GPU Cloud Provider · Unknown
RunPod
RunPod offers on-demand GPU clusters optimized for AI, ML, large language models (LLMs), and high-performance computing (HPC) workloads. They provide instant deployment, on-demand scalability, and flexible billing measured by the second, without minimum commitments or contracts.
GPUs
10
Founded
Unknown
Countries
6
Data Centers
8
Team Size
51-200
GPU Marketplace
AMD Instinct MI250X MI250XOn-Demand
AMD Instinct MI250 MI250On-Demand
AMD Instinct MI200 MI200On-Demand
NVIDIA A100 80GB SXMOn-Demand
NVIDIA A100 80GB PCIeOn-Demand
NVIDIA H100 PCIeOn-Demand
NVIDIA H100 PCIeOn-Demand
NVIDIA H100 NVL1On-Demand
NVIDIA H100 NVLOn-Demand
Company Profile
Company TypeMarketplace/Aggregator
Provider TypeMarketplace
Legal EntityRunPod, Inc.
FundingSeries A/B/C/D
Total Raised$20M
Team Size51-200
Intel Capital
Infrastructure
GPU FleetAMD Instinct MI250, NVIDIA H100 SXM, NVIDIA H100 PCIe, NVIDIA A100 SXM 80GB, NVIDIA A100 PCIe 80GB, NVIDIA A40, NVIDIA RTX 4090, NVIDIA RTX 3090, NVIDIA RTX A6000, NVIDIA L40S, NVIDIA L40, NVIDIA RTX 4000 Ada, NVIDIA A30
Network FabricInfiniBand, RoCE v2
Connectivity1,600–3,200 Gbps
StorageNetwork Storage with shared filesystems
Data Center TierSecure Cloud: Tier 3 data centers; Community Cloud: peer-hosted, varies
Bare MetalYes, via Secure Cloud (data center-grade) and Community Cloud (peer-hosted) tiers
AvailabilityGA
StartupResearchHobbyistEnterprise
Compute & Deployment
On-DemandYes
Spot / InterruptibleYes (Spot pods available at significant savings, interruptible by host)
Reserved InstancesYes (Savings Plans and longer-term commitments available)
Bare MetalNo
VM-BasedNo
Container-BasedYes (Docker)
KubernetesNo
Serverless GPUYes (RunPod Serverless for inference endpoints)
Spin-Up TimeUnder 2 minutes
TerraformYes (community provider)
GPU Hardware
Latest GenH100 SXM, H100 PCIe, H200, MI300X, MI250, L40S, RTX 4090
Legacy SupportA100 SXM, A100 PCIe, A40, A10G, V100, RTX 3090, RTX 3080
Multi-GPU NodesYes (up to 8x per node)
Max GPUs/Node8
NVLinkYes (NVLink on SXM nodes; H100 SXM supports NVLink 4.0)
InfiniBandYes (available on H100 SXM cluster configurations)
PCIe vs SXMBoth PCIe and SXM
Pricing Model
Per HourYes (primary billing unit)
Per MinuteYes (per-minute billing supported)
SubscriptionNo
Reserved DiscountNo
Spot DiscountUp to ~80% off on-demand (Spot pods available at significantly reduced rates)
Public PricingYes
Hidden FeesNone disclosed; storage and egress billed separately
Pay-as-you-goYes
Credit SystemYes (prepaid credits required to use platform)
Performance & Scaling
Multi-Node TrainingYes (distributed training supported via NCCL, limited orchestration tooling)
Elastic ScalingManual only
Auto ScalingInference only (Serverless auto-scaling for inference workloads)
InfiniBandNo (Ethernet only; InfiniBand not advertised for MI250 pods)
NVSwitchNo (AMD MI250 uses Infinity Fabric, not NVSwitch)
Perf IsolationPartial (dedicated GPU allocation per pod, but shared host infrastructure)
Noisy NeighborPartial (GPU dedicated per pod, CPU/network resources may be shared)
Developer Experience
OnboardingDeploy in under 5 minutes via web UI; instant account creation with credit card
FrameworksAll major AI frameworks compatible via Docker
SDK LanguagesPython
CLI ToolingFull CLI (runpodctl) with SSH tunneling, file sync, and pod management
JupyterNative JupyterLab integration via pod templates
TemplatesPyTorch, TensorFlow, Stable Diffusion, LLM Fine-tuning, AUTOMATIC1111, ComfyUI, Text Generation WebUI, vLLM, FastChat
Model MarketplaceBuilt-in Serverless Endpoints with prebuilt model workers; supports HuggingFace model IDs
DocumentationComprehensive docs with tutorials, API reference, and community guides
API FeaturesSelf-service provisioning through intuitive console
Security & Compliance
SecuritySOC2 Type II compliance
ComplianceSOC 2 Type II
Backed by Intel CapitalWidely used by AI/ML open-source community and indie researchersActive Discord community with 10,000+ membersTens of thousands of registered developersFeatured in major AI developer communities and tutorials
Data Center Locations
Coverage
CountriesUnited States, Norway, Poland, Romania, Sweden, Canada
CitiesDallas TX, Chicago IL, San Jose CA, Oslo, Warsaw, Bucharest, Stockholm, Montreal
Multi-Region FailoverNo (manual pod migration only)
North AmericaEuropeAsia-Pacific
Compliance Regions
EU Data ResidencyPartial (Norway, Poland, Romania, Sweden — no formal GDPR certification published)
US Gov CloudNo
India RegionNo
Datacenter Locations
Key Strengths
Among the lowest GPU pricing in the market including for H100s
Serverless GPU endpoints with auto-scaling to zero
Massive template library for AI/ML workloads out of the box
Dual-tier model: Secure Cloud for reliability, Community Cloud for cost savings
Per-second billing minimizes waste for short jobs
Known Limitations
Community Cloud GPUs can be less reliable and are subject to host availability
No Windows support; Linux containers only
Limited enterprise SLAs and compliance certifications compared to hyperscalers
Network bandwidth can vary significantly on Community Cloud nodes
No native managed Kubernetes or MLOps platform (basic orchestration only)
SOC 2 certification status not prominently published
Additional Information
Support Options
Not specified
Community
Active Discord (10,000+ members), active subreddit presence, GitHub repositories for SDKs and tooling
Core Proposition
Distributed GPU marketplace offering on-demand and spot GPU instances at competitive prices with fast provisioning, targeting AI/ML workloads from hobbyists to enterprises.
Payment Methods
Credit CardCrypto
Last updated March 2026. Information subject to change.