Hyperbolic

Hyperbolic is a GPU cloud that rents capacity from a network of about 30 smaller GPU providers, often called neoclouds, through one console and API. It sells hourly on-demand instances of NVIDIA H100 SXM, H200 and B200 GPUs, with B300 listed in the docs. You reach these instances over SSH and can start one in minutes. Reserved capacity is prepaid at a discount on terms under a year. Private Cloud is single-tenant, for commitments longer than a year. The company also runs an OpenAI-compatible inference API for open-source text and vision models, billed per token from a credit balance. Managed Kubernetes (in beta) runs the control plane and installs the NVIDIA GPU and network operators. An open-source Rental Operator lets teams create rentals from Kubernetes resources. The service suits AI startups and developers who need GPUs for training, fine-tuning or inference without a long contract. Hyperbolic says more than 250,000 developers use it, including teams at Modal and Daytona. Prices start around 3.19 dollars per GPU-hour for an H100 SXM, 3.99 for an H200 and 5.99 for a B200. They are refreshed weekly based on supplier rates. Payment is by card through Stripe, and you are not charged for failed instances. The main drawback is that prices change often. Independent trackers saw the H100 rate more than double, from 1.50 to 3.19 dollars, within about ten weeks in 2026. That rate ranks in the more expensive half of the providers they track. There is no general-purpose web hosting. The docs do not mention object storage, and they do not list regions.