think two things can be true. I think every market is prone to bubbles. Why? Because there's this memetic copying. This private equity fund does a data center fund, another one does a data center fund, and at some theoretical point, you are going to have a bubble. But in this case, there's so much need for compute, it may not come for a decade.

• 11:42Speaker 00

  The demand right now for compute is so large that supply just isn't there. Everyone says, "We're compute constrained, we're compute constrained, we're compute constrained." And that's true across the whole market.

• 11:55David Weisburd?

  These LLM wars are insane. It's now June 2026. How do you look at the ecosystem and what are the implicit strategies by the different competitors?

• 12:05Speaker 00

  I think OpenAI is taking a really good approach where they have secured a lot of compute capacity. Sam's done an incredible job of getting compute capacity before anyone else in the market, and that's why you have OpenAI just leading in terms of how much compute they have available, how many data center partnerships they have available, and how many gigawatts they have online. Anthropic has sort of been behind on this, right? They haven't had the same partnerships that exist for the compute side, and they've been compute constrained for a while. I think recently, with this new SpaceX deal with Elon and his team, they've been able to catch up to OpenAI a bit, but I still think OpenAI is far ahead in terms of Anthropic on their amount of compute that they have available for their clients, and they're able to train their models and all sorts of things. At the end of the day, it's whoever has the most resources, in my opinion, will be able to win this race.

• 12:50David Weisburd?

  How much of a factor is Colossus? They went from creating a data center in three months to one month to two weeks. They just seem to be continuously improving.

• 12:59Speaker 00

  It's incredible. I think Elon's obviously done an incredible job of being able to build data centers faster than anyone in the market. And that's led to a lot of advancements, not just for XAI or SpaceX AI, I think would be calls it now. But also other companies that he licensed those data centers out to. For example, Cursor now has been able to train their composer 2.5 model off of their SpaceX AI data centers. And Anthropic is newest sort of customer for them. So you can kind of see that more and more of these players are becoming Neo Clouds even.

• 13:28David Weisburd?

  We talked about orbital data centers a decade from now. A lot of net new energy sources are coming online as well, nuclear and solar. Maybe you can map that world for us.

• 13:39Speaker 00

  The US has been behind on energy compared to when you look at China for a long time. And that's probably the large bottleneck today that you face when building out a data center, right? I think there's a decent amount of capital that's available today. Maybe that changes in the future, but there's also not enough energy that just exists. Powered land is such a scarce resource to come by now, given all of these data centers.

• 13:58David Weisburd?

  Land that's next to.

• 13:59Speaker 00

  Power, exactly. Exactly, energy source or that you can create a sort of on-prem, behind the meter energy source off of, either using turbines or their own sort of generators. Depends on what the actual build out is, but I think that's going to help a lot in terms of building out as many data centers as possible. Cause all this stuff requires energy. And so we just need to improve our energy infrastructure in the US specifically.

• 14:21David Weisburd?

  You guys just raised with Andreessen Horowitz, you're trying to size the market. How do you go about sizing a market like?

• 14:27Speaker 00

  The projection's like seven trillion. It's large enough. When you, I think the easiest way to see is if you take a look at the world today, where's most of the money being poured into? It's all in data centers. Every single sovereign wealth fund, every single private equity firm, any person with capital has an investment somehow into a data center project, and that just shows the sheer. Amount of construction that we are going to be doing in the next few years.

• 14:53David Weisburd?

  Yeah, multiple firms that are interested. How did you go about picking Andreessen Horowitz?

• 14:58Speaker 00

  I mean, we liked the Andreessen team probably the most out of all the other firms. They've been the most helpful. And then Chris Dixon, Mark, Ben, and our partner Ali has done an incredible job at sort of keeping us on track and focused. And they also have been incredible advisors for us. From all the firms that we had available to choose from, I think Andreessen Horowitz is by far the best. And they've done such an incredible job for us so far on all sorts of things.

• 15:23David Weisburd?

  And tell me about what it means to raise in June 2026.

• 15:26Speaker 00

  It's funny, we started the company back in September. It's been like nine months now. And it was funny because the round was fully oversubscribed and people were asking Wayne and I to buy secondaries from us. And we were laughing because we're like, yeah, we just haven't invested any of our shares. Like, how are we gonna like?

• 15:41David Weisburd?

  We don't even have shares.

• 15:42Speaker 00

  Yeah, like, say, I don't have anything to sell. Like, I don't know what you're asking for. So I think the growth for us has been such, so incredible in such a fast time. And a lot of that just becomes how much we sort of put effort we've put into, how much we focus on quality. I think our biggest factor is a lot of people in this market have tried to do what we do before, but just haven't figured out like what quality metrics you need in this. I think the biggest part is the index that we create, that we track, it's all transaction data. We don't mix it with offer data. We don't give you indicative pricing. And that may be why like our index goes up and down and it's a higher volatility than some other indices. But at least you know that this is real transaction.

• 16:18David Weisburd?

  Ground truth.

• 16:19Speaker 00

  Ground truth. It's what people are paying for compute, not what someone in the market says that compute should be worth.

• 16:26David Weisburd?

  You guys are transacting on what our mutual friend Dr. Alex Wizner Gross calls the inmost loop of the economy. The most strategic part of the economy. How do you think about things like regulation, public policy and those kind of?

• 16:39Speaker 00

  A lot of people are trying to stop the data center buildup because they're scared of. A lot of sorts of factors. I think the biggest thing is like job loss that people are scared of like, hey, like this data center is going to power what replaces me in the future. Regardless, the reason why America needs to sort of stay ahead in this data center race is like, if we don't do it, China is going to do it. They're already building out data centers. They have the chips now to do this and they have their own model companies. And so it's kind of just a race between two countries and two sort of ideologies where you want a sort of surveillance based data center and models that you don't know actually where the data is going or they're based in America. At least you can trust that these models are not sort of spying on you, are not injecting code into your enterprise and different things.

• 17:21David Weisburd?

  It's the hard pill to swallow, which is yes, things will be crazy. Things will be unpredictable. Nobody could predict. Not even Elon Musk knows what's going to happen. And yet we have to push forward, not only push forward, but push forward at breakneck speed in order to win the AI.

• 17:36Speaker 00

  America has the largest capital markets in the world and that's probably our biggest advantage compared to any other nation. And we need to leverage those capital markets in the same way we've done so with all sorts of other resources. And so we're trying to help advance that by helping create capital markets for compute.

• 17:50David Weisburd?

  And how old are you?

• 17:52Speaker 00

  Twenty-two.

• 17:52David Weisburd?

  So you're twenty-two. You just started last September. Talk to me about building a company. How has that been?

• 17:58Speaker 00

  It's pretty funny. So a lot of our engineering team is around like my age, like twenty-five. Some people are twenty, twenty-one. But our go-to-market team and people that have been in the industry for a while and our legal team is obviously much older than we are. It's a little weird when like you come into the office and there's like people that are probably like twice my age, like asking like, what do we do next? I was like, oh, this is like, uh, this is weird. And then we've had to put, it's been interesting though, and it's a lot of like learning for us. And we've focused so much off of like hiring the right people for the right job. Like talent is what the company is today and that's the most important thing. And so we say we're always hiring and that's partly true. If there's the right person that comes along, it's always hiring.

• 18:37Speaker 00

  But it's also maturing very quickly. It's learning how to get deals done and how to negotiate with people and get all sorts of things that you just, you don't learn in college. So it's just things that you kind of have to learn in the real world.

• 18:47David Weisburd?

  You mentioned you have developers roughly the same age, some maybe a little bit older, some maybe even younger. Is it fundamentally different type of developer that you're hiring in this AI world than you would two, three years ago?

• 19:00Speaker 00

  Look, we just want developers that can use AI tools. I'd rather not teach someone that's been in the industry for ten years that refuses to use like Codex or Claude Code or Cursor. Name your favorite like model. It's so much easier to hire someone that already been using it and they know how to use all of these tools. We just focus on quality of output and that's the metric we track, not like anything else. And it just happens to be so that people around my age are better than doing that than people that are older.

• 19:26David Weisburd?

  In theory, coding is now in the English language. Have you seen either non-developers that have been able to succeed? Or somebody that's only been developing for a short amount of period that has been able to accelerate their growth?

• 19:38Speaker 00

  That's definitely true. There's like definitely non-developers that will like learn how to code because it's just easy now. But also a level of thinking that you get as a developer and how to actually build the product out. How do I architect this?

• 19:49David Weisburd?

  That hasn't changed.

• 19:50Speaker 00

  Yes, I think the coding agents haven't caught up to that yet. They will in the future. I like totally agree. I just don't know how long it will take to get there. There's also a lot of technical skills that you learn, not just from like coding classes, but how to think. That's very important. So I'd say coding is like basically, I haven't coded like a line of code in a long time. I use Codex. I prompt Codex different ways to sort of use that. But at the end of the day, it's like how to actually prompt it, what to change, how to get this infrastructure set up. That's some of that stuff hasn't been solved yet. And so just working out to solve that.

• 20:22David Weisburd?

  Let's go back to how you're building the business. What mental models do you use on how to grow a hyper-growth startup in the age of AI?

• 20:30Speaker 00

  Uh, one is customer quality is like a huge thing that we focus on. The reason we were able to work with our customers so fast is we have a five-minute response time. I used to work in consulting and there was like a rule where if you get an email, you have to respond to it in 15 minutes. Like, whether that's for a client or whatever. During the week, during the week. Consultants are more like bankers. We, they work Monday to Friday usually, sometimes Saturday, Monday to Friday. And you just had to respond and we've taken that same sort of approach for our customers in tech. Where I think every single one of our competitors or people that initiate trying to do something that we do, just have like a two, three, five-hour response time. And it sucks if you're like a developer and you're using the product and you're like, hey, like where is this? This doesn't work. Like can I just get this like done now?

• 21:13Speaker 00

  So we focused a lot on just like customer support and quality of what we're offering. And just like quick responsiveness back and forth between people. We say our customers are our friends and many of our customers are in fact investors in the company now too. And that's because they've realized like, hey, these guys have like great quality and they, I mean, they know the market more than us, right? They've been in the space like, and so clearly they see something that we haven't seen ourselves yet either.

• 21:36David Weisburd?

  Taking back to the origin story of Orin, how did Orin start?

• 21:39Speaker 00

  So Wade and I. We were both working full time and I remember in July and August, like Wayne, he's a quant trader before this. He was like, hey, like I've been trading like Terra Wolf and Core Weave and all the stocks. Like these things are going up and that's when all the NeoCloud stocks were just like going straight up. And we're realizing like for, if you look at oil markets, like for Exxon and Shell and all sorts of these public companies, their assets are pretty directly tied to like oil prices, right? The actual, I think the beta is very high relative to what oil prices do and oil companies, how the stocks move. But there's no sort of commodity market for compute. So like, why doesn't this exist? And because if you look at every enterprise today, they buy from multiple clouds. Today, no one really cares where the compute comes from, as long as it's up to a certain quality.

• 22:23Speaker 00

  And then they just care about price. There's like a minimum threshold of quality and then price is the biggest factor after that. And so we've been focused on like, like this makes sense. Like this is essentially just a commodity, like this should exist. And so we took that idea and sort of started creating a, a market for this compute.

• 22:41David Weisburd?

  If you go back just nine months ago, before you started, Orin, what is one piece of timeless advice you'd give yourself?

• 22:47Speaker 00

  Continuously focused on building out the team. I probably was spending way too much time early on focused on like, like working on the product and the business. I'm like trying to do everything myself. And now I've realized you can sort of trust people to get the actual work done. You don't have to do everything yourself. So it's just focus on hiring the right people to do the right job.

• 23:07David Weisburd?

  I've thought a lot about this, which is working on the product, you get immediate feedback. You get progress, you get immediate milestones. Finding people takes hundreds of people to interview and it's a long process. It's a long on-ramp, but then it really compounds much more than obviously your own time, which is finite.

• 23:23Speaker 00

  Exactly. It's something that I wish I realized early on.

• 23:29David Weisburd?

  Well, Kush, uh, thanks so much for jumping on the podcast and looking forward to seeing this grow and the future of Orin.

• 23:35Speaker 00

  Sounds good. Thank you for having me.