JarvisLabs is a paid tool. Please visit their website for pricing details.

JarvisLabs

Pay-as-you-go NVIDIA GPU and CPU cloud with managed templates, root-access VMs, and per-minute billing.

paid

Last Updated 8/28/2026

About JarvisLabs

JarvisLabs rents GPU and CPU compute for AI training, inference, notebooks, and other accelerated workloads. Its GPU catalog covers NVIDIA H200, H100, RTX Pro 6000 Blackwell, A100, A30, and L4 hardware, alongside general-purpose CPU virtual machines. You can choose a managed template with JupyterLab, VS Code, an API endpoint, or SSH, or take a VM with root access for custom kernels, drivers, containers, orchestration, daemons, and CI runners. GPU configurations scale from one to eight cards. On-demand instances are billed by the minute; listed GPU rates run from $0.41 to $3.99 per GPU hour. Storage costs $0.10 per GB each month, and in-region egress is free. Pausing an instance stops compute billing, though storage charges continue. Reserved capacity, clusters, and deployments above 25 GPUs require a sales quote. JarvisLabs suits workloads that benefit from quick provisioning and short billing increments. The tradeoffs are ongoing paused-storage costs, region constraints for some products, and no free trial.

Best for

AI researchers who need short-lived access to high-memory NVIDIA GPUs without a long-term compute commitment.

Engineering teams that want either managed notebook templates or root-access virtual machines for custom infrastructure.

Startups planning multi-GPU training or inference that can use per-minute billing before negotiating reserved capacity.

Best use cases

Launch a managed JupyterLab or VS Code template for an experiment, keep its storage between sessions, and pause compute when work stops.

Provision a root-access VM to install custom drivers, run Docker or Kubernetes, and connect a self-hosted CI workload.

Scale a training or inference job from one GPU to an eight-GPU configuration, then contact sales when capacity exceeds 25 GPUs.

Compare before choosing

Managed templates reduce operating-system and CUDA setup work, while VMs trade that convenience for root access and custom infrastructure.

Per-minute on-demand billing fits bursty workloads better than a fixed subscription, although paused storage continues to accrue charges.

The catalog spans budget A30 and L4 instances through H100 and H200 hardware, so cost and memory can be matched to each workload.

Features

LLM training and fine-tuning

Run single-GPU or multi-GPU training and fine-tuning jobs on H200, H100, A100, A30, or L4 hardware with configurations of up to eight GPUs.

Notebook-based experiments

Launch managed containers with JupyterLab or VS Code, persistent storage, and a preconfigured CUDA stack for iterative deep-learning work.

Custom GPU infrastructure

Use an on-demand VM with root SSH when a workload needs custom drivers, kernels, Docker or Kubernetes, daemons, or a self-hosted CI runner.

Inference and API endpoints

Deploy low-cost inference workloads on A30 or L4 instances, or expose a managed template as an API endpoint without managing the base operating system.

Pros

Per-minute on-demand billing limits compute charges for experiments that run for only part of an hour.
Managed templates handle the operating system, drivers, CUDA, and container while preserving storage across stop and start.
Root-access VMs support custom kernels, drivers, Docker, Kubernetes, daemons, and self-hosted CI runners.

Cons

Paused instances still incur storage charges at $0.10 per GB per month, so idle projects are not cost-free.
CPU virtual machines are listed only for the India (Noida) region, which can constrain latency or data-location needs.
There is no free trial; users must fund the account before testing a real workload on the service.

Pricing Plans

H200 SXM

$3.99per GPU hour

141 GB VRAM
368 GB RAM
28 vCPU
1, 2, 4, or 8 GPUs

H100 SXM

$2.69per GPU hour

80 GB VRAM
200 GB RAM
16 vCPU
1, 2, 4, or 8 GPUs

RTX Pro 6000 Blackwell

$1.89per GPU hour

96 GB VRAM
160 GB RAM
28 vCPU
1, 2, 4, or 8 GPUs

A100 80GB

$1.49per GPU hour

80 GB VRAM
112 GB RAM
16 vCPU
1, 2, 4, or 8 GPUs

A100 40GB

$0.89per GPU hour

40 GB VRAM
112 GB RAM
16 vCPU
1, 2, 4, or 8 GPUs

A30

$0.41per GPU hour

24 GB VRAM
112 GB RAM
16 vCPU
1, 2, 4, or 8 GPUs

L4

$0.44per GPU hour

24 GB VRAM
124 GB RAM
32 vCPU
1, 2, 4, or 8 GPUs

CPU VM 2 vCPU

$0.05per hour

2 vCPU
8 GB RAM
On-demand
India (Noida) region

CPU VM 4 vCPU

$0.10per hour

4 vCPU
16 GB RAM
On-demand
India (Noida) region

CPU VM 8 vCPU

$0.20per hour

8 vCPU
32 GB RAM
On-demand
India (Noida) region

CPU VM 16 vCPU

$0.40per hour

16 vCPU
64 GB RAM
On-demand
India (Noida) region

CPU VM 32 vCPU

$0.79per hour

32 vCPU
128 GB RAM
On-demand
India (Noida) region

* Pricing information is for reference only. Please refer to the latest announcement on JarvisLabs's official website.

FAQ

Running on-demand GPU instances are billed by the minute at the listed per-GPU hourly rate. You can pause and resume an instance; paused instances continue to incur storage charges.

Templates are managed containers with pre-built JupyterLab, VS Code, API endpoint, or SSH-only stacks. VMs provide root access for custom kernels, drivers, Docker, Kubernetes, daemons, and CI runners.

JarvisLabs does not offer a free trial. The service is pay as you go, and the homepage says an account can begin with a $10 credit top-up.

Storage is listed at $0.10 per GB per month, while in-region egress is free. GPU compute charges apply only while an on-demand instance is running.

Tool Information

DeveloperJarvisLabs

Release Date7/20/2026

Pricingpaid

Popular Trending

AI Function Categories