Artificial intelligence infrastructure startup Hyperbolic Labs Inc. said today it has closed on a $7 million seed funding round to transform the way AI developers access compute and inference resources.

Today’s round was led by Polychain Capital and Lightspeed Faction, and saw the participation of a host of other investors, including Chapter One, LongHash, Bankless Ventures, Republic Digital, Nomad Capital, CoinSummer Labs and Third Earth Capital, as well as angel investors such as Balaji Srinivasan, Illia Polosukhin, Sandeep Nailwal, Casey Caruso, Tekin Salimi and Santiago Santos.

Hyperbolic has attracted such widespread attention from the investment community because of its promise of delivering “open access” to the essential AI computing resources that so many companies are scrambling to obtain today.

In its pitch, Hyperbolic explains that some AI companies spend as much as 80% of their capital on computing resources such as graphics processing units. They typically rent GPUs from public cloud infrastructure providers such as Amazon Web Services Inc., Google Cloud and Microsoft Azure, or more specialized GPU cloud companies.

But the problem with obtaining GPUs from these sources is the huge costs involved. Because of a global shortage of GPU capacity and challenges around distribution, GPU access has become incredibly expensive and unsustainable for many companies.

Hyperbolic is looking to provide an alternative, with a globally distributed network of GPU resources. Its network taps the aggregated compute power of millions of idle GPUs from around the world.

For instance, the startup points out that there are more than two billion personal computers in the world, and that the vast majority of these sit idle for more than 19 hours a day. Many of those PCs have GPUs inside. In addition, Hyperbolic draws on the underutilized GPU resources of data centers spread across the world.

The startup’s AI compute network leverages blockchain technology to aggregate these idle GPU resources and ensure that its network of nodes operates in a way that’s verifiable and secure. It says its AI inference services can be accessed at a small fraction of the cost of traditional GPU resources. To encourage users to donate their GPU resources, it’s creating a GPU marketplace where anyone can rent and supply their unused computing hardware.

Hyperbolic refers to its approach as “heterogeneous compute,” and it says it can maximize the performance of any kind of GPU. The system is powered by a decentralized orchestration layer called Hyper-dOS, short for Hyperbolic Distributed Operating System. To ensure AI workloads can run efficiently on its GPU resources, Hyperbolic uses a special compiling stack that helps to optimize AI models to run on an assortment of GPUs, including those sold by Nvidia Corp. and Advanced Micro Devices Inc.

The company also provides access to open-source large language models such as Meta Platforms Inc.’s Llama 3.1 405B, as well as image generation models such as Stable Diffusion, vision-based models such as Cambrian and audio generation models that include MeloTTS.

Hyperbolic co-founder and Chief Executive Jasper Zhang said his company was formed by a team of researchers and developers, united in a mission to make AI services more accessible, verifiable and trusted. “The AI industry remains in its early stages and we’re confident that Hyperbolic will play a crucial role as a reliable infrastructure partner that drives forward the next iteration of the industry’s lifecycle,” he said.

The funds from today’s round will be used to accelerate the development of Hyperbolic’s open-access AI cloud and strengthen its engineering and development teams. The startup also intends to build a more supportive ecosystem for developers, it said, without elaborating.

One of its immediate next steps for Hyperbolic will be to launch its proof-of-sampling consensus protocol on its own, interoperable blockchain. This will act as the foundation of its decentralized network of nodes, paving the way for everyone to contribute their unused GPU resources and monetize them.

Polychain Capital founder and CEO Olaf Carlson-Wee said he’s backing Hyperbolic because it offers an alternative to the “abuse of power” by a select few cloud providers and proprietary model creators. He pointed out that it’s advocating for a more transparent and open approach to AI development.

“Without this transparency, auditing profit-focused tech giants becomes exceedingly difficult,” Carlson-Wee said. “Hyperbolic envisions a future where distributed compute makes services like inference cheaper, enabling researchers and developers globally to experiment at larger scales.”

Image: SiliconANGLE/Microsoft Designer