Hyperstack

Hyperstack is NexGen Cloud's GPU-as-a-Service platform for AI, machine learning and HPC workloads. It provides NVIDIA GPU virtual machines through a console and an API-first model, with data centres in Europe and North America. The platform also lists Kubernetes, object storage, and an AI Studio for training, deploying and monitoring custom models.

What It Is

Hyperstack is a GPU cloud where users deploy virtual machines on NVIDIA hardware such as H100, H200, A100, L40 and RTX A6000, with newer Blackwell-generation options listed on the site. Workloads include model training, fine-tuning, inference, rendering, virtual workstations, data analytics and simulation. Billing is stated to be accurate to the minute.

Platform and Tooling

The site lists role-based access control, resource callbacks, grouping of VMs into environments, custom-build flavours and one-click deployment. Hyperstack says it has no egress or ingress fees. Developer tooling includes documentation, a Terraform provider, Go and Python SDKs, and an open-source LLM Inference Toolkit.

LLM Inference Toolkit

The MIT-licensed toolkit deploys open-source LLMs locally or on Hyperstack. It provides a UI for managing models, a playground, API key generation with usage limits, a vLLM-backed proxy API, and monitoring of usage data. The default cloud deployment creates a proxy VM plus two inference VMs hosting Llama 3.1 8B Instruct and Phi-3 medium.

Current Status

The homepage announces NVIDIA B300s coming on-demand in August, with reserved private clusters in Q4. AI Studio is presented as an early-access offering. The site also warns of a fraudulent site impersonating Hyperstack.