Founding Story

The Hyperbolic story didn’t start with us. It started with the stories of others:

A Stanford postdoc was forced to pause his research after being priced out of the thousands of GPUs needed to push the limits of LLMs.
A Berkeley professor struggled to find a sufficient machine for her research on confidential computation of AI, aiming to ensure user privacy isn’t compromised by companies.
A startup tried to train an AI model for their own use case, but gave up after realizing how much renting machines would cut into their runway.
A bright developer built an app similar to Perplexity using the OpenAI API, only to be forced to drastically limit user access due to API rate limits and excessive costs.

From vantage points within startups and academia, Hyperbolic founders Jasper Zhang and Yuchen Jin saw talented people with promising ideas not coming to fruition. Their bottleneck to progress was not a lack of talent or enthusiasm, but a lack of compute. So they looked for it.

It wasn’t hard to find, it just wasn’t accessible. There are over two billion[^1] personal computers in the world and most of them sit idle for more than 19 hours a day[^2]. Additionally, companies often reserve data-center machines for years, then abandon projects and leave those resources untouched—an issue worsened by continual efficiency gains.

Then came DeepSeek R1. A clever team disrupted the AI space without massive budgets, proving that grit and creativity matter more than capital. If they’d used Hyperbolic’s GPUs, they could’ve cut their costs in half. Imagine what you could build with that kind of leverage.

“Understanding the scale of underutilized GPUs was a eureka moment for me. If we could aggregate global compute and deliver it in a way that’s both performant and affordable, we could significantly increase access to supply for developers and researchers worldwide.”
— Jasper Zhang, Hyperbolic CEO and Co-Founder

AI isn’t just for the giants anymore. You don’t need to be OpenAI or Anthropic to build at scale. You don’t need a massive burn rate to train a model. And you definitely don’t need to compromise just to get your project live.

We built Hyperbolic for developers who ship. Builders who don’t want to wait. Researchers who won’t be priced out.

“As a researcher, I’ve hit walls from compute limits more times than I can count. As a CTO, I’m obsessed with removing those walls for others. We built Hyperbolic so that any developer—whether you’re training a model, running inference, or scaling a product—can access the infrastructure they need, without delays, surprises, or constraints.”
— Yuchen Jin, Hyperbolic CTO and Co-Founder

No red tape. No waitlists. Just raw power when you need it, where you need it.

So the question is, what idea will you take Hyperbolic?

[^1]: “Computers Sold.” Worldometer.

[^2]: Desroches, Louis-Benoit, et al. “Computer usage and national energy consumption: Results from a field-metering study.” eScholarship.