Built for Malaysian researchers finishing a thesis or FYP, and startups that had to ship a live model yesterday. Train and host on a fractional GPU, billed by the hour, in MYR or by card. No contract, no waitlist, no midnight file wipe.
Built by a small team already hosting real thesis projects and startup MVPs. Tell us what you're building and get free credit to try it.
Get Early Access Credit"Colab kicks you out, Kaggle caps your hours, our AI lab wipes files at midnight. This just runs overnight without any of that."
Masters student, Swinburne Sarawak
PyTorch and the usual stack, pre-installed. No setup day.
Run jobs overnight from the terminal. Pay only for hours used.
Turn a trained model into an API from the same screen.
Runs, checkpoints, and live models, all in one place.
Right-sized GPU for your model, not a full card you don't need.
DuitNow, TnG, or card. MYR locally, any currency abroad.
RTX 3090 fractional GPUs today. More GPU types are being onboarded as we expand.
PyTorch, Scikit-learn, Tensorflow, Numpy and many more
CUDA 12, kept current as new versions land, so you're not stuck on an old toolkit.
Start a script from the terminal and let it run unattended, including overnight.
Skip training. Call Qwen or OpenAI's open weight models straight from a single endpoint.
Qwen and OpenAI open weight models live now, more added regularly.
One API endpoint for every model.
Pay per token. No GPU to reserve.
Runs on our own infrastructure. Delete a session, your data's gone, encrypted, no access even from us.
1 · Select your GPU
2 · Choose your VRAM
Spin up a fractional GPU, run your job, shut it down. No daily minimums, no idle charges.
Take a small fraction for a small model, or a full GPU for a bigger one. Resize as your project grows, right from your workspace.
No annual commitment, no sales call, no fine print. What you see is what you pay. Good for a semester or a seed round.
In Malaysia. Your workspaces, volumes, and the open-weight models behind our LLM gateway run on GPU clusters in Malaysia, operated by our data center partners. Your data is handled by Sovernstack, encrypted, and kept in isolated containers, so it is not shared with other users or accessible to the data center operators. You can see who hosts the hardware on our partners page.
Create an account and you get free credit on registration. From your workspace, start a Jupyter notebook or an SSH session, pick any available instance, choose how much RAM and VRAM you need, and start building. When you are done, shut it down and you stop paying.
AI developers, data scientists, and postgraduate and undergraduate students working on AI projects or research. If you need a GPU for training, fine-tuning, or experimenting, and you don't want to pay for a whole card or sign a contract, it's built for you.
Yes. Start your script from the terminal in your workspace and it keeps running whether or not you keep the browser open. An 8 hour or longer job is a normal use case, not an edge case.
We charge per hour, per VRAM slice. Pick the slices you need and pay only while your instance is running. We also offer dedicated plans. Get in touch and we'll tailor one to your workload, so you don't end up paying for things you don't need.
You are billed for the hours your fractional GPU is running, at the rate for the fraction you chose. An 8 hour run costs roughly 8 times the hourly rate. Use our cost calculator to see what your specific job would cost before you run it.
We partner with data center operators only in Malaysia to run the instances, while your data is securely handled by Sovernstack. You can see who we work with on our partners page.
Any credit you already have stays valid. You are not locked into the hardware available on the day you signed up.
We host open-weight models on our partner GPU clusters, so you can call them without managing any infrastructure. If one cluster goes down, another takes over automatically, so your requests keep working.
Yes. All data is encrypted, and every workspace runs in an isolated container. Your code, datasets, and models are not shared with other users and are not accessible to Sovernstack or to the data center operators hosting the hardware.
If you delete a volume, its data is gone and we don't keep a copy. If you delete your account, we delete everything. Back up anything you want to keep before you delete it.
Notebook, training, and hosting in one project.
Made for builders, not a procurement cycle.
Pay only for the GPU fraction you actually use, from RM0.60/hour.
MYR billing locally, card billing abroad.
Talk to the people building it, not a ticket queue.
Open a workspace and see if it fits what you are building. If it does not, tell us why, we are still shaping this.
Working on a research project, a thesis, or a startup idea? Send us a message and we will get back to you.
Data center operator or ecosystem partner in Malaysia? Visit our partners page.
