01:57:35 · John Coogan, Jordi Hays, Amiram Shachar, Mark Gurman
[01:57:35] S1: So your job is to take the prompt from the CEO and give it to Codex and Claude Code. [01:57:39] S0: And I could just dictate. [01:57:40] S1: To the agents. Kind of... It's [01:57:43] S9: more or [01:57:44] S1: not. Yeah. [01:57:44] S9: And then I asked him, so, like, so that's what we're doing right now. Okay. So, like, we're just prompting. Yeah. And then he said, like, yeah, like, my coding, like, time has dropped down by 90 [01:57:55] S1: Woah. [01:57:55] S9: But the review time is up by 600%. [01:57:58] S1: Sure. Sure. Yeah. So So part of the midnight oil. [01:58:01] S9: So I think, like, you have to be really, really smart on, like, how to review that. [01:58:03] S1: Yeah. I think Karpathy had a take about this where he said, like, your [01:58:07] S0: Review this. [01:58:07] S1: Your code Review this, ability atrophies while your code reading ability increases. And that's probably more important than ever in the modern era. Fascinating. Well, tell us about the round. A series c, how big was it? How much did you raise? [01:58:22] S7: 400,000,000. Let's [01:58:25] S0: go. Congrats. He's not playing. He's playing. Let's not go. [01:58:29] S7: That's... You know, that... That's a [01:58:31] S9: lot of money. That's a lot of money. And I I I can tell you that this is what's needed [01:58:34] S1: Yeah.
01:58:34 · John Coogan, Amiram Shachar, Jordi Hays, Jasper Zhang, Zach Rash
[01:58:34] S1: Yeah. [01:58:35] S9: To operate in that very important space. Yeah. Like, it's a super important space. You have to build. You have to innovate. We have big dreams now. Like, we can translate those dreams into execution. We have the right amount of cash and lots of trust from investors, a lot of trust from customers. And yeah, it puts us as like one of the most capitalized companies on the planet to secure that cloud and AI. [01:59:01] S1: Yeah. We need it more than ever now. We saw it with the PHP hack. So thank you for everything you're doing. [01:59:05] S0: Super important. [01:59:06] S1: Good luck, and thank you so [01:59:07] S0: much for your [01:59:07] S1: time to come on the show. Have a great rest of your day. [01:59:10] S9: Appreciate you guys. Thank you. [01:59:11] S1: Thank you. Dude. Let me tell you about the New York Stock Exchange. Wanna change the world? Raise capital at the New York Stock Exchange. You're trying to get more than 400,000,000. You gotta head to the New York Stock Exchange. Our next guest is Jasper Zhang from Hyperbolic, cofounder and CEO. His first appearance on here. Jasper, how are doing? Welcome to the show. [01:59:33] S4: Hey. How's it going? [01:59:34] S8: Sorry for keeping you here, guys. [01:59:35] S1: Thank you so much for hopping on. Please introduce yourself and the company.
01:59:35 · John Coogan, Jasper Zhang, Jordi Hays
[01:59:35] S1: Thank you so much for hopping on. Please introduce yourself and the company. [01:59:41] S4: Hey, everyone. I'm Jasper. I'm the CEO of Hyperbolic. I grew up doing math competitions and won a Fugal medals at Chinese Math Olympiad, Alibaba Global Math Competitions. I finished my math PhD at UC Berkeley in two years, which made me the fastest person to finish a PhD in official Berkeley. [01:59:56] S0: What stopped you from doing it in... What stopped you, just out of curiosity, from doing it in one year? [02:00:02] S4: Like Oh, Berkeley has a residency requirement. Oh. Technically, I need two years. Hands retired. [02:00:07] S1: That's hands retired. [02:00:08] S0: So without that requirement, you think you could have done it in a couple months? Maybe a month? [02:00:13] S4: With AI now, I think so. Yeah. [02:00:14] S0: Oh, really? Okay. Okay. Did you... Have you ever faced Scott Wu? [02:00:19] S1: Oh. Yeah. [02:00:21] S4: Oh, yeah. I mean, I heard about him. Yeah. Haven't haven't met him in person. [02:00:25] S0: No. No. I meant, like... But you guys were kind of competing in different Yeah. Regions, it sounds like. I was hoping that we could find a video of you guys, like Going head [02:00:33] S1: to head. Yeah. Like like old LeBron videos. Yeah. Exactly. High school. Sparring. Yeah. That'd be great. But tell us about the company. What are you what are you building? How how how has it been going? [02:00:45] S4: Yeah. It's been pretty great. I mean, Hyperbolic will aggregate and turn heterogeneous compute [02:00:51] S1: Mhmm.
02:00:51 · John Coogan, Jasper Zhang
[02:00:51] S1: Mhmm. [02:00:51] S4: Into one unified platform. [02:00:52] S1: Okay. [02:00:53] S4: So it's a platform where people can just come access compute in a very flexible way. Yeah. They can manage and orchestrate the compute globally and then sell the compute back [02:01:00] S1: Yeah. [02:01:01] S4: When they don't need it. [02:01:01] S1: How heterogeneous are we talking about? Because we saw a a cluster of 9,000 a one hundreds on eBay earlier today for something like $60,000,000. Then, you know, you got people with a gaming PC that might have a, you know, fifty ninety NVIDIA card in it. How low can you go? What is the practical scale to that heterogeneity? [02:01:23] S4: Yeah. Totally. So our technology support any kinds of, GPU chips. Mhmm. But then, technically, I think most customers want at least the the a 100 plus [02:01:33] S1: Okay. [02:01:33] S4: Chips. We're also starting seeing people using Pro 6,000 for inference as well. Right. We can go as high as to Blackwell and Vera Rubin. [02:01:41] S1: Yeah. How much spare compute is out there? Because you you you you... It feels like every day there's a new story about, like, oh, there's such a compute shortage. Like, tokens are selling. Like, you know, the the the spot market is in Contango and, like, the... There's so much AI demand, and yet you're still finding folks out there that have spare capacity to bring on the platform. Is that correct? [02:02:06] S4: Yeah. I would say we're not just tackling spare capacity. Even for, like, a big cluster, our goal is to maximize the revenue per [02:02:12] S1: GB hour.
02:02:12 · John Coogan, Jasper Zhang, Mark Gurman
[02:02:12] S1: GB hour. [02:02:13] S4: Right? Like, technically, you can sell off take for b 300, $445, $5. [02:02:18] S1: Mhmm. [02:02:19] S4: But it's, like, very low margin business. Mhmm. If everyone... Even if you're on Neo Cloud, you want, like, to maximize the revenue to $10, $20 per GP hour. Sure. You can work with us. [02:02:28] S1: Yeah. And so are you seeing... Like, what is driving that incremental offtake? Is it that there's a company that is renting a cluster and they're rent... Doing a training run, then they pause for a little bit to review the results, and so you can go do some inference right there? Or is it you're doing stuff on the other side of the world when everyone's working with the big models during their workday, but then there may be the GPUs aren't on fire in the middle of the night so you can sell inference somewhere else? Like, what are the use cases that are... That that your customers are coming to you for specifically? [02:03:01] S4: Yeah. So I would say what are the other cases I mentioned are valid for us. [02:03:05] S7: Mhmm. [02:03:06] S4: And I would say we start with first helping customers to get access to compute first. [02:03:09] S1: So,
02:03:09 · John Coogan, Jasper Zhang
[02:03:09] S1: So, [02:03:10] S4: you know, like, there are more than 300 plus neo clouds. The market is super fragmented. Yeah. And it's the first time that technology company, founder, need to start worrying about sourcing procurement, but also the cash flow, the capital, the hedging, and risk management. Yeah. And so what we do is that we kind of have, like, the system, which is called Forge, which is a unified control plan that allow us to take any bare metal cluster, attach it to our platform, and now it can be shared globally with different customers in a very flexible way. So they will start using the compute from our platform. And then let's say if they lock in for three year contract or, like, a few a few year contract Yeah. Then like what you said, if they are doing training run after the training run, they don't need that. They can also sell it back to the market and the same for Visa, Inference too. [02:03:59] S1: So Intel's CEO, Lip-Bu Tan, is backing up the truck for this company. What is your interaction with him and Intel? He's backing up the truck investing. But, you you you know, obviously, he's, you know, an incredibly important CEO in the AI world, but it's not like he's running a neo cloud that needs offtake capacity necessarily. Like, there's not a super logical business relationship that I can think of. But do you see this growing into a strategic relationship, or is it just you're... You got along great and he had the capital to invest?