Hyperstack by NexGen Cloud
Enterprise-grade GPU cloud built for AI and HPC workloads
GPU cloud platform with datacenters in Norway, Sweden, Canada, and the US. Competitive per-GPU on-demand pricing with zero egress fees, spot at 20% off, no commitments, and a full AI development suite — fine-tuning, model evaluation, Hugging Face import, and a unified inference playground. Pricing is identical across all regions.
Among the lowest per-GPU rates for H200 and H100 SXM tracked. Zero egress/ingress fees — no hidden bandwidth costs.
Engineer-led support with defined escalation paths directly to infrastructure specialists. Newer provider without a hyperscaler-length track record.
Full AI Studio (fine-tuning, Hugging Face import, evaluation, playground), GPU-backed Kubernetes with free master nodes, object storage, VM snapshotting, and hibernation.
| GPU Model | VRAM | On-Demand | Spot (−20%) | 1Y Reserved |
|---|---|---|---|---|
| B200 192GB | 192 GB | $6.00 | $4.80 | $5.10 |
| H200 141GB | 141 GB | $3.99 | $2.80 | $2.79 |
| H100 SXM 80GB | 80 GB | $3.20 | $1.92 | $2.72 |
| H100 NVLink 80GB | 80 GB | $2.60 | $1.56 | $1.82 |
| H100 PCIe 80GB | 80 GB | $2.50 | $1.52 | $1.75 |
| RTX Pro 6000 SE 96GB | 96 GB | $1.85 | $1.44 | $1.30 |
| A100 SXM 80GB | 80 GB | $1.60 | $1.28 | $1.36 |
| A100 PCIe 80GB | 80 GB | $1.40 | $1.08 | $0.95 |
| L40 48GB | 48 GB | $1.00 | $0.80 | $0.70 |
| A6000 48GB | 48 GB | $0.50 | $0.40 | $0.35 |
| A4000 16GB | 16 GB | $0.15 | $0.12 | $0.11 |
All prices per GPU/hr. Spot is 20% below on-demand. regularly tracked · last tracked Jul 1, 2026 · confirm current rates before procurement.
A fully integrated AI development environment built into the Hyperstack console. Training data, fine-tuning jobs, evaluation, and production inference all in one place.
Wide range of open-source models (Llama, Mistral, Qwen, etc.) available via a single unified API.
Import any model from Hugging Face directly, including LoRA adapters. Validate and import in one step.
Upload training data (JSONL), create a fine-tuning job, and view cost/time estimates before committing.
After training, view the loss curve and full performance indicators inside the console.
Evaluate your fine-tuned model against industry-standard benchmark datasets — no need to leave the platform.
Define your own evaluation criteria and test against your own uploaded dataset.
Assign human-readable alias names to fine-tuned models organised by purpose or task.
Built-in leaderboard to compare your fine-tuned model against competitive models.
Run inference on any model. Compare two models side-by-side to quantify fine-tuning gains.
Click 'API' on any model to get ready-to-paste code. Integrate directly into your application.
GPU VM live in under a minute. Choose GPU type, count, region, OS, SSH key, firewall rules — all in one flow.
Available for every GPU at 20% below on-demand. Best for testing, development, and short-term workloads.
Pause a VM to minimize costs and resume from the exact same state later — useful for batch jobs with idle periods.
Save the full machine state. Restore in place or use as a base image to spin up identical VMs.
Move data in and out freely. No bandwidth charges — a meaningful cost advantage for data-heavy AI and HPC workloads.
Pay by the hour. Stop when done. No contract, no minimum term for standard on-demand VMs.
H200 SXM, H100 SXM, and A100 SXM are network-optimised for multi-GPU distributed training workloads.
Ubuntu, AlmaLinux, and Debian images — with version selection at deploy time.
Import existing SSH keys or generate new ones. Assign per-VM with optional public IP.
Configure access rules at deploy time or update them while the VM is running.
All resources — VMs, volumes, costs — in one view. Wallet balance always visible.
- ›GPU-backed Kubernetes clusters using the same GPU lineup as VMs
- ›Free master nodes — only pay for worker nodes
- ›Choose Kubernetes version at deploy time
- ›Configuration mirrors VM setup: GPU type, region, OS, SSH keys
- ›Scales for deploying and managing applications across multiple machines
- ›Dedicated GPU clusters for large-scale AI training and ML workloads
- ›Submit a request form specifying GPU series, duration, and requirements
- ›Hyperstack team contacts you directly with full assistance
- ›Separate from Kubernetes-based clusters — for bare-metal-grade dedicated resources
- ›Object Storage: S3-compatible buckets, managed via Hyperstack API or console
- ›Access key per region — credentials usable from any S3-compatible client
- ›Volume Storage: attach persistent block volumes to VMs
- ›Set volume size (e.g., 100 GB) for hosting large AI models and datasets
Hyperstack is independently listed on GPUAdvisor. Listing decisions are based on publicly verifiable signals only. Deploy links may be affiliate links — rankings and scores are never influenced by commercial relationships.