Amazon SageMaker Studio Lab gives you free compute: about 4 hours of NVIDIA T4 GPU per day plus about 8 hours of CPU per day, with 15 GB of persistent storage. If you’re searching for AWS free credits but you mostly need a no-cost place to train and test models, this is one of the cleanest options.
ML engineers prototyping in notebooks, students grinding through assignments, and founders trying to stretch runway all get the same core benefit. It’s browser-based JupyterLab 4, with real GPU access, and you don’t need an AWS account.
This guide covers eligibility, the exact signup flow, daily limits, what’s included (and what isn’t), and a few practical ways to squeeze more work out of your time.
Program at a Glance
| Provider | AWS |
| Credit Amount | 4 GPU hours/day + 8 CPU hours/day |
| Duration | Daily reset (per 24-hour period) |
| Eligibility | Individuals; one account per person/email |
| Credit Card Required? | No. No AWS account needed. |
| Difficulty | Intermediate; approval + phone verification required |
| Best For | Learning ML, prototyping, small model training |
| Official Page | AWS Program Page |
What You Actually Get
SageMaker Studio Lab is a completely free, browser-based ML development environment built on JupyterLab 4. You can start a CPU runtime (T3.xlarge: 4 vCPUs, 16 GB RAM) for up to 8 hours per day, or a GPU runtime (G4dn.xlarge with an NVIDIA T4 and 16 GB VRAM) for up to 4 hours per day. You also get 15 GB of persistent storage where notebooks, files, conda environments, and installed packages persist across sessions and reboots. Common frameworks are already available (PyTorch, TensorFlow, Keras, NumPy, scikit-learn, Pandas), and you can install others using conda, pip, or micromamba.
In real terms, this is enough to train small-to-medium deep learning models, fine-tune pre-trained models in short runs, and do GPU inference (Stable Diffusion is specifically called out as a reasonable fit). The bigger value, honestly, is persistence: you install packages once, set up your environment, and come back tomorrow without rebuilding everything from scratch like you often do on other free notebook platforms.
Who Qualifies (and Who Doesn’t)
SageMaker Studio Lab is meant for individual users who want a free ML environment without setting up an AWS account. The core constraint is account-level: AWS expects one account per person and per email, and you will have to complete a one-time phone verification at first runtime launch.
- You submit an access request and wait for AWS approval (often hours, sometimes a few days).
- A valid email address is required because you verify it during the process.
- You must be able to receive an SMS on a mobile number for the one-time phone verification.
- Stick to one account per person/email, because that’s explicitly part of the program rules.
If you try to create multiple accounts for extra GPU time, expect trouble. Also, if you’re in a region with known SMS delivery issues (AWS notes reports in China, Colombia, UAE, and Jordan), signup can fail even if everything else is fine.
How to Sign Up
Plan for a few minutes of form-filling, plus an approval wait.
- Go to studiolab.sagemaker.aws and click “Request free account”.
- Fill in the request form with your email address, first/last name, country, organization name, and occupation.
- Click “Submit request”, then check your email and click the verification link to verify your email address.
- Wait for approval. AWS says requests are reviewed within about 5 business days, but many people get approved within a few hours to a few days.
- Once approved, open the email with your registration link and claim your account within 7 days (the link expires).
- Create your Studio Lab account by choosing a username and password (this is separate from any AWS account).
- Verify your email again via the confirmation email.
- On your first runtime launch, complete one-time phone verification: enter a mobile number, receive a 6-digit SMS code, and verify.
- Choose a CPU or GPU runtime and click “Start runtime” to load JupyterLab in the browser.
After you’re in, your JupyterLab environment loads in the browser and you can switch between CPU and GPU between sessions. If your approval link expires after 7 days, you’ll need to submit a new request (annoying, but common).
What the Credits Cover
Studio Lab “credits” aren’t dollars you can spend across AWS. They’re fixed daily compute time on specific instances, plus persistent storage for your files and environments. The environment is real JupyterLab 4, which means terminals, extensions, Git workflows, and multiple notebooks all work the way you’d expect.
| Service / Feature | What It Does | Included? |
|---|---|---|
| CPU runtime (T3.xlarge) | Notebook compute for preprocessing, training, and scripts. | ✓ |
| GPU runtime (G4dn.xlarge, T4 16 GB) | Accelerates training and inference on a T4 GPU. | ✓ |
| Persistent storage (15 GB) | Keeps notebooks, files, and environments across sessions. | ✓ |
| JupyterLab 4 + extensions + Git | Full IDE-like experience with built-in Git integration. | ✓ |
Notable exclusions: you don’t get SageMaker production features like Pipelines, real-time endpoints, GroundTruth labeling, built-in algorithms/estimators, fine-grained IAM controls, or configurable instance types and storage. Studio Lab is a lab. Not a full cloud platform.
Limitations to Know About
Every free program has catches. Studio Lab’s are mostly about time, capacity, and the fact that it’s intentionally not “full SageMaker”.
- GPU usage is limited to about 4 hours per 24-hour period, and the session limit is 4 hours.
- CPU usage is limited to about 8 hours per 24-hour period, and the session limit is 4 hours.
- Only one runtime session can be active at a time, so you cannot run CPU and GPU simultaneously.
- Compute availability is not guaranteed; during peak demand you may not be able to start a GPU session right away.
- Time limit increases are not supported, even if you “need it for a project”.
- Storage is capped at 15 GB and there is no option to expand beyond that.
- File edits are periodically auto-saved during a session, but are not saved when the runtime ends (manual saves are recommended).
When your session time runs out, all running computations stop immediately. The good news is your files and installed packages are saved to persistent storage, so you can resume later, but you should expect to restart training jobs and rerun cells. Also, keep an eye on that save behavior: hit Ctrl+S before the session ends, because auto-save won’t rescue you after shutdown.
Have Unused AWS Credits?
Studio Lab itself is free time, not a bucket of spendable AWS credits. But lots of teams also have “real” AWS credits sitting around from startup programs or enterprise agreements, and they sometimes expire before the company can use them. If you’re staring at credits you won’t burn down in time, selling them is better than letting them die on the vine. AI Credit Mart lets you list unused AWS credits and recover a chunk of the value (often up to about 70% of face value).
Need More AWS Credits?
If you outgrow Studio Lab, the next step is usually paid AWS: bigger instances, longer runs, and production deployment. At that point you don’t necessarily have to pay full price, because discounted AWS credits are often available from companies with surplus allocations. On AI Credit Mart, AWS credits typically trade around 30% to 70% below retail, depending on size and terms.
Tips for Getting the Most Out of Your Credits
- Use checkpoints for GPU training, because you’ll want to resume in the next 4-hour session instead of restarting.
- Do data preprocessing on the CPU runtime so your GPU hours go to training and inference, not CSV wrangling.
- Install packages once and keep them, since conda/pip installs persist across sessions (unlike many free notebook options).
- Clone GitHub repos to keep runs reproducible, and use the built-in Git UI to push/pull without extra setup.
- Keep an eye on storage: model weights and checkpoints fill 15 GB fast, so delete old artifacts regularly.
- If the GPU won’t start due to demand, try again during off-peak hours (late night or early morning US time is often better).
- For workshops or classes, ask AWS for referral codes, which can bypass approval wait and grant instant access.
- If you plan to migrate to full SageMaker later, use the SageMaker Distribution environment to stay compatible.
Frequently Asked Questions
They aren’t dollar credits; you get about 4 GPU hours/day (NVIDIA T4 16 GB) plus about 8 CPU hours/day (T3.xlarge) and 15 GB persistent storage. In practice, that’s enough for repeated small training runs, short fine-tunes, and GPU inference experiments without paying anything. The real “value” comes from persistence: your conda envs and installed packages stick around, so you don’t burn time rebuilding your setup each session.
No.
Studio Lab’s compute limits reset each 24-hour period (4 GPU hours/day and 8 CPU hours/day), and storage persists while you have access to the service.
Yes. If you have AWS credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.
AI Credit Mart has discounted AWS credits available from companies with surplus allocations. Prices are typically 30-70% below retail.
When Studio Lab session time runs out, your running computations stop, but your files and installed packages remain in your persistent storage.
No. Only one runtime session can be active at a time, which means you have to choose CPU or GPU per session.
Studio Lab requires approval, and AWS says review can take up to about 5 business days (though many requests clear faster). If you’re in a hurry, a referral code from a workshop or hackathon can bypass the wait and grant instant access. Also check your inbox carefully: after approval, the registration link expires in 7 days, and if it expires you have to submit a new request. One more gotcha is SMS verification on first launch; AWS supports 240+ countries but has reported delivery issues in some regions, and VoIP numbers typically don’t work.
Studio Lab is real ML compute for free: a predictable T4 GPU, a solid CPU box, and storage that actually persists. Use it to learn and prototype fast, then graduate to paid AWS (or discounted credits) when you need production horsepower.
Your AI credits are losing value every day
Join the marketplace and start trading unused credits today.