GPU Cloud Provider · San Francisco, CA, USA
Lambda
Lambda offers cloud-based GPU clusters and superclusters optimized for AI and HPC workloads, featuring high-density, liquid-cooled NVIDIA GPUs. Their offerings focus on flexibility, performance, and ease of use for ML researchers and teams needing scalable compute resources.
GPUs
1
Founded
Data Not Available
Countries
1
Data Centers
4
Team Size
201-1000
GPU Marketplace
NVIDIA HGX Rubin NVL8On-Demand
Company Profile
Company TypeScale-up
Provider TypeCloud Provider
FoundedData Not Available
HeadquartersSan Francisco, CA, USA
Legal EntityLambda, Inc.
FundingSeries C
Total Raised$562M
Team Size201-1000
NVIDIAGradient VenturesRazerUS Innovative Technology FundMercato Partners
Infrastructure
GPU FleetNVIDIA H100 SXM 80GB, NVIDIA H100 PCIe 80GB, NVIDIA A100 SXM 80GB, NVIDIA A100 PCIe 40GB, NVIDIA A10, NVIDIA V100, NVIDIA RTX A6000, NVIDIA 3090
Network FabricInfiniBand,Ethernet
Connectivity400Gbps, NVLink, and up to 3200 Gb/s node-to-node compute between worker nodes
StorageNVMe, Shared file system
Data Center TierTier 3 equivalent colocation facilities
Bare MetalYes — bare metal GPU instances available for select configurations including H100 clusters
AvailabilityGA
EnterpriseResearchStartupGovernment
Compute & Deployment
On-DemandYes
Spot / InterruptibleNo
Reserved InstancesYes (1-year and 3-year reserved instances available)
Bare MetalYes (1-click clusters and bare metal GPU servers available)
VM-BasedYes
Container-BasedYes (Docker)
KubernetesYes (managed K8s via Lambda Cloud clusters)
Serverless GPUNo
Spin-Up Time1-3 minutes
TerraformYes (community provider)
GPU Hardware
Latest GenH100 SXM, H100 PCIe, H200 SXM, B200
Legacy SupportA100, A10, V100, RTX 6000 Ada
Multi-GPU NodesYes (up to 8x per node)
Max GPUs/Node8
NVLinkYes (NVLink 4.0 on SXM nodes)
InfiniBandYes (HDR 200Gbps on cluster configurations)
PCIe vs SXMBoth PCIe and SXM
HGX PlatformYes (HGX H100 8-GPU)
Pricing Model
Per HourYes (primary billing unit)
Per MinuteNo
SubscriptionYes (reserved instances with 1-year and 3-year terms)
Reserved DiscountUp to ~45% off with 1-year or 3-year reserved contracts
Spot DiscountNo spot pricing
Public PricingYes
Hidden FeesNone disclosed
Egress ChargesFree
Pay-as-you-goYes
Credit SystemNo
Performance & Scaling
Multi-Node TrainingYes (multi-node supported via NCCL and MPI across on-demand and reserved clusters)
Elastic ScalingNo (manual provisioning, no dynamic node add/remove)
Auto ScalingNo
InfiniBandYes (InfiniBand on select H100 SXM5 and A100 SXM4 clusters)
NVSwitchYes (on SXM nodes, H100 SXM5 and A100 SXM4)
Perf IsolationYes (dedicated bare metal instances, no hypervisor)
Noisy NeighborYes (bare metal, no sharing with other tenants)
Developer Experience
OnboardingDeploy in under 5 minutes via web UI or API; instant account approval for standard tiers
FrameworksPyTorch, TensorFlow, CUDA, cuDNN
SDK LanguagesPython
CLI ToolingLambda Cloud CLI available for instance management, SSH key management, and file operations
JupyterNative JupyterHub integration available on cloud instances
TemplatesPyTorch, TensorFlow, CUDA, Stable Diffusion, LLM Fine-tuning
Model MarketplaceNone — infrastructure-focused, no built-in model marketplace
DocumentationComprehensive docs with tutorials, API reference, and quickstart guides for common ML frameworks
Security & Compliance
Security
Backed by NVIDIASeries C funded ($562M total)In operation since 2012 — one of the longest-running GPU cloud providersSOC 2 Type II compliantSupplies GPU hardware to major AI labs and enterprisesUsed by Microsoft and other enterprise customers
Data Center Locations
Coverage
CountriesUnited States
CitiesAustin TX, San Jose CA, Washington DC, Salt Lake City UT
Multi-Region FailoverNo
North AmericaEuropeAsia-Pacific
Compliance Regions
EU Data ResidencyNo EU presence
US Gov CloudNo
India RegionNo
Datacenter Locations
Key Strengths
Competitive H100 pricing — among the lowest list prices for H100 SXM clusters
Also sells on-premise GPU hardware (workstations and servers), creating a unique hybrid offering
Research and developer-friendly with simple, transparent pricing
NVIDIA-backed with strong GPU supply relationships
Long-standing ML/AI focus since 2012 with deep domain expertise
Known Limitations
Limited geographic presence — primarily US-based regions
No spot or preemptible instances for cost savings
No Windows GPU instances
Smaller ecosystem and fewer managed services compared to hyperscalers (no managed databases, object storage tiers, etc.)
H100 cluster availability can be limited during high demand periods
No multi-cloud or hybrid cloud orchestration tooling
Additional Information
Support Options
["24/7 Dedicated support"]
Community
Active presence on Twitter/X and Discord; developer community via documentation forums; GitHub repositories for tooling
Core Proposition
GPU cloud purpose-built for AI/ML workloads with on-demand and reserved access to high-end NVIDIA GPUs at competitive pricing with no egress fees.
Notable Customers
Microsoft
Tenstorrent
Scale AI
Various AI research labs and startups
Payment Methods
Credit CardWire TransferACH
Last updated March 2026. Information subject to change.