Hyperbolic is a GPU and inference cloud that sources capacity from a global network of compute providers rather than a single data-centre estate. You can rent H100, H200, and B200 instances on demand, reserve dedicated multi-node clusters, or arrange longer-term private-cloud capacity through the same supplier network; the company states more than 250,000 builders use it.
Alongside raw compute, Hyperbolic serves inference over an OpenAI-compatible API across a set of open models, so an application already written against an OpenAI client can point at it without a rewrite. The stated use cases are training, fine-tuning, experimentation, and short- or long-term production inference.
Capacity comes in three commitment tiers. On-demand instances launch in minutes with no commitment and bill by usage, provisioned self-serve from a dashboard without a sales call or quota request. Reserved clusters trade a commitment of under a year for discounted dedicated capacity aimed at steady production workloads. Private Cloud covers dedicated long-term infrastructure for teams that need predictable capacity over the longest commitment.
The platform also runs a supply side: data centres and infrastructure providers can connect spare GPU capacity to the network and sell it through Hyperbolic.
Features
- On-demand GPUs: H100, H200, B200, and other accelerators launched in minutes, billed by usage, with no commitment
- Reserved clusters: dedicated multi-node capacity at discounted rates under sub-one-year commitments
- Private cloud: long-term dedicated infrastructure drawn from the supplier network
- Serverless inference: an OpenAI-compatible API over a catalogue of open models
- Self-serve provisioning: instances launched and managed from a dashboard, scaled up or down without procurement
- Provider network: data centres and infrastructure operators can list spare capacity for sale on the network
