GPU compute,
on demand.
Rent GPU nodes from providers worldwide. Deploy inference endpoints, SSH into live instances, or run batch jobs — billed by the second, with enterprise-grade isolation.
Nodes online
Jobs completed
Running now
99.9%
Uptime
Your GPU control plane
Browse live capacity, deploy in a click, and manage every endpoint, job, and invoice from one console.
HZ AIMarketplace
128 GPUs online across 14 regions
NVIDIA H100 80GBSXM5
16 vCPU · 192GB · Frankfurt
NVIDIA A100 80GBPCIe
12 vCPU · 128GB · Oregon
RTX 4090 24GBAda
8 vCPU · 64GB · Singapore
From zero to GPU in three steps
Live marketplace
H100 80GB
Frankfurt
A100 80GB
Oregon
RTX 4090
Singapore
Built for serious workloads
From fine-tuning LLMs to rendering 3D scenes — compute-intensive work at a fraction of cloud prices.
LLM Inference
Deploy vLLM, Ollama, or TGI endpoints. Serve Llama 3, Mixtral, and Phi on real GPUs.
Model Fine-Tuning
LoRA or full training runs. Persistent volumes keep your checkpoints safe between sessions.
Image Generation
Stable Diffusion, ComfyUI, and more. Serve as an endpoint or batch thousands of images.
Video & 3D Rendering
Blender, Unreal, or custom pipelines distributed across multiple GPU nodes.
Data Processing
RAPIDS, GPU Spark, or custom ETL. Process terabytes with GPU acceleration.
Research
Jupyter, PyTorch, experiment tracking. SSH in and work like it's a local machine.
Engineered for multi-tenant trust
Every layer is designed for isolation, reliability, and zero-trust by default.
gVisor Sandbox
Optional userspace syscall interception — the same isolation model as Google Cloud Run.
Container Isolation
Per-job containers, dropped capabilities, read-only rootfs, enforced PID limits.
Constant-Time Auth
SHA-256 hashed API keys with constant-time validation and instant revocation.
Automatic TLS
Caddy provisions HTTPS automatically. HSTS enforced, mTLS between control planes.
Per-Second Billing
Real provider pricing, 5% platform fee, transparent invoicing with CSV export.
Memory-Safe Agent
Rust agent, zero GC pauses, Docker via library — no shell injection surface.
Rate Limiting
Per-buyer limits and idempotency keys, Redis-backed for sub-millisecond checks.
Full Observability
Prometheus metrics, Grafana dashboards, and alerting rules out of the box.
Have GPUs? Put them to work.
Install the agent, set your price, and earn whenever a buyer runs a job on your hardware. You keep 95% of every transaction.
- One-command install on Ubuntu / Debian
- Set your own price per GPU-hour
- Auto-detected specs shown to buyers
- Fiat payouts, transparent earnings
$ curl -fsSL https://hz-ai.io/install.sh | bash ✓ Installing HZ AI Agent v0.1.0… ✓ Detected 2× NVIDIA A100 80GB ✓ Agent registered and running $ hzai-agent --price 1.89 Listing updated · $1.89 / gpu-hr Waiting for jobs… ↳ Job received · buyer@corp.io Earned $3.78 (2 GPU-hours)▋
Questions, answered
You prepay with credits. When a job runs, compute is billed per second at the provider's GPU-hour rate. Unused credits stay in your account.
Yes. On-demand instances and dedicated reservations give you full SSH access. Your public key is configured in Settings.
Whatever providers list — today that spans consumer cards (RTX 4090, 3090) and data-center GPUs (A100, H100, A10G). The marketplace shows real-time availability.
Each job runs in an isolated container with a read-only root filesystem. Volumes are scoped per buyer and pinned per node, so no other buyer can reach your data.
Register a provider account, run the one-command installer on your Linux machine, set your price, and you're live. The agent auto-detects your GPU specs.
Any public image from Docker Hub, NVIDIA NGC, or GitHub Container Registry. Private-registry support is on the roadmap.