Paperspace review: quick verdict
Paperspace provides a dual-model GPU cloud featuring Gradient for managed Jupyter notebooks and Core for dedicated virtual machines with full root access.[source][source] It is highly suitable for developers who want to avoid data transfer costs due to its zero-egress network policy, but users must weigh this against monthly subscription overhead and physical capacity limits.[source][source] GPU Picks did not run tests, trials, or benchmarks.
Key takeaways
- The service model is divided into Gradient (managed Jupyter notebooks and workflows) and Core (dedicated GPU/CPU virtual machines).[source][source]
- Compute billing is hourly and stops when virtual machines are shut down, but non-compute resources (such as block storage and public IPs) continue to charge hourly until they are destroyed.[source]
- High-end GPU instances (such as NVIDIA A100 or H100) are restricted behind subscription tiers, specifically the Growth plan.[source]
- Block storage is subject to a maximum monthly cap, and there are no fees for data ingress or egress.[source]
- Uptime availability is covered by a 99.99% SLA, and customer support ranges from free Standard support to paid Enterprise assistance.[source][source]
The verdict comes from official pricing, documentation, and trust platform specifications accessed on the dates in the source list. GPU Picks did not run a paid trial or benchmark. Documentation explains how the service is designed, but it cannot show how quickly or reliably a reader's application will run.
How Paperspace works
Paperspace divides its cloud infrastructure into two distinct service lines: Gradient and Core.[source][source] Understanding the operational boundaries of these two models is essential before selecting a GPU environment, as they target different workflows and introduce different billing constraints.
Since being acquired by DigitalOcean, Paperspace has integrated more closely with DigitalOcean's billing systems, documentation standards, and trust platform, yet it retains its specialized GPU-focused compute offerings.[source] The division between containerized, notebook-first workflows and traditional dedicated virtual machines (VMs) remains the defining structural feature of the platform.
The Gradient service model
Gradient is designed around managed Jupyter notebooks, containerized workflows, and serverless endpoints.[source] Rather than managing a full operating system, developers use Gradient to access interactive development environments directly in the browser or deploy models behind API endpoints.
This model is optimized for machine learning researchers and developers who want to write code, train models, and deploy inferences without handling virtual machine configuration, SSH key setups, or manual driver installations. Within Gradient:
- Managed Notebooks allow immediate access to a JupyterLab interface backed by dedicated or shared GPU resources.
- Workflows enable developers to orchestrate multi-step containerized pipelines for data preprocessing, training, and evaluation.
- Deployments provide serverless endpoints that scale workers to handle incoming inference requests.
Gradient relies on containerization. The underlying compute nodes are managed by Paperspace, meaning users do not have root access to the host operating system. This abstraction simplifies setup but limits customization, as users cannot install low-level system packages or modify the kernel.
The Core service model
Core is a traditional infrastructure-as-a-service (IaaS) model.[source] It provides dedicated virtual machines with full root access, allowing developers to configure the operating system, install custom packages, manage drivers, and run arbitrary background processes.
Core virtual machines are standard Linux or Windows instances that can be accessed via SSH, remote desktop protocols, or the Paperspace web console. This model is preferred when:
- The workload requires complex multi-process orchestration, custom networking configurations, or non-containerized software.
- The application relies on persistent background daemons or services that must run continuously without being wrapped in a Gradient workflow.
- The team wants to use their own deployment tools, orchestrators, or custom machine images.
With Core, the user is responsible for managing the operating system, security updates, firewall rules, and software dependencies. While Paperspace provides template images with pre-installed GPU drivers, the operational ownership belongs entirely to the customer.
Paperspace pricing and total cost
A major factor in evaluating Paperspace is the combination of hourly compute billing, monthly subscription tiers, and storage costs.[source][source] A simple comparison of hourly GPU rates will not reveal the actual monthly total.
| Cost area | Official policy | What to verify before deployment |
|---|---|---|
| Compute billing | Hourly billing for running machines; compute billing halts when shut down.[source] | Verify instance availability and check if your VM is fully stopped in the console. |
| Subscription plans | Free ($0/mo), Pro ($8/mo), or Growth ($39/mo) plans determine GPU access.[source] | Check if your target GPU is gated behind a higher subscription plan.[source] |
| Block storage | Tiers from 50 GB ($0.0074/hr, capped at $5.00/mo) to 1 TB ($0.0818/hr, capped at $55.00/mo).[source] | Monitor storage usage and select the appropriate tier for your persistent data.[source] |
| Gradient storage | Included base storage with additional usage billed at $0.29/GB/month.[source] | Track active files and clean up unused datasets to avoid extra monthly charges.[source] |
| Network bandwidth | Zero fees for network ingress or egress to the public internet.[source] | Confirm that your application does not require external paid transit proxies. |
Compute resources are billed hourly, and compute billing halts when the machine is shut down.[source] However, non-compute resources (such as block storage and public IPs) continue to be billed hourly when instances are powered off, until they are destroyed.[source]
Subscription tiers and GPU access gates
Paperspace gates access to specific instances, features, and free GPU instances behind three subscription plans:[source]
- Free ($0/mo): Provides access to basic CPU and low-end GPU instances, standard community features, and basic auto-shutdown limits.[source]
- Pro ($8/mo): Unlocks additional GPU instances, longer running times, and larger storage allowances.[source]
- Growth ($39/mo): Required to access high-end GPUs, including dedicated NVIDIA A100 or H100 instances, and provides maximum auto-shutdown limits.[source]
This subscription structure means that if a team wants to rent an NVIDIA H100, they must add the fixed $39/mo subscription overhead to their hourly calculations.[source] For occasional users, this monthly overhead can make short-term training runs significantly more expensive than on platforms with no subscription requirement.
Zero-egress network policies
Unlike legacy hyperscalers that charge significant fees for data egress, Paperspace does not charge fees for data ingress or egress.[source] Bandwidth to and from the public internet, as well as traffic within a virtual private cloud (VPC), is free of charge.[source]
For workloads that involve transferring large datasets, such as video processing, large-scale model training, or high-throughput inference, this zero-egress policy can result in major cost savings. It eliminates the unpredictability of bandwidth bills, which are often difficult to estimate in advance.
Important features and block storage
Beyond compute and subscriptions, several platform features shape the operational experience on Paperspace, including block storage mechanics, service level agreements, and support structures.
Block storage pricing tiers and caps
Paperspace Core virtual machines use block storage that is billed hourly but subject to a maximum monthly cap.[source] This pricing model provides cost predictability for storage-heavy workloads.
Block storage tiers scale as follows:
- 50 GB Tier: Billed at $0.0074/hr, with a maximum monthly cap of $5.00/mo.[source]
- 100 GB Tier: Billed at $0.0148/hr, with a maximum monthly cap of $10.00/mo.[source]
- 250 GB Tier: Billed at $0.0372/hr, with a maximum monthly cap of $25.00/mo.[source]
- 500 GB Tier: Billed at $0.0744/hr, with a maximum monthly cap of $50.00/mo.[source]
- 1 TB Tier: Billed at $0.0818/hr, with a maximum monthly cap of $55.00/mo.[source]
Once the hourly storage billing reaches the monthly cap during a billing cycle, charges for that storage volume halt for the remainder of the cycle.[source]
For Gradient, storage is governed by the subscription plan's base allowance.[source] Any storage usage exceeding the base subscription allowance is billed at a flat rate of $0.29/GB/month.[source] Users must monitor their workspace storage and clean up large artifacts, weights, or datasets that are no longer needed to prevent unexpected storage bills.
Virtual machine lifecycle and billing states
When a Core virtual machine is shut down or powered off, compute billing stops, but storage and network resource billing continues.[source] This is because the underlying block storage volume and any reserved public IP addresses remain allocated to the user's account.
To halt all billing for a resource, the virtual machine and its associated storage must be destroyed.[source] Users who leave stopped VMs in their console expecting zero charges are often surprised by storage fees. Always check the active resource list in the console to ensure that unused volumes are fully deleted.
Service level agreements and support
Paperspace provides a 99.99% uptime Service Level Agreement (SLA) covering network availability, power, virtual server availability, and storage availability.[source] If availability falls below this threshold, users are eligible for financial credits in accordance with the SLA terms.[source]
Support is structured into two main tiers:[source]
- Standard Support: Free of charge for all accounts, providing ticket-based assistance 7 days a week for basic troubleshooting, billing queries, and platform issues.[source]
- Enterprise Support: Contracted support designed for teams requiring rapid response times, dedicated account management, and deep infrastructure assistance.[source]
Who should use Paperspace
Evaluating a GPU cloud requires matching your workload to the provider's strengths. Paperspace is a strong candidate for several specific use cases:
- Individual Machine Learning Researchers: Researchers who rely heavily on Jupyter notebooks can benefit from Gradient's managed environment.[source] It eliminates infrastructure configuration, allowing them to focus on code and experimentation.
- Teams Seeking Predictable Network Costs: If your workload requires moving terabytes of data, such as fine-tuning models on large remote datasets or serving high-throughput media models, the zero-egress network policy ensures that you will not face unexpected bandwidth charges.[source]
- Developers Requiring Windows VM Environments: While many specialized GPU clouds only offer Linux containers, Paperspace Core supports Windows virtual machines with full desktop access, making it a viable option for game development, rendering, or Windows-only software testing.[source]
- Startups Building on DigitalOcean: Startups that already run their main applications on DigitalOcean can consolidate their infrastructure management, billing, and trust compliance under a single corporate umbrella.[source]
For these profiles, the combination of managed notebook environments, zero egress fees, and traditional VM options matches their technical and financial requirements. If you fit this criteria, you can use the compare live GPU prices tool to check currently available instances.
Who should skip Paperspace
While the platform has several advantages, there are clear reasons to look for alternatives:
- Hobbyists Needing High-End GPUs for Short Runs: If you only need to run a large model on an NVIDIA A100 or H100 for a few hours, the requirement to purchase a $39/mo Growth subscription makes the entry cost disproportionately high.[source] A provider with no subscription fees would be more cost-effective.
- Teams Requiring Bare Metal Orchestration: If your workflow requires bare metal servers, low-level InfiniBand clustering, or orchestrators like Slurm for large-scale distributed training across hundreds of nodes, Paperspace's VM-based Core model is not the right fit.
- Users Seeking Purely Market-Driven Rates: If you want a bidding-style marketplace where spare capacity is sold at steep discounts, you will find Paperspace's fixed subscription and hourly rate model rigid.
- Applications Demanding Guaranteed High-End Availability: High-end GPUs are subject to physical capacity constraints. If your application requires instant, guaranteed scaling of A100 or H100 instances without reservation agreements, you may experience capacity limits.
If these constraints affect your architecture, you should evaluate other options. You can read the GPU cloud provider comparison to see how different service models stack up. Startups should also review the GPU clouds for startups guide to choose a matching financial model.
Paperspace alternatives
For users looking to compare providers, RunPod and Lambda Labs are two of the most relevant alternatives:
- RunPod: A popular choice for developers looking for a mix of managed environments and serverless workers. The RunPod review highlights its container-based Pods and activity-driven Serverless endpoints, which do not require monthly subscription tiers to access high-end GPUs.
- Lambda Labs: Known for its straightforward self-serve virtual machine rentals. The Lambda Labs review explains its simple hourly pricing model with no subscription fees, offering on-demand Linux instances that appeal to teams wanting direct root access without subscription overhead.
When comparing these alternatives, ensure you keep your requirements constant. Evaluate the combined cost of compute, storage, and egress fees over a typical billing cycle rather than comparing hourly GPU rates in isolation.
Pros and cons
The summary below reflects Paperspace's documented product model, pricing guides, and official limitations, rather than measured service performance.[source][source][source]
Pros
- Gradient provides managed Jupyter notebooks and containerized workflows that simplify machine learning setups.[source]
- Core provides traditional Linux and Windows virtual machines with full root access for custom deployments.[source]
- The platform does not charge fees for data ingress or egress, eliminating bandwidth cost volatility.[source]
- Block storage fees are capped monthly, preventing runaway storage bills during long projects.[source]
Cons
- High-end GPU instances are gated behind monthly subscription tiers, increasing the fixed cost for short runs.[source]
- Non-compute resources like block storage and public IPs continue to bill on stopped VMs until destroyed.[source]
- GPU Picks has not measured VM startup speed, network latency, or physical GPU availability.
Methodology and sources
This review of Paperspace is based on official pricing documentation, subscription guides, service level agreements, and trust platform disclosures published by DigitalOcean and Paperspace.[source][source][source][source] GPU Picks did not provision virtual machines, deploy containerized workflows, measure network speeds, or submit support tickets to verify response times.
We encourage readers to check our GPU Picks research methodology to understand how we maintain our truth boundaries. Always verify the current rates, instance availability, and terms on the official Paperspace console before deploying resources, as provider terms and pricing can change.
Frequently asked questions
Is a subscription required to use Paperspace?
A subscription is not strictly required for basic resources, as Paperspace offers a Free tier at $0/mo.[source] However, higher-end GPU instances (such as NVIDIA A100 or H100) and advanced features are gated behind the Pro ($8/mo) and Growth ($39/mo) subscription plans.[source] You must factor in these subscription costs when estimating your budget.
Does Paperspace charge for data transfer?
No. Paperspace has a zero-egress network policy, meaning they do not charge fees for network ingress or egress to the public internet or within a VPC.[source] This makes the platform cost-effective for workflows that involve transferring large datasets or serving high-bandwidth models.
Do stopped virtual machines on Paperspace cost money?
Yes. When a virtual machine is shut down or stopped, compute billing halts, but you continue to pay for non-compute resources such as block storage and public IPs until they are destroyed.[source] For example, a 50 GB block storage volume will continue to bill at $0.0074/hr up to its $5.00/mo cap even if the VM is stopped.[source]
What is the difference between Gradient and Core?
Gradient is a managed platform for Jupyter notebooks, workflows, and endpoints where the infrastructure is abstracted and managed by Paperspace.[source] Core provides traditional virtual machines with full root access, allowing you to configure the operating system, network, and software stack yourself.[source]
Does Paperspace support Windows virtual machines?
Yes. Paperspace Core supports Windows-based virtual machines in addition to Linux instances, providing desktop access for rendering, development, and testing.[source] Users should check the console for specific Windows template availability and licensing details.
What happens if my storage exceeds the subscription allowance?
For Gradient users, any storage usage that exceeds the base allowance included with your subscription plan is billed at a flat rate of $0.29/GB/month.[source] You should monitor your active workspaces and delete unused datasets or models to avoid these extra monthly fees.