Hyperbolic

Hyperbolic (also referred to as Hyperbolic Labs) operates an open-access AI cloud that provides access to high-performance NVIDIA GPUs, including H100, H200, and B200, sourced through a global network of compute providers. The platform is organized around three consumption models: On-Demand GPUs, which can be launched in minutes with no commitment and usage-based pricing; Reserved Clusters, offering dedicated capacity at discounted rates under commitments of less than one year; and Private Cloud, which gives customers access to Hyperbolic's supplier network for dedicated, long-term infrastructure at its lowest cost. Alongside raw compute, Hyperbolic serves models through an OpenAI-compatible API and offers dedicated model hosting.

Customers provision and manage instances through a self-serve dashboard without sales calls, procurement cycles, or quota requests, and can scale usage up or down. Stated workload targets include foundation-model training on reserved multi-node H100/H200/B200 clusters with high-performance interconnects, fine-tuning and LoRA on bare-metal single-node and small multi-node capacity, batch compute jobs such as evaluations and data processing, long-running training jobs, agent automation that programmatically launches and manages GPU resources, and private infrastructure with dedicated hardware and isolated networking.

The company also runs a supply-side program for data centers and infrastructure providers, who can list available GPU capacity via Forge and Hyperbolic's managed compute platform to serve on-demand, reserved, and private cloud deployments.

Business model

Hyperbolic operates a two-sided compute marketplace and cloud platform: it aggregates GPU capacity from data centers and infrastructure providers and resells it to AI teams as on-demand instances, reserved clusters, or private cloud deployments, alongside model-serving via an API.

Usage-based pay-as-you-go pricing for on-demand GPU instances, discounted rates for reserved capacity under commitments of less than one year, and lowest-cost pricing for long-term private cloud contracts; compute providers monetize idle GPU capacity through the network.

Traction

The website states more than 250,000 builders/engineers use Hyperbolic's AI infrastructure. Referenced users and advisors include the LMSYS Chatbot Arena (running FastChat and SGLang workloads), researchers at UC San Diego, and startups Voio, Eigen AI, and Reve AI; Reynold Xin (Databricks co-founder) and Prof. Yi Ma (HKU) are listed as advisors.

Latest developments

Product updates published on the company site include Hyperbolic Organizations (December 2025), Dedicated Model Hosting (January 2026), Reserve On Demand (February 2026), and Forge, a platform for compute providers to list GPU capacity (June 2026), plus market analyses on GPU pricing and compute planning through August 2026.