Hi, I'm Lip-Bu Tan, founding managing partner at Walden Catalyst and CEO of Intel.
We are proud to partner with Jasper and his team to lead Hyperbolic's $30 million new funding round. Hyperbolic is tackling a foundational challenge in AI infrastructure, bringing fragmented GPU capacity together so AI developers can easily access, manage, and orchestrate their workloads.
Adding more hardware alone won't solve the shortage. We need a better platform to make global compute capacity more liquid, unified, and productive. We believe Jasper and his team are building a disruptive platform, and we are excited to support their hyper growth as they make AI compute flow across the system.
Congratulations to Jasper and the Hyperbolic team.
Lip-Bu, thanks for inviting us, and we're very excited to join the Walden Catalyst family. We are delighted to be a part of that financing and to support you in this hyper growth opportunity that you have.
Thanks. And a lot of people are thinking about the compute market in general, and they're starting to hear news about people worrying about overbuilding the AI infrastructure. What's your take?
I think this AI initiative and the momentum will continue, even bigger than the internet. And I think the question mark is how quickly can you scale?
And then a couple of things: one is the memory shortage, the other is the GPU shortage, not available, and of course the power shortage. And now the other big thing is basically the financing, and that becomes the bottleneck of the growth. And so I think you're going to see some correction, but overall momentum will continue because this is just the beginning, and you're going to have thousands, millions of agents, and then how do you manage that? And then the other part that will be challenging is the security.
So if you have so many zillions of agents, how do you safeguard them and not have them move into different uncharted areas that they're not authorized to do? So those are the things — it's just some of the correction, and we need to address that. Eventually, we'll continue to grow.
Cool. What makes you excited about Hyperbolic?
First of all, I think we are so impressed with you. Thanks. You finished a PhD at Berkeley in two years — that is unheard of. And then secondly, you're a mathematician; you are well known for that, and we were very excited about that, and also your experience in this field and also at Citadel as a market maker. How do you drive that change? And then your idea about this whole GPU exchange, and then providing the infrastructure, the software to meet some of the developer requirements for this hyper growth. You know, from the beginning, only 14 months, you're already at a $100 million run rate, and that's amazing. And we want to be part of it and to be financing you and supporting you.
How did you come up with this idea?
So I would say when I was doing my PhD, besides math, I was also doing a little bit of AI. So I first experienced the compute shortage. The lab only had 30 to 40 GPUs at a time, and you couldn't actually do any meaningful experiments for AI. And so I kind of realized that most AI innovation will be bottlenecked by compute. And then when I started working at Citadel Securities as a quant researcher on market making, I got my first experience of the market, like trading, and understanding the market. I kind of got to understand what is liquidity and how to actually provide liquidity to connect supply and demand. So when ChatGPT came out, a lot of people were thinking about building different AI companies, but then they kind of started realizing that actually the idea is not theirs, not their expertise in AI — it's how they can get access to compute. And they were starting worrying about balance sheet, they're worrying about burn, they worry about risk. And so I feel like, yeah, if you are a talented researcher, you shouldn't worry about those infrastructure financing stuff. You should just spend most of your energy building the next big thing. And so we really wanted to spend time to help them solve that problem. So when I looked at the market, I realized that a lot of people are building GPU clusters, which is good, but gradually the market is getting more and more fragmented.
There are like 300 plus neoclouds across the world, and the hardware is very heterogeneous, right? Different GPU types, different networking, different storage, even the access to the compute varies differently. And so we came up with the idea: okay, we really wanted to build a solution that can aggregate and turn heterogeneous, fragmented compute into one unified platform.
So that any researchers, developers, any AI companies will be able to access the compute in a very flexible way. They can manage all the compute, they can orchestrate the workload across the globe's compute, and then whenever they don't need the compute, they also can sell it back. And ultimately, we just want to put the world's compute behind the ones who are bold enough to train the world, and they can just focus on that.
In fact, you know, I'm talking to some of the AI developers, and they all mentioned that they have a hard time getting GPUs. So I think they all point to you to get the GPU from you. So clearly your idea of that exchange, that you can provide to the developer the GPUs they need for their development and then for the model. And then the other part is, you know, a lot of customers I talk to, even though they have GPUs, they are all worried about the pit group. So some of the time they only use 25, 30%. And so here you come, that you can provide that extra and create more value for some of the developers. They're going to start, they want to have more compute, you know, for the GPUs, and you provide that exchange. But more important to me, if I understand correctly, and then you also provide not just the hardware, you also provide the software and the whole full stack, and so that they can manage more effectively for running compute requirements. Am I correct?
Yeah, totally. So I would say in our company we are building two layers. The bottom layer we call the first global compute grid, just like we believe that compute is the new generation of power, just like electricity, and it's going to power the AI era. And no one wants to get stuck with a three-year contract. They want to be able to draw compute elastically, and that's why we built the unified control plane solution called Forge.
That allows us to take any AI compute and attach it to our control plane, virtualize it, so now it can be shared with multi-tenants globally in a very flexible way. And then, like you said, if there are companies who don't need the compute off-peak hours, then they will be able to also sell back to the grid.
On top of that, we know that AI companies and even enterprises, they don't want to just get access to hardware; they need a lot of AI solutions, and that's what we feel very excited about for the next phase. We're already building very, very fast on providing developers the necessary tools that they need, for example, managed Kubernetes. We launched inference like more than two years ago, and we attracted 250k developers on our platform. And now with the new financing, we're expanding our team on the technical side, and we're leveraging agents as well and building fast on providing all the nice services that the customers need.
