Skip to main content
Serverless AI is a Nebius AI Cloud service for running containerized AI workloads without creating or managing virtual machines (VMs) or clusters. It supports the full AI/ML development workflow: build prototypes and run experiments in interactive Devlab environments, run training, fine-tuning and data processing workloads as jobs, and serve models and applications through endpoints. Provide a container image or select a template, and Serverless AI provisions the compute and storage resources required for each workload. Devlabs, endpoints and jobs run on Compute container VMs, with usage-based, per-second billing. The underlying VMs are managed by Serverless AI. You can connect to them via SSH, but they aren’t shown as Compute resources and can’t be managed or modified directly like other Compute VMs.

​
Devlabs, endpoints and jobs

Serverless AI supports the end-to-end AI/ML workflow through three deployment types. Choose the deployment type that best fits what you want to do:
  • Experiment and prototype: Write code, explore models and data in a browser-based IDE or notebook → Devlab.
  • Run a task to completion: Train or fine-tune a model, preprocess data, run batch inference or evaluation → Job.
  • Serve requests: Host a model or application at a URL for real-time inference → Endpoint.
Rather than providing three separate paths, these deployment types are designed for three stages that most projects move through. You can prototype in a Devlab, submit the working code as a job and then serve the result as an endpoint. Starting with one doesn’t prevent you from using the others. The following table compares the three deployment types:

​
Observability and debugging

Each Serverless AI deployment (a Devlab, an endpoint or a job) has a status that indicates the current stage in the lifecycle. You can view its status and logs on the resource page under Devlabs, Jobs or Endpoints in the web console. Deployments also report GPU and vCPU utilization metrics on the Metrics tab of this page. To investigate errors or unexpected outcomes, see Debugging failed deployments. For metric definitions and details, see Monitoring deployments.

​
Pricing and quotas

Serverless AI follows Compute billing and quota rules. Billing depends on whether the deployment is running or stopped:
  • While a Devlab, job or endpoint is running, you are billed for its computing resources and storage.
  • While a Devlab is stopped, you are billed for the entire container disk, but not for computing resources.
  • While an endpoint is stopped, you are not billed for computing resources or storage.
Mounted Object Storage buckets and shared filesystems are billed separately, including when the deployment is stopped. For more details, see Pricing and quotas. “Jupyter” and the Jupyter logos are trademarks or registered trademarks of LF Charities, used by Nebius B.V. with permission.