San Francisco Compute's Evan Conrad wants to build a utility-grade compute market — starting with the biggest supercomputer ever

Speaker 1: He's the co founder and CEO. What's up? Overdue.

Speaker 2: So happy to only about a year overdue.

Speaker 1: Yeah. So But thank you so much for Okay. Coming on. I already know what we're gonna talk about, but please introduce yourself and the company a little bit to give everyone some lay of the land.

Speaker 6: Hey. I'm Evan. We make a company called San Francisco Compute. We're the inventor of compute markets.

Speaker 1: Okay. And what is the current state of compute markets? Mature are Where do you want these to go? Break us down for the status quo and then we can start talking about the the the future.

Speaker 6: So our impression is that right now, compute kind of looks like at the early days before there was a power grid. Mhmm. And when that happened, every factory ran its own generator.

Speaker 1: Mhmm.

Speaker 6: And they had to schedule that generator to exist for peak capacity. Okay. And then during idle hours, they would just sort of burn that money. And so what happened in the power markets is that utilities came in, and they pooled demand between all of these different factories. And then by doing that, you made the economics of all the factories better, and this is how we scaled up electrification of the country. This is what people are trying to do with compute markets. It's what we try are trying to do with compute markets. But I think at the moment, there's whole bunch of folks who are entering the space, and the way that they're trying to do it is they're typically trying to create some sort of index price, and then you can create a cash settled future on top of this. Our impression is that by doing this, you're divorcing it from the actual technology under the hood because the index isn't connected to anything. Like, you're not actually buying the compute. You're sort of trading on this, like, magical number.

Speaker 1: Yeah. Is there a solution here where you look at the semi analysis cluster max three, you view that more like a Moody's credit rating and you assign different tranches of compute. Because maybe we're not in a fully homogenous market for compute yet, but it's actually reasonable to trade CoreWeave for Nebius, which both earned the top ranking in ClusterMax three. And then you go down the list and maybe there's a little bit of a delta in value between compute. But within a particular tranche, if they're all rated triple a or triple b, it's okay to actually treat those as fungible. How close are we to that?

Speaker 6: I think the real problem is that you just don't have very many people at the top of that ladder.

Speaker 1: Okay.

Speaker 6: And so one of the things that needs to happen before we get to compute markets Mhmm. Is you need to sort of level up the whole market. Right now, the folks that are easiest able to enter the market are the people with lots of capital, lots of land, lots of power Mhmm. But not necessarily a lot of, like, supercomputing expertise. Mhmm. And so I think that's why we're in the current stage of the market where you have a few excellent players Yeah. Like Cory or like Anivias.

Speaker 1: Yeah.

Speaker 6: But you don't really have enough actual participants Sure. That know what they're doing in order to make it possible to have a market.

Speaker 1: And what is the the the key unlock to to raise the floor here? Is it the actual chipsets, the networking technology, the software that's installed? Because I know that there's variation across basically every dimension, and that feels like the enemy of commoditization. But at the same time, there's a huge amount of companies that are saying like, look, you're not going to commoditize without me, AMD, having a say or me, another networking provider having a say. Everyone wants to bring something to bear.

Speaker 6: Yeah. So our impression or our view at San Francisco Compute Mhmm. Is that you need a market operator that has to be independent from the other con sort of other clowns. Otherwise, if it's not independent, Wall Street won't trust it. Okay. That market operator also needs to do what's called physical settlement. And in this case, settlement, we think, means running the whole damn thing. Sure. You've got to run it all the way down to the the data center itself. That includes even building the data centers in some cases.

Speaker 1: Interesting.

Speaker 6: And if you do that, then you can standardize because you can make the whole operator be the same thing, and then you can plug in capacity. You can plug in land and power and other assets into it, and you get lots of different market participants, but that's the way the grid works. There's a grid operator, and they sort of run and manage things that allows it to standardize. Right now, we're in the world where people are, you're, like, trading capacity. And imagine if we did this in the power market where I traded you a coal plant for a nuclear power plant. That's basically what we're doing right now.

Speaker 1: Interesting. I mean, at the same time, like, are there are, you know, plenty of heterogeneity in the in the oil refinery business. But then once the oil gets to be a standard crude and it's graded, then it can be traded as a commodity. Is there not some like doesn't isn't that a prerequisite that basically the vast majority of compute comes online in some homogenous market to really create like the commodity vision that people are seeing in the future as opposed to just more standardization within individual neo clouds or or compute providers? Doesn't something have to happen at a higher level of abstraction?

Speaker 6: So to some extent, yeah. So this is what we think of ourselves as doing Sure. Is by making that refinery exist. Yeah. On the other hand, it's just way more nuanced than that. These are supercomputers, not soybeans. Yeah. And so there's just like yeah. It is just frankly more complicated.

Speaker 1: Mhmm. Make it easier, if only. So, yeah, what is your what is your plan right now? What's the stage of the company? I mean, you've raised money, but how close are you to do do you think to acquiring power, a data center? Is that is that the plan in the near term?

Speaker 9: Yeah. Okay.

Speaker 6: Can't say that much, but Okay.

Speaker 1: It's in it's in the works.

Speaker 2: Doing pretty well. Computer ever. Yeah.

Speaker 6: Yeah. We're we're shooting for the biggest computer ever.

Speaker 1: Yeah. And and is it is it not viable to I mean, should I think of the the hyperscalers as sort of building a like a fungible layer on top of all the neo clouds? Because you when you when you look at the neo cloud, so many of them have partnerships with AWS, GCP, the big AI labs. And in some ways, could imagine some commodity some commoditization happening like in the AWS console. But but do you think it's viable to try and come over the top and and create a new platform that marshals compute across all the different neo clouds? Or will there be like neo cloud revolt for that or organ rejection? Because they will only do that if it's Amazon and they have to play nice.

Speaker 6: So our impression is that there's a missing utility. Mhmm. We think of like the hyperscalers as kind of the old school factories. Mhmm. And they've got these sort of generators or power companies that are working with them. And they totally they've sort of abstracted over the the power companies, this case the Neo Clouds is sort of that analogy. And now they're producing goods from their factory.

Speaker 1: Mhmm.

Speaker 6: The problem is if you want to really scale everything up, you need to figure out how to make a market that can secure the risk for other people to scale things up as well. And if you do that, that's great for the existing participants. Like, that's gonna be great for the hyperscalers. It's also gonna be great for the neo clouds. But that market needs to be independent in order to exist, and the operation of the market, the technology needs to be co designed with the order book itself. Otherwise, you can't actually do this.

Speaker 1: Yeah. Yeah. That makes a lot of sense. Well, congrats on the progress. Good luck. I'm excited for the next announcement. Come back when it's live. We'd love

Speaker 2: hear Yeah. Anytime. Anytime. Last last Please do. Last question. There's someone on X that's been posting about I'm not just gonna keep posting through this infrastructure build out. I'm getting in in the action. I'm levering up personally to get compute to be able to participate in this. Sure. What are your what are your thoughts on the on the humble, on the humble home supercomputer operator that wants that wants to get into the token flow? Is it an should they just be buying NVIDIA, you know, and be like levered long there and just sit back and do nothing? Or should they have their own rack at home, a little bit of power, and be be feeding the the token stream?

Speaker 6: So I think I know who you're talking about, and I would say they are they are doing more than just, you know, a rack in your your house.

Speaker 1: Mhmm.

Speaker 6: Okay. What what we do at San Francisco Compute is we help people become clouds, and then we help them sell those clouds on the market. We're like a supercomputer developer. And we think well, we know. We've seen this that all the money and all the capital and all the power is sort of trying to figure out how to turn itself into GPUs. Mhmm. Because there's a really strong economic reason to do that right now. And so we're seeing this happen across the market. Our gist as a company is we're going to help them do that and help make it be possible that there's a large amount of new neo clouds, new NCPs in the NVIDIA terminology. And then we're going to help them get financed by building the market that makes it possible to derisk them.

Speaker 1: That makes a lot of sense. Well Thanks for

Speaker 2: I have one Please. My second last question. Okay. And and sorry sorry for keeping you a little bit over. But you you clearly love to use these different historical analogies. And I'm wondering when when people build up bring up sort of like railroad build out and and overbuild and and try to comp it to this present moment. There are certainly variety of of strong comparisons that you can make. What what's your reaction to that? Where do you think that sort of analogy falls short? Would say one one of them is like, if you just build a railroad effectively to nowhere and there's not a lot of economic activity there and you invest all this money into it, and then there's no not that much value creation, it's obvious why those kind of things can blow up. Whereas with compute, if you can build a bunch of compute, the the current state of things is like the compute will the the demand will find it one way or another. Even if you're a tier three, four, five operator, if you have Fundamentally, if you have power and GPUs, someone will find you and and sort of force you into being competent and economically viable one way or another.

Speaker 6: Yeah. So I think kind of what you're getting at is is there an AI bubble? Are we out overbuilding like crazy? This is this is the this is the problem we're trying to solve. Right now, it's important to understand where that bubble is.

Speaker 1: He's solving. He's solving the

Speaker 5: bubble. Thanks.

Speaker 6: Yeah. So it it the bubble itself

Speaker 2: Someone had to do

Speaker 1: it. Yeah.

Speaker 6: It works like this. There's the cloud itself doesn't deploy the GPU cluster until they have the contract in hand from the customer. Mhmm. And then they force the customer to pay them a massive deposit up upfront so that they can give it to the OEM who then gives it off to NVIDIA. Mhmm. But the like, the AI lab itself doesn't have that capital yet because they need it for the future. Mhmm. And so they go off and they raise money at massive valuations. Mhmm. And all the VCs look around and say, shit. This AI lab, it's it's asking for a crazy valuation. And is that is that okay, guys? Like, is everyone cool with that? And they all look around and say, well, I guess we missed out on OpenAI and Entropic. And so maybe these people will be OpenAI and Entropic too, and so we'll go with the the big valuation, but then the venture capitalist owns the risk. So if you're looking for a bubble, it's held currently by the venture capitalists. We're we're trying to solve that problem by derisking compute in the future, by basically making it possible to know, like, what the price of GPUs will be off into the future. And that reduces a lot of this problem because then you don't have to have the big deposit upfront from the customer, which then means you don't have to have the big giant valuation that's probably maybe a little overvalued from where it should be. Yeah. But

Speaker 1: And then you're the king in the castle. King in the castle.

Speaker 2: John's out of control.

Speaker 1: I have a problem with the you asked questions Great

Speaker 2: point, Evan. We'd love to do this again soon, and we'll we'll make sure to get more time next time.

Speaker 1: Yeah. Have a good one. We'll talk

Speaker 2: to you soon. Thanks guys

Speaker 1: to Hank. Goodbye. Cheers. Let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents whether you're writing code, analyzing data, creating content or automating business workflows.