# The Accidental YC Founder Solving AI’s GPU Crisis | SF Compute’s Evan Conrad

Markdown copy of https://realalpha.app/v/uYzQxxAD41g/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Evan Conrad argues that credit, not GPU supply alone, constrains AI compute—and that a physically settled market can make clusters easier to finance.

- Show: Upstarts
- Hosts: [Alex Konrad](https://realalpha.app/people/alex-konrad/)
- Guest: [Evan Conrad](https://realalpha.app/builders/evan-conrad/) is CEO and co-founder of SF Compute.
- Published: Oct 29, 2026
- Length: 39 min

## The best minutes

- [1:11](https://realalpha.app/v/uYzQxxAD41g/?t=71) [GPU markets can make compute easier to finance](https://realalpha.app/s/gpu-markets-can-make-compute-easier-to-finance/) (4 min)
- [7:15](https://realalpha.app/v/uYzQxxAD41g/?t=435) [GPU clusters can fail in ways buyers may not expect](https://realalpha.app/s/gpu-clusters-can-fail-in-ways-buyers-may-not-expect/) (4 min)
- [10:53](https://realalpha.app/v/uYzQxxAD41g/?t=653) [Compute deposits can force startups to raise at huge valuations](https://realalpha.app/s/compute-deposits-can-force-startups-to-raise-at-huge/) (6 min)

## Also worth watching

- [28:16](https://realalpha.app/v/uYzQxxAD41g/?t=1696) [GPU futures need to account for performance differences](https://realalpha.app/s/gpu-futures-need-to-account-for-performance-differences/) (4 min)
- [36:34](https://realalpha.app/v/uYzQxxAD41g/?t=2194) [B300 rental rates and workload value may converge soon](https://realalpha.app/s/b300-rental-rates-and-workload-value-may-converge-soon/) (2 min)
- [22:59](https://realalpha.app/v/uYzQxxAD41g/?t=1379) [A $245 million compute deal is small by industry standards](https://realalpha.app/s/a-245-million-compute-deal-is-small-by-industry-standards/) (3 min)
- [34:11](https://realalpha.app/v/uYzQxxAD41g/?t=2051) [Compute financing could become a US geopolitical lever](https://realalpha.app/s/compute-financing-could-become-a-us-geopolitical-lever/) (3 min)
- [20:25](https://realalpha.app/v/uYzQxxAD41g/?t=1225) [Conrad tried around 40 products while searching for a business](https://realalpha.app/s/conrad-tried-around-40-products-while-searching-for-a/) (3 min)

Skip the intro and sponsor reads, the earlier YC and first-product stories, the investor-pitch discussion, and the detailed market roadmap.

## Summary

Evan Conrad is CEO and co-founder of SF Compute. He says the company grew out of his first AI lab, which had about $500,000 in the bank and a GPU bill of about $400,000 a month after buying too much capacity.

Conrad argues that AI compute’s key bottleneck is credit: lenders hesitate to finance clusters for young labs, so buyers face large deposits and startups raise at high valuations to cover them. He says SF Compute addresses this by operating clusters owned by partners and selling actual GPU capacity in an independent, physically settled market. He sees that structure as a way to let smaller labs access capacity, while noting that compute is technically nonfungible and hard to commoditize. He also predicts continued scarcity and rising GPU prices, and says data-center restrictions could shift financing and compute capacity across geopolitical lines.

### Minute by minute

- [1:11](https://realalpha.app/v/uYzQxxAD41g/?t=71) **Credit constrains compute access.** Conrad says GPU buyers face large deposits because lenders are reluctant to underwrite AI labs’ credit risk. He says SF Compute created an independent, physically settled market so buyers can secure GPU capacity in the future and lenders can assess the asset’s value.
- [4:45](https://realalpha.app/v/uYzQxxAD41g/?t=285) **SF Compute operates clusters for owners.** Conrad describes the company as a property manager for supercomputers: it helps owners design, build and operate clusters, then sells their capacity. Owners include investors and nonprofits; short-term sales can leave capacity available for a portfolio company or grantee that needs it later.
- [7:15](https://realalpha.app/v/uYzQxxAD41g/?t=435) **Reliable delivery requires operators.** Conrad says deployment problems pushed SF Compute from subleasing into operating clusters, including voltage issues and broken InfiniBand networks. He says the company has encountered data centers in chicken coops and flooding during a monsoon, and argues that compute needs an operator to make it a reliable market.
- [10:44](https://realalpha.app/v/uYzQxxAD41g/?t=644) **Credit drives up startup valuations.** Conrad says GPU clouds can make more money by selling long-term capacity to frontier labs, while SF Compute chose a different model. He says deposits can force new AI labs to raise at very high valuations: he recalls initial valuations of $100 million a year or two earlier and about $1 billion at the time, and projects $5 billion, $15 billion and $30 billion.
- [13:29](https://realalpha.app/v/uYzQxxAD41g/?t=809) **Liquid capacity can arm smaller labs.** Conrad says SF Compute wants startups like its former lab, June Lark, to get compute without first raising at a billion-dollar valuation. He contrasts owned GPUs, which allowed xAI to sell capacity, with leases that some labs could not sublease, and says the market should serve more than the largest labs.
- [17:44](https://realalpha.app/v/uYzQxxAD41g/?t=1064) **A therapy exercise led to YC.** Conrad says he went to a Y Combinator interview because it frightened him and he was working through cognitive behavioral therapy. His first YC company was Quirk, a mental health app; he later tried a distributed-systems product and, over about four years, around 40 product ideas. He recalls Dalton Caldwell’s advice to keep going and not die.
- [21:15](https://realalpha.app/v/uYzQxxAD41g/?t=1275) **Survival kept the company going.** Conrad says SF Compute often had a month of runway and that he did not want to fire the team or let people down. He says he had not initially understood the market’s scale and once thought $5 million was a lot.
- [23:10](https://realalpha.app/v/uYzQxxAD41g/?t=1390) **Conrad calls $245 million modest.** Conrad says the company nearly did not announce its $245 million contract because it seemed small by compute-market standards, though it mattered to SF Compute. He says the company is still in survival mode and does not want to become a neocloud that owns and resells most of its hardware. In his view, an independent operator is needed to build a trusted forward market.
- [25:56](https://realalpha.app/v/uYzQxxAD41g/?t=1556) **The market resists a simple pitch.** Conrad says the company’s first description was a power grid for compute, but that phrase led people to ask why compute needed a grid. He says investors sometimes mistook SF Compute for a competitor or a simple GPU-matching marketplace, even though operating a physical market involves finance and supercomputing.
- [28:16](https://realalpha.app/v/uYzQxxAD41g/?t=1696) **Compute is not fully fungible.** Conrad says compute futures cannot work like wheat because GPU capacity varies by operator and hardware version. He cites a 40% difference in GPU kernel-launch performance between some VM implementations and says early H100s performed worse than later ones. SF Compute’s market delivers GPU time, rather than settling a bet against a price index in cash.
- [31:43](https://realalpha.app/v/uYzQxxAD41g/?t=1903) **Lower deposits depend on a forward market.** Conrad says he wants buyers to find available capacity, pay for shorter lead times and face smaller or no deposits. He argues that lenders could finance clusters based on expected future compute prices rather than each buyer’s credit. A liquid market could also let operators replace failed nodes sooner and reduce delays.
- [34:21](https://realalpha.app/v/uYzQxxAD41g/?t=2061) **Compute could become geopolitical leverage.** Conrad predicts GPU prices will rise and says supply will remain behind demand for at least four to five years. He argues that credit and revenue streams, not land and power alone, constrain new clusters. He says data-center moratoriums could push compute outside the United States and make capital flows a geopolitical asset.
- [37:24](https://realalpha.app/v/uYzQxxAD41g/?t=2244) **GPU economics favor owners with advantages.** Conrad says a B300 could generate $20 to $50 an hour depending on its use, while its price was $5.50 an hour, or perhaps $6. He expects those figures to move closer together. He says buyers with strong credit, power or land—or a portfolio reason to reserve GPUs—should consider owning them; others should probably lease.

## Transcript

Upstarts, hosted by Alex Konrad, with Evan Conrad.

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[0:00](https://realalpha.app/v/uYzQxxAD41g/?t=0) A mid-sized deal in Compute is bigger than many venture funds' entire AUM. Finding GPUs is difficult, buying them is expensive, and operating them is risky. San Francisco Compute calls itself the power grid for Compute, and CEO and co-founder Evan Conrad says he has invented a new kind of market for supercomputing, which is even more complicated than it sounds. For like the last 30 years, companies were really simple ideas, and you could easily explain them in like a single sentence. Welcome to fucking Compute. It is not a simple sentence. Today on the podcast, we'll talk about the services that people who own supercomputing hardware really need, how to survive the GPU race one month at a time, and why SF Compute thinks it's important to arm the rebels.

[0:43](https://realalpha.app/v/uYzQxxAD41g/?t=43) Plus, how to get into Y Combinator without really trying. I went and did a YC interview because it seemed very terrifying. It was like a therapy exercise. This episode is brought to you by Jira by Atlassian, where teams and agents get the context, coordination, and control to move work forward. Try it free at Jira.com. That's J-I-R-A dot com. Evan, welcome to the podcast. Thanks for having me. You are also a Conrad, but spelled the more common way, I guess. It's not a bad way. First double Conrad we've had on the show. I feel like you've got to get other Conrads. I've had other Alex's, but I haven't had other Conrads.

[1:16](https://realalpha.app/v/uYzQxxAD41g/?t=76) This is a really special moment for me. Tell our audience, who probably know broadly what is compute, but they don't know what San Francisco compute does, where are you playing in this space? We invented markets for supercomputing. So if you go buy a big GPU cluster today, you're going to be asked to place down a massive deposit that is your entire bank balance.

[1:36](https://realalpha.app/v/uYzQxxAD41g/?t=96) **Evan Conrad:** The reason why you had to place that deposit in the first place is because the lender doesn't want to underwrite your credit risk as an AI lab.

[1:44](https://realalpha.app/v/uYzQxxAD41g/?t=104) To solve that problem, you have to get that lender comfortable with the asset value. You have to make them feel comfortable that GPU prices in the future will be what they expect it to be. If you do that, you don't have to pay the big deposit or a smaller deposit. To make that possible, you need a compute market to exist. To make a compute market exist, it has to be physically settled, meaning that somebody's operating in that cluster, and when you go to buy the compute in the future, you know what you're getting. And that you can get it then. And that you can get it. To make a compute market and to make it viable for the rest of the world, it has to be independent.

[2:17](https://realalpha.app/v/uYzQxxAD41g/?t=137) So it can't be done by the existing clouds. That's what we do. We're a physically settled compute market. We believe we're the only physically settled compute market. It's extremely difficult to do, and it solves the sort of key problem of scaling up compute. mainly credit. Why are you guys the only ones doing it? Do people think you're nuts that this is the best approach? I think it's just very hard to do. A long time ago, we were this little AI lab that was trying to make audio models. We bought too big of a GPU cluster. And then we had about 500k in the bank. And our bill was about 400k a month. And we signed up for about 12 months, which meant that the first month went great. And the second month did not. And then every month, we would have to sell out our cluster or go bankrupt and because of that we went on this sort of long and perilous journey of trying to fix that and ended up inventing compute markets because of it when people go to buy gpu clusters today they run into a bunch of different problems that we hit i think a little bit earlier than maybe some other folks had hit it before those problems are one if you go out to buy gpus everyone is sold out there isn't really on demand

[3:23](https://realalpha.app/v/uYzQxxAD41g/?t=203) for GPU clusters today, it can't really be financed. If you go buy a new GPU cluster, you have to buy a new one, which means there's a lead time. That lead time is gonna be quite long because there's a physical process. It's not going to be click here to get a button, or click here to get a supercomputer. You're gonna pay a massive deposit. You have to pretty much put your entire bank balance on the line, and you may even need to raise more money. So this is why folks raise big giant rounds at pretty high valuations relative to where they are. finally once your cluster is here the likelihood that you correctly predicted your demand unless you were extremely sophisticated is very low um you're either going to wildly overshoot it um meaning that you have too much capacity and you're gonna end up like us um in the beginning and you're gonna have to you're gonna figure out how to sublease it or you're just gonna go under or you're gonna undershoot your demand um and you're gonna look at all your competitors who bought tons of capacity, and you're going to say, shoot, I fucked up, and I should have bought way more, that problem we think is all the same problem.

[4:28](https://realalpha.app/v/uYzQxxAD41g/?t=268) It's just risk, and to solve that risk, we believe that you should make a compute market. For people who understand, okay, we need these GPUs, we need these clusters to be training models, doing a lot of stuff in AI today, and companies big and small want these, are you guys... also sort of building this capacity and operating it? Are you more matching capacity to the companies? How should we be thinking about this stack and where SF Compute fits? Many people in the world have land... our capital, credit, or co-location space, or maybe they went out and bought a big GPU cluster and they would like to become a cloud or they would like to own their own hardware, that's what we do.

[5:11](https://realalpha.app/v/uYzQxxAD41g/?t=311) We're like a property manager for supercomputers. So we go to folks who want to become a cloud and we help them do that. So for context, our CTO is the co-founder of Voltage Park, a multi-billion dollar GPU cloud. So we've done this before at quite large scale. Well, we help people who want to have GPU clusters, design them, build them, operate them, and then sell them to people. We can work in some capacity like Stripe. So we operate GPU clusters for folks, and then you can put your own website in front of us. And the customer never needs to know that San Francisco Compute exists. We just sell the capacity, and then you can sort of treat us like an API, but we're the ones operating the supercomputer.

[5:53](https://realalpha.app/v/uYzQxxAD41g/?t=353) Or you can directly go to sfcompute.com and buy from us. The reason why people work with us typically is either, one, they need the offtake. They have some asset. They would like to turn it into a GPU cluster, and they want somebody to buy it. Two, they want the GPUs because they want it for their portfolio. An example is a venture firm or a nonprofit. They're worried about a compute shortage in the future, and so they have us build a cluster for them that they own. And then in the meantime, we're just selling the capacity for them. And when their portfolio, either some non-profit grantee or, you know, a venture firm's portfolio company, starts to fly off and really need the GPUs, we can stop selling the GPUs to whoever we were selling it to because we're only selling them on short-term contracts and then hand them back to the portfolio company or the portfolio project that needed that capacity.

[6:43](https://realalpha.app/v/uYzQxxAD41g/?t=403) Also, there's just a lot of people who would like a pure financial return. They fully understand that compute is a very big asset class. There's a really high IRR if you do it right. But they don't know how to operate a supercomputer. And so they look for a partner who can operate a supercomputer for them, and we're that partner. I mean, that's a lot of stuff. I guess the initial use case when you're this small startup, you have other GPU compute that you've basically overbought and you need to find takers for it. is that how then the business looked and then you've kind of added more of the full stack or has it been full stack kind of throughout? So what happened when we were subleasing our cluster is we immediately realized that nobody knows what they're doing.

[7:25](https://realalpha.app/v/uYzQxxAD41g/?t=445) What was the moment that you felt that way? We've had clusters that we were working with. other folks on that when they arrived they browned out because they were installed in the wrong data center and so they had voltage issues that caused the cluster when you ran a burn in or ran linpack on it would basically fall apart we've had clusters where the infiniband fabric was broken in a way that was unsellable to customers the party that owned it wasn't aware we then fixed it for the party. And then they broke contact with us and walked away with cluster on the moment we fixed

[8:06](https://realalpha.app/v/uYzQxxAD41g/?t=486) **Evan Conrad:** it. We've seen basically everything that can go wrong. We've seen data centers that were literally chicken coops. We've gotten hit by a monsoon at one point and got flooded during a site visit. It's just the number of things that can go wrong when you're deploying a GB cluster

[8:23](https://realalpha.app/v/uYzQxxAD41g/?t=503) is very high. And early days of SF compute, when we were just trying to buy capacity, what happened is basically like first we were the AI lab then we were trying to sublease it then we would go deeper into the stack to figure out like how do we solve this particular problem and we just kept doing that until we eventually ended up at the bottom where we realized okay no you can't actually do this unless you operate the entirety of the service the benefit of that from the financial perspective is if you're operating the entirety of the service but also it's not on your balance sheet somebody else's then what you can do is you can offer an actual commercial forwards market.

[8:58](https://realalpha.app/v/uYzQxxAD41g/?t=538) One of the ways to think about this is in the power market. Electricity is just straight up super complicated. It doesn't work like soybeans. You need an operator, like a great operator, basically, in order to turn it into like a valuable commodity. That's what we do. Sometimes this is called physical delivery. That's another way to think about San Francisco Compute is we're the only physically settled compute market. And by physical settlement, we mean we do all of the functions of a cloud, but we ourselves are not a cloud. Can you explain what that means? We don't own the hardware. And what we do is we work with people who want to own the hardware for financial reasons, and then we're their partner for operating the cloud portion of what they're doing.

[9:35](https://realalpha.app/v/uYzQxxAD41g/?t=575) But it's not ours. They set the prices, they determine how the rules work, and so on. And if they want, depending on if they've sold it or not sold it, they can recoup the capacity. Can you give me just a quick hypothetical situation to maybe make this more tangible for folks? So someone wants to buy thousands of GPUs from NVIDIA. like they want a bunch of Blackwell GPUs to set up a new cluster so they buy that hardware do they then call you and say we need help setting this up? Or what's the next step? If you come to us today, almost certainly, we will want to be involved in what you're buying. It's very easy to fuck up a purchase of a supercomputer. And many of the problems happen right when you are designing the bill of materials, like what you're going to literally buy. And so we typically get involved before anything is purchased. We then do a consulting period with somebody where we help them avoid really common mistakes. An example is putting the wrong thing in the wrong cluster, or like putting a cluster in the wrong data center. Once the hardware is installed, we then take over and operate it like a cloud for that partner. We can either then sell that capacity to a customer. We

[10:37](https://realalpha.app/v/uYzQxxAD41g/?t=637) can sell it to their customer and just operate it for them and they don't have to ever, like nobody has to know San Francisco Compute exists. Or we can operate it for them on their behalf for like a portfolio company or a grantee or something like that. There are companies trying to build, I mean, there are very highly venture backed, like multi-billion dollar companies and also public companies who are trying to build that extra compute access, right? Why are they not getting it done? If you were a GPU cloud today, you should be selling all of your capacity to the largest labs on a long-term contract. You will make unbelievable money doing that, like $50 billion of money doing that. Anything else is an afterthought.

[11:15](https://realalpha.app/v/uYzQxxAD41g/?t=675) Trying to solve any of those problems is like a fool's errand if you were a normal cloud. Why aren't you just full-time selling to the labs yourself? That's why this company is extremely difficult. We had to build all the capacity to become a GPU cloud, and then we had to decide not to become a GPU cloud. It's why we say that we have no competitors, because anyone who competes with us is basically stupid. But what makes you smart? Most of the time, it was just because we didn't have any other better option. For the majority of the company, it was that we were going to go bankrupt if we didn't do it.

[11:47](https://realalpha.app/v/uYzQxxAD41g/?t=707) Now, I think it's basically like, okay, we know that this is our path. We would like to solve this problem. If we don't solve this problem, what happens is all of the capacity in the world gets locked up by only a few buyers without a grid, without something like this. What happens is every AI lab, when they need to start, has to pay for a massive deposit. They have to pay for that massive deposit because they have no credit. And so when the GPU cloud looks at them, they go back to their bank and they say, I would like to finance a new cluster. Can I finance it with this customer? And the bank looks at the customer and says, that is a two-person AI lab in a San Francisco Victorian.

[12:27](https://realalpha.app/v/uYzQxxAD41g/?t=747) In order for them to actually get the cluster, the two-person... AI lab in a San Francisco Victorian, the new startup, has to go out to the venture capitalists

[12:34](https://realalpha.app/v/uYzQxxAD41g/?t=754) **Evan Conrad:** and has to raise, you know, a billion dollar valuation right off the gate in order to pay for a big deposit, which then makes them, makes the whole thing financeable. This is the current

[12:44](https://realalpha.app/v/uYzQxxAD41g/?t=764) state of the market. As we scale up into superintelligence, that's only going to get worse. So, you know, a year or two ago, it was a hundred million initial valuation. Now it's like a billion initial valuation. It's going to be five billion and it's going to be 15 billion and it's going to be 30 billion and at some point it's going to become unreasonable that you cannot actually do that the way to solve that is you have to make it possible for the lender to think okay i know what this asset this gpu cluster is worth in the future and so i don't care about the credit risk of the buyer i don't care that it's a two-person ai lab because if you're not selling to the two-person ai lab you're going to sell to somebody else because this thing has an actual asset value that's what we're doing. If there's not enough compute, if the AI labs have a rapacious appetite for compute and can pay top dollar for that compute, what drives you to not just create a more efficient market to funnel them even more compute and maximize your own profits in doing so?

[13:47](https://realalpha.app/v/uYzQxxAD41g/?t=827) We expect at some point to sell to Frontier Labs. The thing we want to make sure happens is that that capacity is liquid. It's available to be sold to other people as well. I hear you. But are you intentionally arming the rebels or are you arming the rebels as just sort of the way to get started to kind of get to the scale that you did out of necessity at first? We're intentionally arming the rebels to some extent. The thing we want to make sure is that if you were starting June Lark today, June Lark was our AI lab, you would actually be able to do that. Right now, what happens is the only way you're going to start something like June Lark today is you go out and you raise a billion-dollar valuation right out the gate.

[14:27](https://realalpha.app/v/uYzQxxAD41g/?t=867) Right now, we're living in the good times. We're living in super high valuations are possible, markets are hot. That's not going to be sustainable for a long period of time. We don't think that's possible. When it is no longer sustainable, what that means is all of the startups that could have existed are impossible to start because you can't go get a cluster. You can't raise money for the deposit. Therefore, you don't exist. We would like it to be possible for startups to exist. And so from our perspective, we were like, okay, step one, make June mark. Oh, fuck. okay, well, I guess if we don't make this, we can't go and make DuneLark.

[15:05](https://realalpha.app/v/uYzQxxAD41g/?t=905) Like, it won't exist. Like, it won't be possible to make. That was basically our logic. That if you don't solve this problem, there isn't another startup that we could do. But the hope is to empower more DuneLarks, not just more power law, like, big lab winners, basically. I think the big labs have the same problem. They just have it at a bigger scale. If you look at, like, XAI, XAI is the situation in which you overbuilt compute, and then they were forced to sell it. The reason why they could sell it is they owned the hardware. They bought it. And so they had the residual value on that. If you were in a lease, you would not necessarily be able to sell that capacity, which is what happened to many AI labs.

[15:42](https://realalpha.app/v/uYzQxxAD41g/?t=942) They bought a lease of capacity, and they were not allowed to sublease that capacity. They were not allowed out of that. Even if they were able to sublease that capacity, they personally had to go find the subleaser. And because nobody's quite sure if you personally will be able to find the subleaser, This has lots of downstream effects against the lender who's worried that you're going to sell this capacity to someone who needs to sublease at all, who's in a not necessarily great situation. And because they're worried that, oh shit, these people may not be able to afford this in the future, they're going to ask you to pay a big giant deposit to de-risk them, or they're going to not sell it to you.

[16:20](https://realalpha.app/v/uYzQxxAD41g/?t=980) And they're going to sell more of it to the folks who have really great credit. How do you move fast but also stay in control? You have to ship quickly, but you want to ship with confidence. And a lack of context is breaking that balance. Your AI is only as good as the context it has. And for most teams, that context is scattered across tools, threads, and DMs. But you don't have to ship with Blindspots when you use Jira by Atlassian. What's the goal behind a project? What got decided in Slack last week? Atlassian's teamwork graph pulls together across Jira, Confluence, GitHub, Slack, and more.

[16:55](https://realalpha.app/v/uYzQxxAD41g/?t=1015) so nothing falls through the cracks. Teams get better answer quality while also using fewer tokens. And you can share that context with the agents you already use like Claude, Cursor, or GitHub Copilot or connect your own tools through Atlassian MCP. Keep the speed, stay in control, and ship with confidence with Jira by Atlassian. At Upstarts, we document startup journeys that upend the status quo while building great businesses. And iterating on business models that best reflect your product innovations is a big part of doing that. Chargebee is the billing, metering, and monetization infrastructure that enables just that for some of the fastest-scaling AI-native startups.

[17:34](https://realalpha.app/v/uYzQxxAD41g/?t=1054) Charge for active agent minutes, multimodal usage credits, different seat types, or any other custom combination by heading to chargebee.com slash upstarts. That's C-H-A-R-G-E-B-E-E dot com slash upstarts. June Lark was, I believe, audio-focused AI Lab? Yeah. Was this something that you had been interested in for a while, or how did you end up kind of a founder in the first place? I had very severe panic disorder for a while. Like, I had lots of panic attacks. I did a thing called cognitive behavioral therapy, which got over the panic attacks. But one of the things that you do when that happens is you do things that make you afraid.

[18:11](https://realalpha.app/v/uYzQxxAD41g/?t=1091) I wandered in to a coffee shop, and I talked to these folks who were talking about their YC interview, and it sounded really terrifying. And so I went and did a YC interview. No way. Because it seemed very terrifying, and it was like a therapy exercise. I, like, looped in my brother to do it, and I was working on this, like, CBT app at the time. And so I went in and pitched that. We got into YC. I then went back to my— Did you tell them, by the way, I was only doing this as a therapy staff? I think I did later. I don't actually remember if I did. Like, I liked making products. 99% of applicants or whatever to YC never get a callback or never get in.

[18:46](https://realalpha.app/v/uYzQxxAD41g/?t=1126) You do it almost accidentally, it sounds like. I mean, like, I had a thing. But do you think maybe you were more confident in the pitch because you were like, I'm not—my hopes and dreams are not tied to the result here. I would not describe myself either now or before as confident. Well, obviously, you showed something to get into YC, though. I think I just liked making things. Since I was a young kid, I was just making shit all the time. Okay. Yeah. Like what? I would constantly make different types of mobile apps. And then in college, my roommates and I had this consulting company in which we would make stuff for other people.

[19:19](https://realalpha.app/v/uYzQxxAD41g/?t=1159) We made a point-of-sale system for a golf cart where there would be a teenager or something that would come around in a golf course. and take orders for people from the kitchen back at the main spot. Oh, got it. And it would coordinate with... You were DoorDash for golf courses. We were DoorDashes for golf courses, basically. We made DoorDashes for golf courses. That's a really good, extremely good description. Well, I'm just thinking of the YC pitch you didn't have to make for that one. Exactly, yes. So did you go to YC with this idea, or what happened next? Yeah, so we went to YC with Quirk, which was our cognitive behavioral therapy app, and pretty much immediately ran into the problem that every mental health app has, which is like if you do well at all like if your thing works your customers churn And so there's really negative incentives in that.

[20:02](https://realalpha.app/v/uYzQxxAD41g/?t=1202) Kind of like, I guess, dating app challenges as well, right? Similar to dating app challenge, yeah. So we kind of went back to people and said, what should we do? And they said, well, I guess you could try other shit. And so we tried a thing called a room service. A room service was like a distributed systems company. We built the component piece of Figma that makes things multiplayer. We went with this, and then after nine months of this, realized that the people who were most likely to buy it were the people who were most likely to build it themselves internally. And so I went back to our investors again. And one of them told me that I should stare into the ocean and then I should come back and just try shit for like forever.

[20:39](https://realalpha.app/v/uYzQxxAD41g/?t=1239) And so I was like, yes, sir. And then Dalton Caldwell said, don't die. And I was like, yeah, don't die. Got it. Shout out Dalton. Shout out Dalton. And internalize that turns out way

[20:52](https://realalpha.app/v/uYzQxxAD41g/?t=1252) **Evan Conrad:** more things you should know other than just don't die. And tried like 40 products of different things. Gave up on some of them way too early because somebody told me that they were stupid.

[21:02](https://realalpha.app/v/uYzQxxAD41g/?t=1262) And I was like, oh, this person must be very smart. And so therefore, it's done. That was the wrong decision. So I went on that for like four years or so. And I think the thing that it drove into my head is if you just do a lot of iterations and loops, you realize what works and what doesn't. It also sounds like the don't die was really important there too, because part of me when I hear your story, I'm like, okay, you didn't have to go resell all this compute. You could have taken the bankruptcy, taken the L, moved on to another startup and not wasted time on something that theoretically you didn't enjoy. Obviously, you must enjoy it now or you wouldn't be doing it. But there must have been a moment there where you guys at least discussed, should we bother doing this, right? In the early days of San Francisco Compute, we hired a bunch of people And they were going through a lot.

[21:49](https://realalpha.app/v/uYzQxxAD41g/?t=1309) Like, it was not a good time in the early days of San Francisco Compute because our economic incentives around us was insane. We had like a month of runway at any given point. And so just the pressure and stress of that was like throughout the company. But why were you subjecting them to it? Why do it? What had made you not want to just shut down and take the loss on the cluster you'd overcommitted to? In my head, I just didn't want to fire everybody. Like, I didn't want to let everybody go and, like, let everyone down. Like, if I left, then everyone else would have gotten, like, screwed in some capacity.

[22:22](https://realalpha.app/v/uYzQxxAD41g/?t=1342) Never crossed your mind. Let's just sunk costs, you know, move people to something else. There's a bigger goal that I have. I think I... I assumed that if, like, the first company basically didn't work, and the reason the first company didn't work in some ways is that I kept trying things over and over again and just, like, doing something different, and I think I was assuming that if I quit this one, even though to some extent, like, this wasn't a pivot, this was, like, a forced into this sort of thing, I think I just assumed that I would, like, never be able to get a job or fundraise or do anything ever again.

[22:52](https://realalpha.app/v/uYzQxxAD41g/?t=1372) I don't think I'm very good at working for other people. Could you tell it was going to be a huge market if you guys could survive for, you know, as Because you're going month to month in those early, early moments. I don't think I actually thought that much about the scale in the early times. Because at the time, we had thought that $5 million was a lot. Compute does not work like other markets. A mid-sized deal in Compute is bigger than many venture funds' entire AUM. A mid-sized deal in Compute is bigger than all the money that will ever be made by a whole category of SaaS companies. So you just announced this $245 million contract, right?

[23:25](https://realalpha.app/v/uYzQxxAD41g/?t=1405) **Evan Conrad:** We were almost not going to announce that because I thought it was like, it just isn't that big of a deal in compute.

[23:31](https://realalpha.app/v/uYzQxxAD41g/?t=1411) Like, it's not that large of a contract relative to other folks in the space. It is one that we take really seriously. It's one that's very important to us. Relative to the space, it is not that large. Like, we are not that big of a player in compute. How long do you think it took you to go from survival mode to expansion, trying to own more of that stack? Think about sort of the big vision of the company. When would that have been? We're still in survival mode. Most of San Francisco compute has just basically been, okay, we have this cluster. Oh, shit, we don't have enough access to this cluster.

[24:01](https://realalpha.app/v/uYzQxxAD41g/?t=1441) We can't do the thing we want to do unless we like dive deeper and deeper and deeper and deeper and deeper. But you hit bottom eventually. Yeah, so bottom for us is probably like land and power. But even land and power, like there's stuff that goes into getting land and power. I think there's more places that we can go. But the thing that's difficult here is we don't want to become a cloud. We're not a neocloud. How do you define a NeoCloud for folks who are not as plugged in? So I would say like a NeoCloud is somebody who goes and buys hardware and then sells the hardware. We don't do that. We do that for like test clusters.

[24:34](https://realalpha.app/v/uYzQxxAD41g/?t=1474) We do that for POCs. But we don't like the vast majority of our capacity is stuff where we're working with partners. With the NeoClouds, would they work with you or are they just a whole different part of the ecosystem then? We think the NeoClouds will eventually work with us and we would like to work with the NeoClouds. We think we're very complementary to them. Because if we exist, then what happens is GPU clusters become wildly easier to finance. And that is the biggest constraint in setting up new clusters. It's just... credit. The problem, though, is the neoclouds themselves can't make a forwards market like this because if you own the assets, you can't be the market. No one will trust you. It has to be an independent thing. Like Wall Street won't trust the market if it's manipulatable, basically.

[25:15](https://realalpha.app/v/uYzQxxAD41g/?t=1515) Thus, somebody needs to make it. And in order for them to make it, you have to make physical delivery, which means that somebody has to operate a GPU cloud or do many of the parts of operations of a GPU cloud. If you go out to build the capacity within your company to build GPU clouds, then you have to turn down the opportunity to be a GPU cloud. And that is such a ridiculous financial incentive to not do that that almost everyone who ever looked like they were building a compute market either did one of two things. They either went up the value chain and they became a platform as a service company or they just became a GPU cloud because you make so much money to do that.

[25:52](https://realalpha.app/v/uYzQxxAD41g/?t=1552) To actually build a physically settled compute market, you have to be turning down money left and right, basically. Is that where people might have historically thought you guys were crazy? Or where do you think San Francisco Compute has been most an upstart in kind of the philosophy or the approach that you're taking? For like the last 30 years, companies were really simple ideas. And you could easily explain them in like a single sentence. Welcome to fucking compute. It is not a simple sentence. Like, I'm sorry, there isn't an elevator pitch. You're going to have to understand finance. You're going to have to understand supercomputing.

[26:27](https://realalpha.app/v/uYzQxxAD41g/?t=1587) It's the fucking supercomputing company. It's not going to be a simple explanation. It's really easy to explain the problem. It's really easy to say, look, everyone sold out, you know, big, long lead times. You have massive deposits. Everyone can see the problem. When we've explained the company to people, most people didn't want to engage because it was a really fucking complicated problem and a really fucking complicated company. And was this investors? Was this employees, potential customers? Who were the people who were shying away? Our early rounds were quite difficult because of this. We would go and try and talk to folks, and if you couldn't explain it in a simple sentence, most people either assumed that we were competitive with something we were not competitive with.

[27:05](https://realalpha.app/v/uYzQxxAD41g/?t=1625) They assumed that they just didn't understand the company. Good portion of that's my fault. There's both the complexity of the problem, and then also you do actually have to be able to explain this to some extent. But some part of the company is just irreducible complexity. If you had to describe the company as much as possible in a sentence today, have you practiced it? Can you give us a shot? With a power grid for compute. I mean, that seems pretty simple to me. Super simple. But it took you a while to get there. No, it was the immediate first sentence we would say. Really? And that just didn't resonate?

[27:36](https://realalpha.app/v/uYzQxxAD41g/?t=1656) The immediate next question is, why would you need a power grid for compute? And if you keep pulling on that thread, you end up realizing that the entire company is extremely complicated. I think most people would like to look at the company and see a traditional software company that has traditional software problems. Or just a very simple marketplace where it's like, okay, there are GPUs over here, everybody wants them, these guys matchmake, they take a spread. simple, right? Yes. Not so simple. It's not so simple. Nothing about the way that you would think you would make a GPU market, nothing that intuitively makes sense works.

[28:09](https://realalpha.app/v/uYzQxxAD41g/?t=1689) And we experienced that really directly because we started with step one and then we just kept going and going and going. So many of the things that we do don't make a lot of sense to people if you just think we're like Airbnb for computer or something. Does this become more obvious if compute does become more of a futures market or a commodity like the sort of, you know, the electricity or corn or something like that? So I actually think that futures markets are a good example of the simplistic thing doesn't work. I'll give you an example. Let's say we're doing futures for wheat. What happens in the futures for wheat is there is some world and maybe you can actually get wheat in the future, like you actually get delivered that wheat.

[28:52](https://realalpha.app/v/uYzQxxAD41g/?t=1732) The problem that doesn't really work for compute is that compute is not very fungible. It's kind of fungible. You can make it fungible, but that fungibility depends on a massive number of different types of variables that are not factored in when people mostly think about the compute price as just a single number. One example is that depending on how your VM implementation works,

[29:19](https://realalpha.app/v/uYzQxxAD41g/?t=1759) **Evan Conrad:** you can have a 40% performance difference on GPU kernel launches from one operator to another.

[29:26](https://realalpha.app/v/uYzQxxAD41g/?t=1766) Another example is if you had the early H100s, they were shit. And if you had the later H100s, they were great because there were like problems in the initial ones and they got fixed over time. If you're talking about just like an H100 price, that's not the same thing. Like it matters which one you get. It depends on which one you get. Many people today are trying to do a cash-shuttled feature. And the way that works is that you go and you develop some index, you build some sort of tracker of price. and then you go and try and build a cash-shuttled future on top of it, which means that if you can bet on what that index price is, and if you're right, you make money, you get paid in cash.

[30:13](https://realalpha.app/v/uYzQxxAD41g/?t=1813) You don't get paid in compute. The problem of that is, let's say you end up in a situation where you would like to take that cash and turn it into compute. The price that you want might be completely different based on the actual cost of compute. And so you have de-risked yourself on an index price that is divorced from the actual cost of compute. That's kind of the problem of making a cash-settled index if you don't have physical settlement. What is it that you guys are doing that is different from that cash-settled futures market that you don't think is working? If you buy GPUs from us that start in November, you get GPUs.

[30:51](https://realalpha.app/v/uYzQxxAD41g/?t=1851) You don't get cash. That's the difference. And so that means that you can use that to risk and de-risk yourself with actual settled GPUs. Like you get the VM time or you get the hardware time. What about if I could use the compute, but other people will pay me such a premium for that compute that it would make sense for me to put it on the market anyway? Today, a lot of people, when they're sort of running their product on GPUs, they make some margin. Let's say that margin is 50%. One option that you could do is that you could run your product, make 50% margin. Another option that you could do is that you could sell that compute for more than 50%.

[31:31](https://realalpha.app/v/uYzQxxAD41g/?t=1891) And then you make a profit either way. One of the things that SF Compute allows in our market is for people to buy that capacity and then as prices go up to sell it, which is a sort of attractive opportunity for you. You sort of de-risk yourself against your own product not making as much money as you thought. Your vision of the future, when you guys are at the scale you want to be, what does this market look like and what is SF Compute fully empowering there? We want it to be the case that if you go buy GPUs, there's a price you can buy it at. Right now, that is just straight up not the case.

[32:01](https://realalpha.app/v/uYzQxxAD41g/?t=1921) On anywhere other than SF Compute, almost everybody sold out. On SF Compute, you can buy capacity. Like, that's cool. Step one. Step one, guys. We did it. Then, if you want capacity, and you want a lot of capacity, you can pay more for a lead time that is shorter. On SF Compute, there's a maximum amount of capacity that you can get. You can't go beyond a certain amount because we either don't have that capacity for sale, or maybe we just don't have that capacity total. We want it to be the case that you can pay for a shorter lead time. There's a bunch of different ways that we can make this happen, but that seems like an advantageous thing that should exist.

[32:35](https://realalpha.app/v/uYzQxxAD41g/?t=1955) Then, we want it to be the case that when you go to buy capacity, you don't have to pay as big of a deposit. Either you don't pay any deposit, or you pay very little deposit. The reason why that's possible is that if you know what the price of compute will be in the future, the lender doesn't have to underwrite the credit risk of the customer, they have to underwrite the price of compute. That is a super advantageous thing to exist, and we think that if you make a commercial forwards market, But you can create a future in the future that solves that problem. Once you have the capacity, we want it to come online sooner and be delayed less often.

[33:09](https://realalpha.app/v/uYzQxxAD41g/?t=1989) The way that you make delays not happen is that the cloud was able to go buy the capacity before they had the customer in hand, which they were able to do because the lender was pretty confident on what the price of compute would be in the future. So they knew what it was going to be sold at. When the cloud can place the purchase well before the customer has even seen it, then by the time they're selling the capacity, they already have the hardware, potentially. Or the construction process is very de-risked. And so the delays are way less likely to happen. Right now, this is like financially infeasible for most people.

[33:45](https://realalpha.app/v/uYzQxxAD41g/?t=2025) Once you have gotten the hardware, we want to make it possible for you to have fewer types of failures. And the way you can make fewer types of failures is you have a market that has very liquid. You can pay to swap in a node when another one fails. Right now, what happens is when you get a cluster, you first do a burn-in, which kills a bunch of nodes, and then those nodes that die go through an RMA process, meaning you have to go get new nodes, and sometimes you'll have spares, but not necessarily always, which means you have to wait for more capacity to come online. And you should be able to pay for that capacity to come online now, which then reduces the experience of broken clusters.

[34:21](https://realalpha.app/v/uYzQxxAD41g/?t=2061) For people who are like, okay, well, I keep hearing that compute is going to be scarce for the foreseeable future. At the same time, data centers have become a big political talking point. And I hear about these huge cloud players talking about doing these multi-billion dollar projects that then may or may not actually happen. Yeah. What is your bet for sort of how this all shakes out and where you guys fit in that sort of bigger context moving forward? First is I think GP prices are going to go up and to the right. I think it is correct that we are in a scarcity that demand is outpacing the supply.

[34:56](https://realalpha.app/v/uYzQxxAD41g/?t=2096) And that that continues. And that continues, yeah. I think that's locked in for the next, like, four to five years at least. When we think about compute as a commodity, what are the impacts, the ramifications of it being so important, so valuable? If you think that compute is going to be bigger than oil, you should probably assume that there are geopolitical impacts of it being bigger than oil. And right now, one of the things that is happening is you've got data center moratoriums that are pushing compute out of the United States, But the AI labs are in the United States. The bottleneck for setting up new clusters is not actually power and land, but that's very useful to have.

[35:38](https://realalpha.app/v/uYzQxxAD41g/?t=2138) It's the credit and the revenue streams that finance the compute.

[35:44](https://realalpha.app/v/uYzQxxAD41g/?t=2144) **Evan Conrad:** And so one of the things I would assume will happen is the leverage that the United States has to shift where those capital flows go is going to probably become a geopolitical asset.

[36:01](https://realalpha.app/v/uYzQxxAD41g/?t=2161) And the way to think about this is you've got projects like Stargate that are $500 billion. $500 billion is the entire GDP of many countries. You're talking about projects that could 5%, 10% growth of other folks' GDP. and if that project lands in their country, that is a really big deal. Our states are probably not acting in their best interest when they're blocking data centers, but the rest of the world is going to act in their best interest. And that becomes a leverage point for the United States to use in geopolitical negotiations. Startups obviously think often in 10-year horizons. If, let's say, Compute did switch into abundance at some point, would the role of your company really change or no?

[36:48](https://realalpha.app/v/uYzQxxAD41g/?t=2208) Okay. So, we were in abundance at one point, so SFC went through a glut, or the whole market did. There was a period in which H100s were just, it was really hard to sell H100s. How long ago was that? A year and a half ago. Okay. You guys still found business otherwise? Yeah. Basically, one part of the market at any given point sort of is stronger than the other. We're asset light, so there's some trade-offs of that, which is that we don't make as much money as you would if you were asset heavy and you were experiencing the upside. The benefit of that is that you get to build something very durable over a long period of time because you aren't, you're sort of immune to market swings.

[37:27](https://realalpha.app/v/uYzQxxAD41g/?t=2247) If there was one takeaway. For a wider startup audience, that you could teach everybody to just have a better understanding of compute and how this market works, what would be the top thing?

[37:39](https://realalpha.app/v/uYzQxxAD41g/?t=2259) **Evan Conrad:** Right now, if you have a B300, depending on what you're running and what you're doing, you can make somewhere between like $20 to $50 an hour on that B300.

[37:50](https://realalpha.app/v/uYzQxxAD41g/?t=2270) Right now, B300 prices are $5.50 an hour, maybe $6 in some cases. I would expect those two numbers to come much closer together quite soon. If you are buying GPUs today, I would take that into account. And I sort of recognize that I am a participant in that, but I also just want to be super frank. The alternative for anyone who owns a GPU is to simply sell that GPU instead for $20 to $40 or something an hour, rather than sell it to somebody else for $5.50 or $6. Is this a good time to want to need those GPUs, or is it a better time to own the GPU? If you have really great credit, if you have power or land, or you have a really strong reason to want those GPUs for maybe your portfolio, then you should totally own GPUs.

[38:50](https://realalpha.app/v/uYzQxxAD41g/?t=2330) And otherwise, you should probably lease them. All right. Thank you, Evan. Of course.