The best Paperspace alternatives in 2026 are Nodus for training runs that have outgrown a notebook, RunPod for cheap per-second GPU pods, and Kaggle Notebooks for students who need a free GPU. Nodus lists the NVIDIA H100 at $2.60 per GPU-hour on its pricing page, against $5.95 on Paperspace, and restores your latest checkpoint on a replacement machine when one is lost. Stay on Paperspace if its $8 or $39 monthly plan already gives you free notebook GPUs that cover your hours.
We build Nodus, so read this list with that in mind. Every fact about another company links to its own pricing page or docs, and each sum is written out so you can redo it with today's prices.
What does Paperspace offer and charge today?
Paperspace is now part of DigitalOcean, and its notebooks still exist. DigitalOcean's Paperspace docs list four products: Notebooks (a browser Jupyter IDE), Deployments, Workflows and Machines, which are Linux virtual machines with GPU options. The Notebooks, Deployments and Workflows products and the subscription plans still carry the Gradient name, per DigitalOcean's documentation index.
Two prices stack up. A Gradient subscription plan is Free, Pro at $8 per month or Growth at $39 per month, and each one includes a set of free notebook machines. Paid machines then bill by the hour: an A100-80G is $3.18/hr, an A6000 $1.89/hr and an H100 $5.95/hr, and an H100 x 8 machine is $47.60/hr, all from DigitalOcean's Paperspace pricing page.
The free machines come with limits. Per the Machines features page, they shut down after at most six hours, notebooks idle for an hour shut down, and only one notebook can use a free machine at a time. On the Free plan, free machines are always public.
H100s take a request. The H100 reference says they run only in the NY2 region and need approval from support, and the Machines limits page says requests typically take 1 to 2 business days.
How do the Paperspace alternatives compare?
They split along one line: a notebook you sit in, or a run that finishes without you. Paperspace and Kaggle are built around the first. RunPod gives you a GPU pod and leaves the run to you, and Nodus runs a command or training job to completion.
| Platform | What you get | Free option | A100 80 GB per hour | H100 per GPU-hour | If the machine is lost |
|---|---|---|---|---|---|
| Paperspace | Notebooks and virtual machines | Free plan with an M4000 notebook GPU | $3.18 | $5.95 | Not stated in the docs we read |
| Nodus | Jobs, Workspaces with JupyterLab, TrainingJobs | None; prepaid credit | $1.090 | $2.600 (H100 SXM) | Latest checkpoint restored on a replacement machine |
| RunPod | GPU pods billed per second | Not listed on the pricing page | $1.79 Secure, from $1.39 Community (A100 SXM) | $3.99 Secure, from $2.69 Community (H100 SXM) | Not stated on the pricing page |
| Kaggle | Notebooks | P100 or 2 x T4, 12-hour sessions | Not listed | Not listed | Not stated; sessions end at 12 hours |
Prices checked November 11, 2026: Paperspace pricing, Nodus pricing, RunPod pricing and Kaggle Notebooks docs. Nodus lists GPU rates and CPU-plus-memory rates as separate lines; Paperspace lists one hourly price per machine.
If you are leaving Paperspace for a plain virtual machine at its parent company, our DigitalOcean GPU Droplet alternatives post covers that comparison. For pods, our RunPod alternatives guide goes deeper on RunPod's two clouds.
What does it cost to turn a notebook into a training run?
On listed rates, the same run costs less than half as much on Nodus as on a paid Paperspace machine. We priced two runs a notebook user typically graduates to, using each company's public pricing page on November 11, 2026, GPU compute only.
The first is a 10-hour fine-tune on one A100 80 GB. Paperspace's A100-80G at $3.18/hr comes to $31.80. RunPod's A100 SXM is $17.90 on Secure Cloud at $1.79/hr and from $13.90 on Community Cloud at $1.39/hr. Nodus at $1.090/hr comes to $10.90, which is $20.90 or 66% less than Paperspace.
For a single-GPU fine-tune, a paid Paperspace A100 costs about three times the Nodus rate.
The second is a bigger run: 8 x H100 for 10 hours, or 80 GPU-hours. Paperspace's H100 x 8 machine at $47.60/hr comes to $476.00. RunPod is $319.20 on Secure Cloud at $3.99 per GPU-hour and from $215.20 on Community Cloud at $2.69. Nodus at $2.60 comes to $208.00, which is $268.00 or 56% less than Paperspace.
At list prices, 80 H100 GPU-hours cost $268.00 less on Nodus than on a Paperspace H100 x 8 machine.
Two caveats keep this fair. Nodus adds CPU and memory lines on top of the GPU rate, so run nodus run --dry-run and compare that estimate, not this table. RunPod's Community rates are "from" prices, so your pod may cost more.
When is a Paperspace plan cheaper than paying by the hour?
A Paperspace plan wins once your free-machine hours pass a break-even point. Pro costs $8 a month and includes a free A4000 notebook machine. Growth costs $39 a month and includes free A6000 and A100-80G machines, per the free machines table on the Paperspace pricing page.
Divide the plan price by the Nodus rate for the same card. Against the RTX A4000 at $0.173/hr, Pro pays for itself after 46.2 hours a month ($8 / $0.173). Growth breaks even after 35.8 hours of A100 80 GB at $1.090/hr ($39 / $1.090), or 117.1 hours of RTX A6000 at $0.333/hr ($39 / $0.333). Counting Nodus CPU and memory lines would make each break-even come sooner.
If you live in an A100 notebook more than about 36 hours a month, the $39 Growth plan is the better deal.
The catch is that free sessions are capped. Each one stops after at most six hours, and the paperspace.com pricing page describes faster free GPUs as "based on availability".
So here is the decision rule we use. Interactive notebook hours belong on a Paperspace plan. Any run longer than six hours, or one that must finish by morning, belongs on a platform that bills only for the run and restarts it after a lost machine.
Where is Paperspace still the better choice?
Paperspace is better for cheap, steady notebook time. A student on the Free plan gets a free M4000 GPU and 5 GB of storage for $0. Nodus has no equivalent: our credits docs say the public starter offer has ended and new accounts add prepaid credit first.
It is also the simpler bill if you want a monthly ceiling. Plans and storage on the pricing page have a maximum monthly price, after which they stop charging for the cycle. Paperspace Machines come with free unlimited bandwidth, per the Machines features page, and compute billing stops when you power a machine off.
Pick Paperspace too if your work is mostly in a notebook and fits a six-hour session, or if your team already shares projects and datasets there. Moving costs you setup time, and the savings only show up once you run long jobs.
Which Paperspace alternative is best for students?
For a student with no budget, Kaggle Notebooks is the strongest free option we found. The Kaggle Notebooks docs offer a free NVIDIA Tesla P100 or two T4 GPUs, 12-hour sessions and 20 GB of auto-saved output. Kaggle says free GPU availability is limited, so you may wait in a queue at busy times.
Paperspace's Free plan is the second free option, with shorter six-hour sessions and public notebooks. Once a class project needs an A100 or H100 for a few hours, paying by the hour beats a monthly plan. Three hours on a Nodus A100 80 GB at $1.090/hr is $3.27, against $39 for a month of Growth.
How do you move a Paperspace notebook to Nodus?
Start a Workspace, which is the closest Nodus match to a Paperspace notebook. Per the Workspaces guide, it is a GPU or CPU machine with JupyterLab, browser VS Code and SSH. Its /home/nodus directory is saved whenever it stops, and by default it stops after an hour idle.
$ pip install nodus-compute $ nodus login $ nodus create workspace lab --gpu H100 Upload your notebook and data to the home directory and develop as you did on Paperspace. When the code is ready for a long run, send it out as a Job instead of leaving a notebook open overnight:
$ nodus run --from workspace/lab -- python train.py The Job runs on a saved copy of your Workspace home, so you can stop the Workspace while it trains. To survive a lost machine, write model, optimizer and step files under /nodus/state and load them on start, as the checkpoints guide shows; our post on resuming training after spot preemption walks through a full loop.
FAQ
Is Paperspace still available in 2026?
Yes. DigitalOcean's Paperspace docs still document Notebooks, Deployments, Workflows and Machines, and its pricing page was updated in July 2026. Windows templates were retired for new users in July 2024, so new accounts get Linux machines.
Can I connect my Paperspace account to Nodus?
No. Paperspace is not one of the providers on the Nodus cloud accounts page. DigitalOcean is, but that connection covers CPU and GPU Droplets, not Paperspace Machines or Notebooks.
Does Paperspace have multi-GPU machines?
Yes. The pricing page lists 8-GPU machines of the A6000, V100-32G, A100, A100-80G and H100, and the H100 x 8 at $47.60/hr is exactly 8 times the single-GPU rate. DigitalOcean's H100 reference says the H100 x 8 supports NVLink and the H100 x 1 does not.
Is Nodus a good fit if I only use notebooks?
Only if you also run long jobs. Without a free tier, Nodus suits people who pay for compute anyway and want runs that finish; a free Kaggle or Paperspace notebook is cheaper for occasional exploration.
Originally published on the Nodus blog.
