Dr. Jasper Zhang

Dr. Jasper Zhang is the co-founder and chief executive of Hyperbolic, an AI infrastructure company building an open-access AI cloud from computing capacity distributed across independent providers. He argues that AI’s compute shortage is partly a coordination failure: valuable GPUs sit idle while researchers and startups struggle to obtain affordable, dependable infrastructure.

Zhang, also known professionally as Yue Zhang, studied pure mathematics at Peking University and completed his mathematics doctorate at the University of California, Berkeley, between 2018 and 2020. His doctoral research on hyperbolic manifolds, supervised by Ian Agol, inspired Hyperbolic’s name. He also won gold medals in the Chinese Mathematical Olympiad and Alibaba Global Mathematics Competition.

After Berkeley, Zhang worked as a quantitative researcher at Citadel Securities, applying mathematics and machine learning to financial markets, then became a senior blockchain researcher at Ava Labs. He met Hyperbolic co-founder Yuchen Jin in late 2022. Their company initially offered spot GPU capacity and serverless inference before expanding into on-demand infrastructure, bare-metal and virtual-machine clusters, and longer-term reservations. In December 2024, Zhang announced a $12 million Series A, bringing Hyperbolic’s disclosed funding to $20 million.

The infrastructure ideas driving Hyperbolic

• Compute scarcity is a coordination problem. Zhang advocates combining new data-center construction with better use of existing hardware. His case for aggregating fragmented GPU supply addresses idle machines, power constraints, environmental costs, and the difficulty of comparing unfamiliar providers.

• Flexible GPU markets follow real workloads. His GPU marketplace supports spot, on-demand, and reserved capacity as teams move between training and inference. Customers can temporarily add machines and relist unneeded capacity, making expensive infrastructure more adaptable to changing demand.

• Distributed orchestration makes independent clusters usable. The Hyper-dOS distributed operating system connects Kubernetes-ready clusters into a broader network; MicroK8s can make additional machines compatible. Hyperbolic’s Monarch coordination service routes customer requests to providers, which provision machines and establish SSH access.

• Reliable compute can become a liquid market. Zhang pairs affordability with hardware benchmarking, throughput validation, and network checks. More recently, he has outlined unified provisioning and Forge, a multi-tenant system for incorporating bare-metal clusters into marketplace inventory, alongside capacity resale, dynamic pricing, and potential forward markets for compute.