Paperspace Review 2026: Pricing, Features, Pros & Cons
Paperspace was the friendliest way to rent a GPU. Then DigitalOcean bought it, and in 2026 the standalone product is being sunset into Gradient GPU Droplets — with a $39/month gate on high-end silicon and a rate card that changed again on August 1. Here is what the platform actually costs now, what still works well, and when to migrate.
Quick Verdict
Best for: students, researchers and small teams doing bursty notebook work on mid-range GPUs, where the free and $8 tiers are excellent value. Not for: production training on H100s — the subscription gate plus above-market hourly rates make RunPod or Vast.ai the cheaper answer.
What Is Paperspace?
Paperspace is a GPU cloud built around a managed notebook and persistent-machine workflow. Instead of provisioning infrastructure, you pick a GPU, open a Jupyter environment, and start working — with storage and packages that survive a shutdown. Gradient, the ML platform layer on top, adds workflows, distributed training, a model repository and deployment endpoints, sold as Free, Pro and Growth subscriptions with hourly utilization billed separately.
The defining fact of 2026 is ownership. DigitalOcean acquired Paperspace in 2023 and has been folding it into its own platform as Gradient GPU Droplets. The standalone Paperspace experience is being sunset, current-generation hardware — HGX H100, H200, AMD MI300X and MI325X — lives on the DigitalOcean side, and the pricing you see depends on which of the two consoles you are standing in.
That migration is the whole evaluation. The workflow Paperspace was loved for is still the best of its kind, and the free tier is still one of the very few ways to touch a real GPU without a card. But the economics have moved: the good GPUs sit behind a $39/month subscription, on-demand H100 capacity is $4.41 per GPU-hour after the August 1, 2026 price change, and the cheapest rates require a 12-month commitment at a moment when the rest of the market is getting cheaper and more flexible.
Running a GPU cloud or an ML platform? People land on this review while they are actively shopping for one.
Add it to the developer tools category — a free listing publishes after review, and it is the same page ChatGPT, Perplexity and Google read when someone asks which GPU cloud to use. Want it live in minutes with a Verified badge instead? That option is on the form, one-time, no subscription.
Paperspace Pros & Cons
✓ Pros
- •The notebook experience is still the friendliest in GPU cloud: a Jupyter environment with a chosen GPU attached is a few clicks away, with no Kubernetes, no container registry and no infrastructure ticket in between
- •A genuinely free GPU tier survives — public projects, 5GB of storage and a 12-hour auto-shutdown — which remains one of the few no-card ways to touch a real GPU before committing budget
- •Pro at $8/month is priced as a convenience fee rather than a platform fee: it buys private projects, 15GB storage, configurable auto-shutdown and priority on the faster free GPUs, and you still pay hourly only for what you run
- •Persistent machines behave like a workstation, not a job runner — your environment, packages and data survive a shutdown, which is a real advantage over pods that reset on every launch
- •DigitalOcean's backing means billing, support and account management are now a real company's processes rather than a startup's, and the same account reaches Droplets, Spaces and managed databases
- •GPU Droplets give access to current-generation silicon — HGX H100 and H200, plus AMD MI300X and MI325X — which the old Paperspace fleet never carried
- •Per-second billing on Droplets and per-hour machine billing are both transparent, and the docs publish the rate card openly instead of hiding it behind a sales call
- •For teaching, prototyping and coursework the platform is hard to beat: the gap between 'I have an idea' and 'a GPU is running my notebook' is smaller here than on any hyperscaler
✗ Cons
- •The standalone Paperspace product is being sunset and folded into DigitalOcean's Gradient GPU Droplets, so anything you build against the old console is on borrowed time — this is the single most important fact about the platform in 2026
- •High-end GPUs sit behind a $39/month Growth subscription, which means the A100/H100 tier carries a platform fee on top of the hourly rate before you have run a single job
- •DigitalOcean changed GPU pricing effective August 1, 2026, and the current rate card puts on-demand HGX H100 at $4.41/GPU/hour and H200 at $4.47/GPU/hour — well above what dedicated GPU marketplaces charge for comparable silicon
- •The cheapest headline rates require a 12-month reserved commitment ($3.26/GPU/hour for H100, $3.40 for H200), which is the opposite direction from where the rest of the market is moving
- •DigitalOcean states plainly that it does not offer refunds on GPU Droplets, so a misconfigured multi-GPU run that idles overnight is simply money spent
- •A powered-off Droplet keeps billing because its resources stay reserved until you destroy it — a billing model that catches out anyone who assumes 'stopped' means 'free'
- •Storage beyond the plan allowance bills at $0.29/GB/month, which quietly becomes the dominant line item on a project with large datasets or many checkpoints
- •Documentation and branding are split across paperspace.com, docs.digitalocean.com and DigitalOcean's own product pages, and the inconsistencies between them are exactly what you would expect from a platform mid-migration
Paperspace Pricing 2026
Free
- •Free GPU tier, availability-based
- •Public projects only
- •5GB storage
- •Auto-shutdown after 12 hours
- •Basic instances only
Learning, coursework and proving the workflow fits
Pro
- •Private projects
- •15GB storage
- •Configurable auto-shutdown
- •Mid-range instances
- •Priority on faster free GPUs
Solo researchers running mid-range GPUs occasionally
Growth
- •Everything in Pro
- •Access to high-end GPUs (A100, H100 class)
- •Larger storage allowance
- •Team collaboration features
- •Plus hourly utilization on every instance
Teams that need top-tier silicon inside this console
GPU Droplets
- •AMD MI300X $1.91, MI325X $2.88 per GPU/hr
- •On-demand HGX H100 $4.41/GPU/hr
- •On-demand HGX H200 $4.47/GPU/hr
- •12-month reserved H100 $3.26/GPU/hr
- •1 or 8 GPUs per Droplet
Production training and inference on current silicon
Subscription tiers are billed on top of hourly instance utilization — the monthly fee buys access, not compute. Storage above the plan allowance is $0.29/GB/month. GPU Droplet rates reflect DigitalOcean's price change effective August 1, 2026; reserved rates require a 12-month commitment and DigitalOcean states it does not offer refunds. Verify current rates before purchase.
Paperspace vs RunPod vs Vast.ai
| Feature | Paperspace | RunPod | Vast.ai |
|---|---|---|---|
| Subscription fee before any compute | ⚠️ $8 Pro, $39 Growth for high-end | ✅ None | ✅ None |
| Free GPU tier | ✅ Yes, availability-based | ❌ No | ❌ No |
| Managed notebook experience | ✅ Best in class | ✅ Good | ⚠️ Bring your own image |
| Current-gen GPUs (H100/H200/MI300X) | ✅ Via GPU Droplets | ✅ Yes | ✅ Yes |
| Hourly price for H100-class | $4.41 on-demand | Typically lower | Typically lowest (marketplace) |
| Hardware reliability guarantee | ✅ First-party datacentre | ✅ Secure cloud tier | ⚠️ Varies by host |
| Refunds on unused spend | ❌ Explicitly none | ⚠️ Credit-based | ⚠️ Credit-based |
| Platform stability in 2026 | ⚠️ Mid-migration into DigitalOcean | ✅ Stable | ✅ Stable |
Moving Off Paperspace: What Actually Changes
The migration most people face is not Paperspace to a competitor — it is Paperspace to DigitalOcean, whether they choose it or not. Practically, that means the notebook-first console gives way to GPU Droplets, which behave like ordinary cloud instances with GPUs attached. You get current silicon and per-second Droplet billing; you lose the click-to-notebook simplicity, and you inherit a billing model where a stopped Droplet still charges until it is destroyed.
If you are leaving the ecosystem entirely, the export burden is low by cloud standards. Your dependencies live in a Dockerfile or a requirements file, your data lives in object storage you can sync elsewhere, and your checkpoints are just files. The lock-in that hurts is habitual rather than technical: teams that treated the persistent machine as their workstation have configuration living only in that machine's disk. Snapshot it into a reproducible image before you decide anything.
The switcher hesitations worth naming honestly: marketplace providers such as Vast.ai are cheaper but hardware quality varies by host, so long unattended training runs carry more risk. RunPod sits in the middle — no subscription gate, reliable secure-cloud instances, container-shaped workflow. And the free tier is a genuine reason to keep an account open even after you move production elsewhere; nothing else in the category hands a student a GPU for nothing.
Who Should Actually Use Paperspace
Use it if: you are learning, teaching or prototyping and want a GPU attached to a notebook in under a minute; your usage is bursty and mid-range, which is exactly the shape the $8 Pro tier is priced for; you value a persistent environment that survives shutdown; or you are already a DigitalOcean customer and want GPU capacity on the same bill and the same support contract.
Skip it if: your workload is sustained training on H100-class hardware, where the $39 Growth gate plus a $4.41/GPU-hour on-demand rate compounds badly; you need the lowest possible price per GPU-hour, which is a marketplace answer, not a managed-platform answer; or you cannot tolerate building on a console that is explicitly being retired during the life of your project.
Frequently Asked Questions
How much does Paperspace cost in 2026?
There are two price layers and you pay both. The subscription layer runs Free ($0, public projects, 5GB storage, 12-hour auto-shutdown), Pro at $8/month (private projects, 15GB storage, mid-range instances) and Growth at $39/month, which is the tier that unlocks high-end A100 and H100 class GPUs. On top of that you pay hourly utilization for every paid instance you run. On the DigitalOcean GPU Droplet side, pricing changed effective August 1, 2026: on-demand HGX H100 is $4.41 per GPU per hour and HGX H200 is $4.47, while 12-month reserved plans bring H100 to $3.26 and H200 to $3.40. AMD MI300X is $1.91 and MI325X is $2.88 per GPU per hour. Storage above your allowance bills at $0.29/GB/month.
Is Paperspace shutting down?
The company is not shutting down, but the standalone Paperspace product is being sunset and absorbed into DigitalOcean's Gradient GPU Droplets after the 2023 acquisition. In practice that means the old console, the Gradient subscription tiers and the DigitalOcean GPU Droplet rate card all coexist in 2026, with new capacity and current-generation hardware landing on the DigitalOcean side. If you are choosing a platform today, choose it knowing the workflow you learn in the legacy console is the one being retired — build against GPU Droplets, not against Gradient notebooks you expect to keep for three years.
Paperspace vs RunPod — which is cheaper?
RunPod, in almost every configuration, because there is no subscription gate. Paperspace charges $39/month for Growth before you can touch high-end silicon, and the DigitalOcean on-demand H100 rate of $4.41/GPU/hour sits above what dedicated GPU clouds and marketplaces charge for the same card. Where Paperspace still wins is the managed experience: a notebook with a GPU attached, persistent storage that survives shutdown, and no container plumbing. If your work is bursty research on mid-range GPUs and you value the environment surviving between sessions, the $8 Pro tier is good value. If you are running long training jobs on H100s, the arithmetic favours RunPod or Vast.ai clearly.
Does Paperspace bill you when a machine is stopped?
Yes, and this is the billing trap that catches most new users. Paperspace CPU and GPU machines are billed differently from DigitalOcean Droplets, and a powered-off Droplet continues to incur compute charges because its resources stay reserved until you actually destroy it. Stopping is not the same as deleting. Combine that with DigitalOcean's stated no-refund policy on GPU Droplets and the practical rule is simple: destroy resources you are not using, and rely on auto-shutdown rather than on remembering.
Is the Paperspace free GPU tier actually usable?
For learning, yes. For work, no. The free tier gives you a GPU subject to availability, 5GB of storage, public projects only and a hard 12-hour auto-shutdown. That is enough to run a tutorial, fine-tune something small, or decide whether the notebook workflow suits you. It is not enough for a real project: 5GB does not hold a serious dataset plus checkpoints, and public projects rule out anything with proprietary data. Treat it as an evaluation environment, which is exactly what it is designed to be, and budget for Pro at $8/month the moment you need privacy.
Should I migrate off Paperspace in 2026?
Migrate if you are running production training on high-end GPUs — the $39 Growth gate plus above-market hourly rates plus a 12-month commitment for the good prices is a stack of costs that dedicated GPU providers simply do not impose. Stay if you are a researcher, student or small team whose work is bursty, notebook-shaped, and mid-range: the free tier and the $8 Pro tier are genuinely well priced for that shape of usage, and the persistent environment saves real setup time. Either way, plan for the console migration rather than being surprised by it, and keep your environment reproducible in a Dockerfile or requirements file so moving is a Tuesday afternoon rather than a project.
Compare GPU Clouds & AI Infrastructure
See how Paperspace compares to RunPod, Vast.ai, Lambda Labs and Modal before you commit to a rate card.
Affiliate disclosure: Some links on this page are affiliate links. If you sign up through them, AISO Tools may earn a commission at no extra cost to you. This never affects our rankings or reviews.
📬 Get the best new AI tools delivered weekly
One concise email with fresh launches, trending picks, and featured standouts.
Join thousands of professionals who discover the best AI tools every week. No spam — unsubscribe anytime.