Skip to main content
📖 The AI Tool Bible
Velda preview image
Velda logo

Velda

Not yet source-checked

Serverless GPU Cloud without overhead of containers or cluster management.

Machine-drafted from the vendor's site · not yet source-checkedProfile updated Link checked Pricing captured How we rate

Freemium· Free plan; shell pool: $0.40 per hour for 4 vCPU, 16 GB RAMFine-tuning6.7 / 10

In short

Velda is a serverless GPU cloud for fine-tuning that charges only for active compute. It eliminates container management and idle costs for training and inference jobs.

Velda allows users to run training and inference jobs without wasting GPUs on idle time or setup. Velda charges users only for compute when jobs run, not during development or between workloads.

Users can prefix the vrun command with any command to instantly run workloads from the cloud or cluster. The platform eliminates the need for image builds, manifests, and dependency drift.

Developers can launch VS Code in the browser or connect with their preferred IDE. Velda provides prebuilt templates that users can tailor to their specific workflows.

The Enterprise tier provides self-hosted or dedicated infrastructure with premium support for organizations of any size.

Editor's take

According to Velda: Velda allows users to run training and inference jobs without wasting GPUs on idle time or setup. Its pages list this strength: Users pay only for compute when jobs are running, not during development or idle time. They also state this limitation: Instances currently have a storage limit of 50 GiB.

— The AI Tool Bible editorial team

Pros

  • ✅ The vendor says that no image builds, manifests, or dependency drift required.
  • ✅ Users pay only for compute when jobs are running, not during development or idle time.
  • ✅ Browser-based VS Code access or connection to favorite IDEs.
  • ✅ Instant start with prebuilt templates that can be tailored to specific workflows.

Cons

  • ⚠️ Instances currently have a storage limit of 50 GiB.

Pricing (captured )

TierMonthlyKey features
h100-1$4.90/hour
  • 1x H100 80 GB
  • 16 vCPU
h200-1$5.90/hour
  • 1x H200 141 GB
  • 16 vCPU
l40s-1-16d$2.10/hour
  • 1x L40s 48GB
  • 16 vCPU

Prices change; we show the grid as captured on Nov 15, 2026. Check current pricing on the vendor's site →

Facts

Platforms
web
Open source
Yes
Hacker News
2 stories · top 3 points · last May 14, 2026 · checked Oct 22, 2026
Earliest Wayback capture
2025-02-20 (earliest archived capture of the site, not a launch date)

Where this tool sits

Use cases

Training and inference jobsServing and data pipelinesMulti-processing data tasksDistributed data processing

Frequently asked

How much does Velda cost?
Velda offers a freemium model with a free plan. The shell pool costs $0.40 per hour for 4 vCPU and 16 GB RAM. Users are charged only for compute when jobs run, not during development or idle time.
Is Velda difficult to set up?
Velda eliminates the need for image builds, manifests, and dependency drift. Users can prefix the vrun command with any command to instantly run workloads. It also provides prebuilt templates to tailor to specific workflows.
Can I use my own IDE with Velda?
Yes, developers can launch VS Code in the browser or connect with their preferred IDE. This allows for flexible development environments while utilizing Velda's serverless infrastructure for running jobs.
What types of jobs can I run on Velda?
Velda supports training and inference jobs, serving and data pipelines, multi-processing data tasks, and distributed data processing. It is designed to prevent wasting GPUs on idle time or setup.
Does Velda offer enterprise options?
Yes, the Enterprise tier provides self-hosted or dedicated infrastructure with premium support. This option is available for organizations of any size that require specific infrastructure configurations.

Compare Velda with similar tools

All alternatives →

Explore related

Reviews