🇺🇸 USA · RunPod
Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: GPU cloud + serverless inference. Famous for cheap spot pods, community templates, easy deployment of any Docker container with GPU. Popular with indie AI developers + ML hobbyists.
Front-matter facts
| Field | Value |
|---|---|
| Vendor | RunPod Inc (USA) |
| Country / origin | 🇺🇸 USA |
| Recommended for Australian users? | ✅ Yes — accessible from AUS |
| Privacy summary | Standard cloud provider; your workloads on rented hardware |
| Free tier | None for GPUs; some free credits on sign-up |
| Paid tiers | Per-hour rentals + serverless per-second; spot pods for cheaper variable-availability |
| First released | 2022 |
| Last reviewed | 2026-06-26 |
| Official site | https://runpod.io |
What it is
RunPod is a GPU cloud focused on flexibility + cost + ease of use. Two main products:
Pods (rented GPUs)
- Rent GPUs by the hour or second
- Spot Pods — cheaper, can be reclaimed (~50% off On-Demand)
- On-Demand Pods — guaranteed availability
- Community templates (one-click deploy of popular ML tools: Stable Diffusion WebUI, Ollama, Whisper, etc.)
- Multi-GPU support
Serverless
- Pay-per-second functions on GPU
- Cold-start delays (~5-30s)
- Good for variable inference loads
- Auto-scaling
Why RunPod is popular:
- Spot pods are very cheap — among the cheapest H100 access
- Easy template deployments — Stable Diffusion / Ollama / fine-tuning workflows pre-configured
- Community / hobbyist friendly — Discord support, lots of tutorials
- Multiple regions / hardware tiers — A100, H100, H200, B200 availability
What you’d use it for
- Indie AI developer projects — cheap GPU access for prototypes
- Stable Diffusion workflows at scale without local GPU
- Fine-tuning open-weight models — cheap H100 spot pods
- Hobbyist ML experimentation
- Serverless inference for variable-load AI products
- Quick model deployment via community templates
When NOT to use:
- For production-grade SLAs (RunPod is reliable but not enterprise-class)
- For AUS data residency
- For broader cloud needs (compute-focused)
- For tightly-coupled enterprise compliance (less mature than hyperscalers)
How to use from Australia
- Sign up at runpod.io
- AUS card accepted
- Pick a pod template (Stable Diffusion / Ollama / Custom Docker)
- Spin up — typically 30 seconds for warm pods
- SSH or web UI access
- Shut down when done
What it costs
On-Demand Pods (verify current)
- H100 80GB SXM: ~US$2.99/hour
- A100 80GB: ~US$1.74/hour
- RTX 4090: ~US$0.34/hour
- L40S: ~US$1.30/hour
Spot Pods
- ~50% cheaper than On-Demand
- Can be reclaimed; design workflows accordingly
Serverless
- Per-second pricing
- Cold start latency
- Auto-scaling
How it compares to alternatives
| Aspect | RunPod | Lambda Labs | Modal | CoreWeave |
|---|---|---|---|---|
| Spot pricing | Best (cheapest) | Good | N/A (serverless) | Limited |
| Hobbyist friendly | Best (templates + community) | Researcher | Developer | Enterprise |
| Serverless option | Yes | API (Inference) | Yes (best) | No |
| Multi-region | Yes | US | US | Multi |
| AUS data residency | No | No | No | No |
| Best for | Indie / hobbyist / spot pricing | Cost-conscious researchers | Pythonic serverless | Enterprise GPU clusters |
For cheapest GPU access with community templates, RunPod is hard to beat.
Privacy / data handling
- Standard cloud — your workloads on rented hardware
- US-based; no AUS residency
- Spot pod data persists per template settings
Recent changes
- 2025-26: Blackwell B200 availability; serverless improvements
- 2024: Spot pod pricing increasingly competitive
Gotchas
- Spot pods can be reclaimed — design for that
- US-only infrastructure — high latency for AUS interactive use
- Cold starts on serverless — 5-30s typical
- For production-grade SLAs, hyperscaler GPU (AWS / Azure / GCP) better
- Discord support culture — friendly but not enterprise-style support
See also
- Lambda Labs 🟩 🟦
- CoreWeave 🟥
- Modal 🟩 🟦
- Together AI 🟩 🟦
- Fireworks AI 🟩 🟦
- Replicate 🟩 🟦
- Nvidia AI 🟩 🟦
- Ollama 🟥