Lightning AI
AI cloud from the PyTorch Lightning team: on-demand GPUs, clusters, Studios dev environments and LitServe serving; merged with Voltage Park in 2026
Overview
Lightning AI is an AI cloud for building, training and deploying models, made by the team behind the open-source PyTorch Lightning library. It offers on-demand GPUs (B200, H200, H100, A100, L4 and others), multi-node clusters, bring-your-own-cloud, and Studios: browser-based dev environments with SSH access where you prototype, train and serve models, plus open-source tools such as LitServe for model serving, Fabric, LitGPT and Thunder. Enterprise plans add SSO, role-based access, audit logs and VPC deployment. In January 2026 Lightning AI completed a merger with GPU cloud Voltage Park; the combined company runs under the Lightning AI name and says it owns more than 35,000 GPUs. Logos on its site include NVIDIA, Cursor, Runway and Stability AI.
The Verdict
Who Should Use Lightning AI?
Best For
- PyTorch Lightning users who want the same team's managed cloud
- Researchers who want a GPU dev environment without managing machines
- Teams moving from prototype to multi-node training and serving in one place
- Students and hobbyists starting on the free credits
Not Ideal For
- Buyers hunting the cheapest raw H100 hour
- Teams that need clear, consistent published plan terms before signing up
- Anyone relying on a large base of independent reviews
What's Great
- Free tier with starter credits, two seats and a free CPU Studio
- Open-source core libraries (PyTorch Lightning, LitServe) under Apache-2.0
- Studios cover the whole loop: data, prototype, train, evaluate, serve
- Enterprise controls: SSO, RBAC, audit logs, VPC
- Users credit it with saving time on GPU setup and standardizing training runs
Watch Out For
- Pricing page contradicts itself (Pro shown at both $50 and $20 a month, two H100 prices, two CPU counts for Teams)
- H100 at $5.45 an hour on demand is above many GPU marketplaces
- Free Studios restart every four hours
- G2 has too few reviews for buying insight
- Merger with Voltage Park means changes in scale and ownership
Pricing
GPU time is billed on top: the main table lists H200 at $4.50, H100 at $5.45 and A100 (40 GB) at $2.19 per hour on demand.
View all features & details
Key Features
- On-demand GPUs and InfiniBand multi-node clusters
- Studios: persistent GPU dev environments with SSH
- Bring your own cloud account
- LitServe model serving and inference endpoints
- SSO, RBAC, audit logs, VPC deployment
Open-Source Projects
- PyTorch Lightning and Fabric
- LitServe
- LitGPT
- Thunder and TorchMetrics
How It Compares
| Feature | Lightning AI | RunPod | Modal | CoreWeave |
|---|---|---|---|---|
| Main model | GPU cloud plus dev Studios | GPU pods and serverless | Serverless Python compute | Enterprise GPU cloud |
| Free tier | Yes, starter credits | See vendor | See vendor | See vendor |
| Own training library | PyTorch Lightning, LitServe | No | No | No |
Sources: Lightning AI homepage · Lightning AI pricing · SiliconANGLE (Nov 2024)
Is this your tool? Add a badge