Lium
GPU cloud for developers and AI agents, billed per second
At a glance
prices checked 4h ago- Cheapest GPU
- $0.12/hr
- NVIDIA RTX A4000 16GB · community
- Fastest GPU
- $4.00/hr
- NVIDIA B300 SXM 262GB · spot
- GPU models
- 16
- 26 configurations
- Largest node
- 8× GPUs
What's good about Lium
- Per-second billing No minimum, no contract
- Built for AI agents CLI, Python SDK and REST API
- Secure and Spot tiers Vetted data centers or a lower spot price
- Bare machines with root SSH and Docker in minutes
- Public live feeds Every free machine, every price
- Global marketplace 19 countries
Available GPUs
Cheapest first, per GPU per hour.
| GPU | Configuration | Memory | Price | Region | |
|---|---|---|---|---|---|
| NVIDIA B300 SXM 262GB Blackwell Ultra | 8 GPUs Spot | 262.5 GB | $4.00/hr | United States | Rent |
| NVIDIA B300 SXM 262GB Blackwell Ultra | 8 GPUs Spot | 262.5 GB | $4.00/hr | India | Rent |
| NVIDIA B200 SXM 180GB Blackwell | 1 GPU Community | 180 GB | $6.10/hr | United States | Rent |
| NVIDIA B300 SXM 262GB Blackwell Ultra | 1 GPU Community | 262.5 GB | $8.00/hr | United States | Rent |
| NVIDIA B300 SXM 262GB Blackwell Ultra | 8 GPUs Community | 262.5 GB | $12.90/hr | United States | Rent |
| NVIDIA H100 SXM5 80GB Hopper | 8 GPUs Spot | 80 GB | $1.60/hr | Finland | Rent |
| NVIDIA H100 SXM5 80GB Hopper | 1 GPU Community | 80 GB | $1.79/hr | Russia | Rent |
| NVIDIA H200 SXM 141GB Hopper | 1 GPU Spot | 141 GB | $3.35/hr | United States | Rent |
| NVIDIA H200 SXM 141GB Hopper | 1 GPU Community | 141 GB | $3.60/hr | Singapore | Rent |
| NVIDIA H200 NVL 141GB Hopper | 1 GPU Community | 141 GB | $3.50/hr | United States | Rent |
Lium head to head
Compare any two
About Lium
Lium is a GPU cloud where you rent bare GPU machines on demand, from a single RTX 3060 to an 8-GPU B300 node, with SSH and Docker access in minutes. The machines come from a global network of independent data centers and providers rather than one operator's fleet, so prices are often well below the big clouds. Lium checks every machine's GPUs and scores their performance, and every pod gives root and Docker-in-Docker access.
Capacity is on-demand only, in two tiers. Secure nodes come from vetted data centers whose operators are penalised if a rental is interrupted. Spot nodes cost less and may be interrupted without compensation, which suits checkpointed training, batch inference and other fault-tolerant work. There is no reserved capacity. Billing is per second at the listed hourly rate with no minimum and no contract, charged from a prepaid balance that is topped up by card, US bank debit or crypto.
Everything can be driven from the web console or the API, which makes Lium a fit for AI agents that rent and release GPUs on their own. The headless CLI and Python SDK return JSON with stable exit codes, rent-by-spec takes the cheapest node under a price ceiling, API keys carry scopes and daily, monthly and lifetime USD budgets, and a pod can be given a time to live so it terminates itself. Public feeds list every free machine and the live price range per GPU model without a key or signup.
Lium is operated by Datura AI Corp, headquartered in Charlestown, Nevis. Machines are available across 19 countries in North America, Europe, Asia and Australia. Pod volumes are encrypted at rest by default, confidential computing nodes keep the host operator out of a running pod, and a provider who accesses a renter's workload is permanently banned from the network.
Pricing options
Services
- No minimum, no contract Per-second billing
- Pods bill per second at the listed hourly rate from the moment they are running, and a pod that never starts is not charged. The balance is topped up by card, US bank debit or crypto.
- CLI, Python SDK and REST API Built for AI agents
- A headless CLI and SDK with JSON output and stable exit codes, rent-by-spec under a price ceiling, API keys with scopes and daily, monthly and lifetime USD budgets, and auto-termination, so an agent can rent and release GPUs on its own.
- Vetted data centers or a lower spot price Secure and Spot tiers
- Secure nodes come from vetted data centers whose operators are penalised for interrupting a rental. Spot nodes cost less and can be interrupted, which suits checkpointed or fault-tolerant work.
- SSH and Docker in minutes Bare machines with root
- Every pod gives root and Docker-in-Docker on a bare GPU machine, from a single RTX 3060 to an 8-GPU B300 node, reachable over SSH shortly after the rent request.
- Every free machine, every price Public live feeds
- A public nodes feed lists every rentable machine with its price, tier, location and specs, and a per-model price feed carries the live range for each GPU. No key or signup needed; this listing is read from them.
- 19 countries Global marketplace
- Machines come from independent data centers and providers across North America, Europe, Asia and Australia, so prices are often well below the big clouds. Lium checks every machine's GPUs and scores their performance.
Regional coverage
Where workloads run: primary datacenters and active deployment zones.
- United States United States Primary
- Canada Canada
- United Kingdom United Kingdom
- Ireland Ireland
- Netherlands Netherlands
- France France
- Finland Finland
- Portugal Portugal
- Latvia Latvia
- Romania Romania
- Bulgaria Bulgaria
- Ukraine Ukraine
- Russia Russia
- Kazakhstan Kazakhstan
- Singapore Singapore
- Hong Kong Hong Kong
- South Korea South Korea
- Indonesia Indonesia
- Australia Australia
Frequently asked questions
Common questions about working with Lium.
Which GPUs does Lium offer?
NVIDIA B300, B200, H200, H200 NVL, H100 SXM, H100 PCIe, A100 80GB (SXM and PCIe), RTX PRO 6000 Blackwell, RTX 6000 Ada, RTX A6000, L40S, L40, RTX 5090, RTX 4090, RTX 3090, RTX A5000, RTX A4000, RTX 5070 Ti, RTX 5060 Ti, RTX 3080 and RTX 3060, as single GPUs up to whole 8-GPU nodes. The machines are listed by independent providers, so what is rentable at any moment is on the live pricing page and in the public nodes feed.
What is the difference between Secure and Spot on Lium?
Both are on-demand with no contract. Secure nodes come from vetted data centers, and the operator is penalised if a rental is interrupted. Spot nodes are cheaper and carry a Spot badge because the pod may be interrupted without compensation, so they suit checkpointed training, batch inference and other work that can restart. There is no reserved capacity.
How does Lium billing work?
Per second, with no minimum and no contract. A pod is charged at its hourly price times seconds running, from the deploy request once it reaches running, and a pod that never starts is not charged. Charges come out of a prepaid USD balance every five minutes; the balance is topped up by card, US bank debit, crypto or TAO. Volumes and backups are metered separately per GB.
Where are Lium machines located?
Across 19 countries: the United States, Canada, the United Kingdom, Ireland, the Netherlands, France, Finland, Portugal, Latvia, Romania, Bulgaria, Ukraine, Russia, Kazakhstan, Singapore, Hong Kong, South Korea, Indonesia and Australia. Each node in the public feed carries its country and city, and the machine list changes as providers join and leave.
Can an AI agent rent a GPU on Lium?
Yes. The CLI and Python SDK are headless, return JSON and use stable exit codes, so an agent with a shell can rent, use and release a GPU on its own. Rent-by-spec takes the cheapest node matching a GPU spec under a price ceiling, API keys carry scopes and daily, monthly and lifetime USD budgets enforced by the server, and a pod can be started with a time to live so it terminates itself. An MCP server exposes live prices and the docs.
Is my data private on a Lium machine?
On a standard node the provider operates the host and could read a running pod, so Lium tells renters to keep private keys and API tokens off it and to fetch secrets at runtime. Confidential computing (CVM) nodes keep the host operator out of the pod. Pod volumes are encrypted at rest by default, and a provider who accesses, copies or interferes with a renter's workload is permanently banned from the network.
Ready to deploy on Lium?
Visit Lium