Hyperstack: Secure Private Cloud and On-Demand GPU Cloud
Official website: Fast, On-Demand Cloud GPU Solutions for AI & ML | Hyperstack
Hyperstack is an NVIDIA Cloud Partner and full-stack AI cloud serving enterprises, AI teams and engineering teams through Secure Private Cloud for dedicated, single-tenant GPU infrastructure and On-Demand GPU VMs for self-service compute. More than 28,000 users run AI workloads on the platform. Both SPC and On-Demand GPU VMs are Hyperstack products (operated by parent company NexGen Cloud).
Secure Private Cloud
Dedicated, isolated single-tenant infrastructure engineered for enterprise AI models and massive deep learning pipelines:
- Dedicated single-tenant infrastructure: Every deployment runs on physical GPU infrastructure reserved for one organisation, with zero oversubscription and workloads that are physically and logically isolated from other customers.
- Private GPU Clusters: Deployments are available on NVIDIA Blackwell / NVIDIA Blackwell Ultra clusters, including NVIDIA B300 and NVIDIA GB300 NVL72 architectures.
- Four deployment models: Organisations choose between Metal Only, Managed Metal, Managed Orchestration, and Dedicated Cloud depending on how much of the stack they want to operate themselves.
- Bare-metal performance: Virtualised environments deliver 100% bare-metal performance, with SIOV NIC passthrough for direct network access.
- Networking and storage: Clusters connect over NVIDIA InfiniBand or NVIDIA Spectrum-X with NVIDIA ConnectX-8 SuperNICs. Storage options include Ceph as standard, with WEKA, VAST and DDN available as NVIDIA-certified RDMA options for training.
- Orchestration: Teams can run Slurm, Kubernetes, AI Studio or Hyperstack's Managed Orchestration on their dedicated cluster.
- Compliance and resilience: Hyperstack holds SOC 2 Type 2 attestation, offers hot, warm and cold standby hardware tiers and provides 24/7 support from a UK-based team.
- Commercial terms: Deployments start at 512 GPUs across 64 systems on terms of 12 months or longer.
On-Demand GPU VMs
Self-service GPU computing with instant API and console provisioning for developers and startups:
- Self-service GPU VMs: Developers can provision GPU VMs on NVIDIA B300, NVIDIA B200, NVIDIA H200 SXM, NVIDIA H100, NVIDIA A100 and NVIDIA RTX PRO 6000 SE. Pricing at the time of writing starts at $7.40/hr for NVIDIA B300, $6.00/hr for NVIDIA B200, $3.99/hr for NVIDIA H200 SXM and $1.35/hr for NVIDIA A100 on demand, with reserved NVIDIA A100 from $0.95/hr.
- Transparent billing: Usage is billed per minute on a pay-as-you-go basis, with no ingress or egress fees ($0 data transfer costs).
- High-speed networking: NVIDIA H100 and NVIDIA A100 VMs support networking speeds of up to 350Gbps.
- Kubernetes and storage: On-demand Kubernetes is available through both the console and the API, alongside S3-compatible object storage.
- AI Studio and inference: AI Studio lets you deploy open-weight models from its catalogue or Hugging Face, including private and gated models, on dedicated GPUs behind a private OpenAI-compatible endpoint, with catalogue models also available per token through the Hyperstack gateway.
- Dynamic availability: GPU capacity varies by region and can be checked in real time through Hyperstack's console and API.
Four Deployment Models Matrix
| Model | Customer Management | Hyperstack Management | Recommended For |
|---|---|---|---|
| Metal Only | OS, drivers, network topologies, storage systems, and ML software | Datacenter, power, cooling, physical cabling, and bare servers | Full bare-metal customization with in-house sysadmin teams |
| Managed Metal | Orchestration runtime, container stacks, and model training jobs | Hardware maintenance, base operating system, driver updates, and firmware | Direct bare-metal compute without hardware ops overhead |
| Managed Orchestration | Training workloads, scripts, and inference endpoints | Full cluster runtime: Slurm, Kubernetes, AI Studio, and storage mounts | AI teams focused purely on model engineering and experimentation |
| Dedicated Cloud | Self-service developer workflows via web console and REST API | Dedicated physical cluster virtualization, SIOV NIC passthrough, VPCs, and storage | Large enterprises wanting on-demand cloud flexibility on single-tenant hardware |
Editorial Verdict: Hyperstack Review
Hyperstack has solidified its position as one of the most reliable and capable AI clouds globally. Serving over 28,000 users as an official NVIDIA Cloud Partner, Hyperstack successfully bridges the gap between scalable dedicated enterprise infrastructure (Secure Private Cloud) and developer-first on-demand computing (On-Demand GPU VMs & AI Studio). With transparent per-minute billing, guaranteed $0 egress fees, and 24/7 technical support from a UK-based team, Hyperstack is highly recommended for developers, startups, and large-scale enterprise AI teams.
