Paperspace

Provides instant access to NVIDIA GPU instances (H100, and other GPU tiers) with per-second billing granularity, allowing users to spin up compute resources without long-term commitments or reserved instance purchases. The platform abstracts infrastructure provisioning through a tiered instance model (Basic, Mid-range, High-end) and claims 70% cost savings vs major cloud providers through optimized pricing and no idle-time waste.

Provides managed Jupyter notebook instances (Gradient Notebooks) running on GPU hardware with automatic environment setup, persistent storage, and collaborative features. Users launch notebooks directly from the Paperspace dashboard without local setup, and notebooks persist across sessions with versioning and lifecycle management built-in. The environment supports standard Python ML libraries (PyTorch, TensorFlow, scikit-learn) with pre-installed CUDA/cuDNN stacks.

Paperspace uses OAuth-based authentication exclusively, allowing users to sign up and log in via Google or GitHub accounts without creating separate credentials. The platform delegates identity management to OAuth providers, eliminating password management and enabling single sign-on for users with existing Google/GitHub accounts. No email/password authentication option is documented, creating a dependency on OAuth provider availability.

Paperspace was acquired by DigitalOcean and is being integrated into DigitalOcean's broader cloud platform, with Paperspace maintaining its branding while leveraging DigitalOcean's infrastructure and services. The acquisition enables cross-product integration (e.g., Paperspace notebooks accessing DigitalOcean Spaces for storage, App Platform for deployment) and unified billing. The integration timeline and specific feature roadmap are not documented.

Gradient Workflows enable users to define and schedule batch training jobs that run on GPU instances with automatic resource provisioning, job queuing, and lifecycle management. Jobs are submitted via the dashboard or API (specifics not documented) and execute training scripts in isolated containers with configurable GPU allocation. The platform handles instance startup, script execution, and cleanup, abstracting away manual VM management for training workloads.

Gradient Deployments convert trained models into REST API endpoints accessible via HTTP, with automatic model versioning, lifecycle management, and scaling. Users upload a trained model artifact (format not specified) and Paperspace provisions inference infrastructure, exposes a public/private API endpoint, and manages model versions. The platform claims 'scalable' endpoints but specific auto-scaling triggers, concurrency limits, and latency SLAs are not documented.

Paperspace supports team workspaces with role-based access control (RBAC) for notebooks, training jobs, and deployments. Users invite team members with specific roles (permissions not detailed) and share resources within a team namespace. The platform provides 'Insights' feature for visibility into team utilization, permissions, and resource consumption, though specific metrics and dashboard capabilities are not documented.

Paperspace supports deploying trained models and running inference on multiple cloud providers (Azure, AWS, GCP) and on-premise hardware (DGX, custom servers), enabling users to avoid vendor lock-in and optimize for cost/latency across regions. The platform abstracts deployment targets through a unified interface, though specific implementation details (API format, supported instance types per cloud, failover mechanisms) are not documented.

Paperspace provides persistent storage for notebooks, training artifacts, and model files with tiered capacity based on subscription tier (Free: 5GB, Pro: 15GB, Growth: 50GB) and overage charges ($0.29/GB beyond included allocation). Storage persists across notebook sessions and training jobs, enabling users to accumulate datasets and model checkpoints without re-uploading. Storage is accessible via filesystem mount within notebooks/jobs and via web dashboard.

Paperspace automatically versions trained models and notebooks with tagging and lifecycle management capabilities, allowing users to track model lineage, compare versions, and promote models through development/staging/production stages. The platform stores version metadata (creation date, creator, tags, performance metrics) and enables rollback to previous versions without manual artifact management. Specific versioning API and metadata schema are not documented.

Paperspace offers a free tier ($0/month) with limited resources (5GB storage, 12-hour auto-shutdown, public projects only) enabling users to try the platform without payment. The free tier includes access to GPU instances (type not specified) with per-second billing, allowing users to run small training jobs and notebooks at no subscription cost. Auto-shutdown after 12 hours of inactivity prevents runaway costs from forgotten instances.

Paperspace offers four subscription tiers (Free: $0, Pro: $8/month, Growth: $39/month, Team: $0–$12/user/month) with increasing storage, auto-shutdown configurability, and team collaboration features. Each tier unlocks additional capabilities (private projects, longer auto-shutdown, team management, expert support) while maintaining per-second GPU billing independent of subscription tier. Tier selection determines platform features but not compute costs, creating clear upgrade path for growing teams.