Home›AI tools›Google ColabDeutsch
colabgooglecloud

Google Colab: the same notebook, on Google's machines, yours, or your cloud project

Colab is Google's hosted Jupyter notebook service. It ranges from a free version with GPUs and Gemini help to Colab Enterprise, which runs in your own Google Cloud project and region.

Colab sorts itself by where the code runs. In the free and paid versions it runs on a Google-managed VM tied to your Google account, and notebooks sit in Google Drive. With a local runtime it runs on your own hardware. In Colab Enterprise it runs on VMs in your Google Cloud project, with IAM, your network and regional storage.

Which model runs is Google's choice: the built-in assistant and the Data Science Agent use Gemini, and there is no documented model picker, as of September 2026. Whose account serves it depends on the version: a personal Google account under consumer terms, or your own Google Cloud project under the Cloud Data Processing Addendum. Any other model, such as an open Gemma model, you load onto the runtime yourself.

The family

Colab[1]

Web
A hosted Jupyter notebook service with free access to compute, including GPUs and TPUs. Notebooks are stored in Google Drive.

Gemini in Colab[1]

Web
Chat, code generation, error fixes and AI autocomplete in the notebook. On by default for eligible users aged 18 and over.

Data Science Agent in Colab[1]

Web
Plans an analysis, runs the code in your notebook and presents the results. You review the plan before it runs.

Colab CLI[2]

CLI
Connects your local terminal, or an agent with terminal access, to a remote Colab runtime to request a GPU or TPU, run scripts and download results.

Local runtime[3]

Web plus your machine
Connects the Colab web interface to Jupyter or Colab's Docker image on your own hardware.

Colab Enterprise[4]

Google Cloud console, BigQuery
A managed notebook environment in your own Google Cloud project. Part of Gemini Enterprise Agent Platform, formerly Vertex AI.

Gemini in Colab Enterprise[5]

Google Cloud console
Code completion, code generation, chat about the notebook and error explanations. Part of Gemini for Google Cloud.

Data Science Agent in Colab Enterprise[6]

Google Cloud console
Works with CSV files and BigQuery tables, including BigQuery ML, BigQuery DataFrames and Spark. Its code runs only in your notebook's runtime.

Plans

Consumer Colab is free, with paid plans for more compute; Colab Enterprise is billed through Google Cloud. Plans as of September 2026.

PlanForWhat changes
Colab (free of charge)[1]Students, hobbyists, anyoneNo guaranteed resources, limits that change over time, very limited GPU access, notebooks run at most 12 hours.
Pay As You Go[7]Occasional heavier useNo subscription. Compute units expire after 90 days. Faster GPUs and a standard Gemini quota.
Colab Pro[7]Regular usersMonthly compute units, faster GPUs, an expanded Gemini quota and high-memory machines.
Colab Pro+[7]Heavy usersMore compute units, priority access to premium GPUs, the highest Gemini quota, and background execution for up to 24 hours with the browser closed.
Google AI plans[8]Google AI subscribersPaid Google AI plans now include premium Colab benefits; Ultra adds background execution and Premium GPU access. Existing Colab subscriptions stay as they are.
Colab Enterprise[9]Organisations on Google CloudBilled on your Google Cloud account for the runtimes you use; the Data Science Agent is billed per token. IAM, regional storage and Google Cloud support.
Workspace accounts[1]Google Workspace usersA Workspace admin switches Colab on or off. Colab Pro for Education is closed to new signups.

Which model runs

Google picks the Gemini model behind the assistant and the Data Science Agent, in Colab and in Colab Enterprise. Which models run in your own code on the runtime is up to you. As of September 2026.

Gemini[1]built-in assistant and Data Science Agent in Colab; Google picks the version
Gemini for Google Cloud[10]Gemini in Colab Enterprise and its Data Science Agent
Open models you run yourself[2]loaded onto your runtime by your own code, e.g. Gemma 3 1B fine-tuned on a Colab T4

What you can configure

IAM[11]Colab Enterprise uses IAM roles such as Colab Enterprise Admin and Colab Enterprise User. You share a notebook with a user, a group or a Workspace domain.
Runtime templates[12]Choose machine type, GPU type and count, disk size and Python environment for Colab Enterprise runtimes.
Network and credentials[12]A runtime template sets the VPC network and subnet, and can turn off public internet access and end-user credentials. Idle shutdown runs from 10 to 1440 minutes.
VPC Service Controls[13]Runtimes sit inside your service perimeter; any other Google API the notebook calls must be in the perimeter too.
CMEK and Access Transparency[4]Customer-managed encryption keys for runtimes, notebooks and the Data Science Agent, and logs of any access by Google staff.
Gemini activation[14]An admin enables the Gemini for Google Cloud API in the project and grants users the Gemini for Google Cloud User role.
Schedules and capacity[4]Scheduled notebook runs store results in Cloud Storage. Compute Engine reservations secure capacity. Shared VPC is supported.
Consumer notebook settings[1]Sharing works like Google Drive. "Omit code cell output when saving this notebook" keeps outputs out of Drive.

What happens to your data

Colab (consumer), AI features[1]Google collects prompts, code, output and feedback and uses them to "provide, improve, and develop Google products". Human reviewers may read them. The data is kept up to 18 months, detached from your account. The FAQ describes no opt-out.
Colab Enterprise, Gemini[10]Gemini for Google Cloud "doesn't use your prompts or its responses as data to train its models". Prompts are handled under the Cloud Data Processing Addendum.
Colab (consumer), notebooks and runtimes[1]Notebooks live in Google Drive. Runtime VMs are private to your account and are deleted after idling or at a maximum lifetime.
Colab Enterprise, storage[4]Notebooks are stored regionally in Dataform; scheduled-run results go to Cloud Storage in your project.

Europe: GDPR, data residency and the AI Act

DPA and GDPR, Colab Enterprise[15]Colab Enterprise is a Google Cloud service under the Cloud Data Processing Addendum.
Terms for consumer Colab[16]Consumer Colab runs under the Google Terms of Service, the Colab Additional Terms and the Google Privacy Policy. Google documents no DPA for it and no regional storage.
Data residency in the EU[17]Colab Enterprise runs in 12 European regions, including Frankfurt, Berlin, Zurich, Paris, Milan, Madrid, Warsaw and the Netherlands. Data at rest stays in the notebook's location.
Gemini processing location[17]Google gives "no guarantee" that Gemini and Data Science Agent requests are processed in the notebook's region. The Data Science Agent is available in 7 of the European regions.
AI Act[18]Google said on 30 July 2025 that it will sign the EU General-Purpose AI Code of Practice. The Colab pages say nothing specific about the AI Act.

What the AI Act and the GDPR ask of you as a user →

Hybrid: the same models in your own account

Colab separates the notebook interface from the machine it runs on. The machine can be a VM in your own Google Cloud project and region, a computer you control, or a GPU your local agent borrows through the CLI.

Colab Enterprise in your own project and region[12]Runtimes are VMs in your Google Cloud project, in a region you choose, with the GPUs, VPC and IAM you set. You can load and run open models there yourself.
Local runtime on your own hardware[3]Run Jupyter or Colab's Docker image on your machine (tested with NVIDIA T4, L4 and A100) and connect Colab to it. Code runs on your hardware, but cell outputs still save to Google Drive unless you omit them.
Runtime on a VM you control[3]Start the runtime on a remote machine, for example a Compute Engine VM, forward the port over SSH and connect Colab to it.
Colab as a remote GPU for your agents[2]The Colab CLI lets Claude Code, Codex, Antigravity or other terminal agents request a Colab GPU and run jobs such as fine-tuning Gemma.
Limit: Marketplace VM route retired[19]Colab via GCP Marketplace, the old custom GCE VM runtime, was deprecated on 21 March 2025. Google Drive mounting does not work on local or Enterprise runtimes.
The same finished product, in three configurations: consumer, business, or on a cloud account you already have.
The same finished product, in three configurations: consumer, business, or on a cloud account you already have.[20]
Recently changed
  • Paid Google AI plans now include premium Colab benefits; existing Colab Pro and Pro+ subscriptions stay as they are. 22 Sep 2026.
  • The Colab CLI brings Colab GPUs and TPUs to local terminals and coding agents. 5 Jun 2026.
  • Vertex AI is now called Gemini Enterprise Agent Platform, and Colab Enterprise is part of it. Sep 2026.

What would help you next?

Learn to decide

The course “From Managed to Hybrid” shows where AI should run: what each layer costs, what you operate and where your data has to stay. In preparation, with a free live session first.

Linda MohamedLinda Mohamed
AI and cloud consultant, AWS Community Hero, Vienna

Roll it out with me

Tell me which tools your teams use or want. In 30 minutes we look at plans, settings, data rules and whether AWS funding can cover part of a pilot. How AWS funding works →

You already know what you need

Book a workshop directly:

ai-solutions.wiki

Read more on ai-solutions.wiki

My open knowledge base, with longer comparisons and sources.

The technology underneath: AI technologies explained →

Sources

  1. Colab, research.google.com. https://research.google.com/colaboratory/faq.html
  2. Colab CLI, developers.googleblog.com. https://developers.googleblog.com/introducing-the-google-colab-cli/
  3. Local runtime, research.google.com. https://research.google.com/colaboratory/local-runtimes.html
  4. Colab Enterprise, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/introduction
  5. Gemini in Colab Enterprise, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/chat
  6. Data Science Agent in Colab Enterprise, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/use-data-science-agent
  7. Pay As You Go, colab.research.google.com. https://colab.research.google.com/signup
  8. Google AI plans, developers.googleblog.com. https://developers.googleblog.com/colab-is-now-part-of-your-google-ai-plan/
  9. Colab Enterprise, cloud.google.com. https://cloud.google.com/colab/pricing
  10. Gemini for Google Cloud, docs.cloud.google.com. https://docs.cloud.google.com/gemini/docs/discover/data-governance
  11. IAM, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/access-control
  12. Runtime templates, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/create-runtime-template
  13. VPC Service Controls, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/service-controls
  14. Gemini activation, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/gemini-in-colab/set-up-gemini
  15. DPA and GDPR, Colab Enterprise, cloud.google.com. https://cloud.google.com/terms/data-processing-addendum
  16. Terms for consumer Colab, research.google.com. https://research.google.com/colaboratory/tos_v5.html
  17. Data residency in the EU, docs.cloud.google.com. https://docs.cloud.google.com/colab/docs/locations
  18. AI Act, blog.google. https://blog.google/around-the-globe/google-europe/eu-ai-code-practice/
  19. Limit: Marketplace VM route retired, research.google.com. https://research.google.com/colaboratory/marketplace.html
  20. Linda Mohamed, course slides “Start here”, MASE-3, FH Burgenland, September 2026
  21. Managed vs reserved vs self-hosted inference, ai-solutions.wiki. https://ai-solutions.wiki/comparisons/managed-vs-reserved-vs-self-hosted-inference/
  22. vLLM, ai-solutions.wiki. https://ai-solutions.wiki/tools/vllm/
  23. Google Cloud services with data residency (Agent Platform, formerly Vertex AI). https://cloud.google.com/terms/data-residency
  24. Colab Enterprise runtimes. https://docs.cloud.google.com/colab/docs/runtimes
  25. EU General-Purpose AI Code of Practice. https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai

Checked against the vendors' own pages in September 2026. Plans, models and features change often: check the linked pages before you decide.