AWSAmazon SageMakerMachine Learning

aws_sagemaker_notebook_instance cost estimation

A managed Jupyter notebook instance billed per hour while running. An ml.t3.medium is ~$37/month if left on 24/7 — and idle notebooks are the classic waste.

An aws_sagemaker_notebook_instance is a managed Jupyter environment for ML development. It bills per instance-hour by type, for every hour it's running — whether or not anyone is using it. An ml.t3.medium is ~$0.05/hour, about $37/month if left on around the clock; an ml.m5.2xlarge is ~$0.461/hour, ~$337/month.

The defining cost problem is idle time. Notebooks get started for an afternoon's work and left running for weeks, billing nights and weekends for an environment nobody's touching. The single biggest saving is simply stopping notebooks when not in use — a stopped notebook bills no instance time (only its EBS volume).

The levers: right-size the instance type to the actual development work (most exploration runs fine on a t3/m5; reserve GPU instances for genuine training), stop notebooks when idle (lifecycle configs can auto-stop them), and move heavy training to transient training jobs rather than a big always-on notebook.

c3x prices the notebook from instance_type, so the running cost — and the cost of leaving it on — is visible up front.

Terraform example

A minimal but realistic configuration that C3X can estimate.

resource "aws_sagemaker_notebook_instance" "dev" {
  name          = "ml-dev"
  role_arn      = aws_iam_role.sagemaker.arn
  instance_type = "ml.t3.medium"
  volume_size   = 20

  lifecycle_config_name = aws_sagemaker_notebook_instance_lifecycle_configuration.auto_stop.name
}

Pricing dimensions

What you actually pay for when you provision aws_sagemaker_notebook_instance.

DimensionUnitWhat's being charged
Notebook instanceper hourPer instance-hour by type, for every hour the notebook is running. Stopped notebooks bill no instance time.
$0.05/hour for ml.t3.medium → ~$36.50/month if on 24/7
EBS volumeper GB-monthThe attached storage volume bills per GB-month, including while the notebook is stopped.

Sample C3X output

An ml.t3.medium notebook left running 24/7 (the avoidable case):

aws_sagemaker_notebook_instance.dev
└─ Notebook instance (ml.t3.medium)   730 hours   $36.50
                                      Monthly     $36.50

Optimization tips

Common ways to reduce aws_sagemaker_notebook_instance cost without changing the workload.

Auto-stop idle notebooks

Up to ~75% on a part-time notebook

A lifecycle configuration that stops the notebook after an idle period is the highest-value fix — it ends the nights-and-weekends billing for an environment nobody's using. A notebook used 8 hours a weekday costs roughly a quarter of a 24/7 one.

Right-size the instance type

Large vs an oversized GPU notebook

Most exploration and light work runs fine on ml.t3.medium or ml.m5.xlarge. GPU and large instances cost many times more — reserve them for work that genuinely needs the hardware.

Run training as jobs, not in the notebook

Notebook stays small; training is transient

Heavy training on a big always-on notebook pays for that instance continuously. Submitting transient SageMaker training jobs (with Managed Spot) runs the compute only for the job's duration.

Consider SageMaker Studio

Workload-dependent

Studio offers more granular, per-app compute control than a standalone notebook instance, which can reduce idle-instance waste for teams.

FAQ

Does a SageMaker notebook cost money when idle?

Yes — a running notebook bills its instance-hour rate whether or not you're using it. Only stopping it ends the instance charge (the EBS volume still bills). Idle running notebooks are the most common SageMaker waste.

How do I stop paying for an idle notebook?

Stop it when not in use, and attach a lifecycle configuration that auto-stops it after an idle period so it doesn't bill overnight and on weekends. This is the single biggest notebook saving.

How does c3x estimate the cost?

From instance_type, pricing the per-hour rate (shown at 24/7 as the ceiling). Real cost depends on running hours — auto-stop and part-time use bring it well below the ceiling.

Related resources

Estimate this resource in your own Terraform

Free, open source, no API key. C3X parses your Terraform and shows line-item cost for every resource, including aws_sagemaker_notebook_instance.