GPU Cloud Provider · Redmond, Washington, USA
Azure
Azure, Microsoft's cloud computing platform, recently introduced the NVIDIA GB300 NVL72 supercluster, marking a significant advancement in AI infrastructure. This new offering includes advanced GPU types and extensive connectivity options, designed to facilitate the training and deployment of frontier AI and high-performance computing workloads.
GPUs
2
Founded
2010
Countries
34
Data Centers
54
Uptime SLA
99.9%
Team Size
10,000+
GPU Marketplace
NVIDIA A100 80GB SXMOn-Demand
NVIDIA GB200 NVL4On-Demand
Company Profile
Company TypeHyperscaler
Provider TypeHyperscaler
Founded2010
HeadquartersRedmond, Washington, USA
Legal EntityMicrosoft Corporation
FundingPublic (MSFT)
Team Size10,000+
Infrastructure
GPU FleetNVIDIA H100 SXM 80GB, NVIDIA H100 NVL, NVIDIA A100 80GB SXM, NVIDIA A100 40GB, NVIDIA V100 32GB, NVIDIA V100 16GB, NVIDIA T4, NVIDIA A10, AMD MI300X
Network FabricInfiniBand, NVIDIA NVLink and NVSwitch
ConnectivityUp to 800 Gbps per GPU and 400 GB/s scale-out networking
StorageNVMe, Premium Blob Storage, High-throughput Block Storage
Data Center TierTier 3+ / Tier 4 equivalent; Microsoft-owned and operated global data centers with N+1 redundancy
Bare MetalYes, via Azure Bare Metal Infrastructure and ND-series VMs with direct GPU access
AvailabilityGA
EnterpriseGovernmentStartupResearchISV
Compute & Deployment
On-DemandYes
Spot / InterruptibleYes (Azure Spot VMs, up to 90% savings vs on-demand, eviction-based)
Reserved InstancesYes (1-year and 3-year reserved VM instances with significant discounts)
Bare MetalYes (Azure Dedicated Host for isolated physical servers)
VM-BasedYes (N-series, ND-series, NC-series GPU VMs)
Container-BasedYes (Docker via Azure Container Instances and AKS node pools)
KubernetesYes (managed K8s via Azure Kubernetes Service - AKS)
Serverless GPUNo (no native serverless GPU; Azure Container Apps lacks GPU support as of early 2026)
Spin-Up Time2-5 minutes (standard GPU VMs); longer for large clusters or quota-limited SKUs
TerraformYes (official HashiCorp-registry provider: hashicorp/azurerm)
GPU Hardware
Latest GenH100 SXM, H100 PCIe, ND H100 v5, MI300X
Legacy SupportA100, V100, T4, A10
Multi-GPU NodesYes (up to 8x per node)
Max GPUs/Node8
NVLinkYes (NVLink 4.0 on SXM nodes)
InfiniBandYes (NDR 400Gbps on ND H100 v5)
PCIe vs SXMBoth PCIe and SXM
HGX PlatformYes (HGX H100 8-GPU)
Liquid CoolingSelect SKUs
Pricing Model
Per HourYes (primary billing unit)
Per MinuteYes (per-second billing, minimum 1 minute)
SubscriptionYes (monthly/annual Reserved VM Instances)
Reserved DiscountUp to 72% off with 3-year commitment; up to 40% off with 1-year commitment
Spot DiscountUp to 90% off on-demand with Azure Spot VMs
Public PricingYes
Hidden FeesPublic IP address charges (~$0.004/hr), load balancer fees, premium storage I/O operations, outbound data transfer charges not always prominently displayed
Egress ChargesFirst 100GB/month free; tiered pricing after: ~$0.087/GB for next 10TB (varies by region)
Pay-as-you-goYes
Credit SystemYes (Azure credits via MSDN, startup programs, and enterprise agreements)
Performance & Scaling
Multi-Node TrainingYes (up to 1000+ nodes with NCCL and MPI via Azure CycleCloud and HPC clusters)
Max Cluster SizeNot disclosed (thousands of GPUs possible via CycleCloud and HPC deployments)
Elastic ScalingYes (add/remove nodes dynamically via Azure VMSS and CycleCloud autoscaling)
Auto ScalingYes (policy-based auto-scaling via Azure VMSS with custom metrics and Azure Monitor)
InfiniBandYes (HDR InfiniBand 200Gbps on HB/HC/ND series; NDR 400Gbps on NDv5/H100 clusters)
NVSwitchYes (on SXM5 nodes such as ND H100 v5 series)
SLA99.9%
Perf IsolationPartial (dedicated VM instances with SR-IOV; bare metal available via Azure Dedicated Host)
Noisy NeighborPartial (SR-IOV and dedicated host options reduce contention; full bare metal isolation via Azure Dedicated Host)
Developer Experience
OnboardingDeploy in minutes via Azure Portal, CLI, or ARM/Bicep templates; enterprise onboarding with account managers available
FrameworksTensorFlow, PyTorch, Jax
SDK LanguagesPython, JavaScript, TypeScript, Java, C#, .NET, Go, Ruby, PHP, C++
CLI ToolingFull Azure CLI (az) with VM management, SSH access, file sync, and Azure Developer CLI (azd) for full-stack deployments
JupyterNative via Azure Machine Learning Studio and Azure Notebooks integration
TemplatesLLM Fine-tuning, PyTorch Training, TensorFlow Training, Stable Diffusion, HPC workloads, Vision AI
Model MarketplaceAzure AI Model Catalog with HuggingFace, Meta Llama, Mistral, OpenAI, and other model integrations
DocumentationComprehensive docs with tutorials, API reference, quickstarts, architecture guides, and Microsoft Learn training paths
API FeaturesAzure CLI, Azure SDK, REST API
Security & Compliance
SecuritySOC2, ISO 27001 compliance, regular penetration testing
ComplianceISO 27001, SOC 2
Public company (NASDAQ: MSFT)SOC 1/2/3 Type II certifiedFedRAMP High authorizedISO 27001, 27017, 27018 certifiedHIPAA/HITECH compliantPCI DSS Level 1Exclusive strategic partner of OpenAIUsed by 95% of Fortune 500 companies
Data Center Locations
Coverage
CountriesUnited States, United Kingdom, Germany, Netherlands, France, Ireland, Sweden, Switzerland, Norway, Finland, Spain, Italy, Poland, Czech Republic, Austria, Denmark, Belgium, Canada, Brazil, Mexico, Chile, Australia, Japan, South Korea, India, Singapore, Hong Kong, Indonesia, Malaysia, UAE, Qatar, South Africa, Israel, New Zealand
CitiesBoydton VA, Chicago IL, San Antonio TX, Phoenix AZ, Des Moines IA, Seattle WA, Washington DC, New York NY, Atlanta GA, Dallas TX, Toronto, Quebec City, Sao Paulo, Rio de Janeiro, Santiago, Mexico City, London, Dublin, Amsterdam, Frankfurt, Paris, Stockholm, Zurich, Geneva, Oslo, Helsinki, Madrid, Milan, Warsaw, Prague, Vienna, Copenhagen, Brussels, Sydney, Melbourne, Tokyo, Osaka, Seoul, Busan, Mumbai, Pune, Chennai, Hyderabad, Singapore, Hong Kong, Jakarta, Kuala Lumpur, Dubai, Abu Dhabi, Doha, Johannesburg, Cape Town, Tel Aviv, Auckland
Multi-Region FailoverYes (automatic failover with Azure Traffic Manager and paired regions)
Latency TiersUltra-low (<1ms intra-DC via Availability Zones), Standard cloud latency across regions, ExpressRoute for dedicated low-latency connectivity
North AmericaEuropeAsia-PacificMiddle EastAfricaSouth America
Compliance Regions
EU Data ResidencyYes (Frankfurt, Amsterdam, Dublin, Paris, Stockholm, Helsinki, Oslo, Geneva, Zurich, Madrid, Milan, Warsaw, Vienna, Copenhagen) with EU Data Boundary guarantee
US Gov CloudYes (FedRAMP High authorized, Azure Government regions in Virginia and Texas, Azure Government Secret and Top Secret for classified workloads)
India RegionYes (Mumbai, Pune, Chennai, Hyderabad)
Datacenter Locations
Key Strengths
Deep OpenAI partnership with exclusive early access to GPT-4 and o-series models
Massive global footprint with 60+ regions and edge locations
Seamless integration with Microsoft 365, GitHub, and enterprise software ecosystem
Azure Arc enables consistent hybrid and multi-cloud GPU workload management
Strong enterprise compliance portfolio (FedRAMP, HIPAA, ISO 27001, SOC 2)
Known Limitations
H100 and high-end GPU availability can be constrained in certain regions
Pricing complexity with many SKU tiers and reservation options can be difficult to navigate
Cold-start latency and quota approval processes can slow large-scale GPU provisioning
Less developer-friendly UX compared to specialized GPU cloud startups
Spot VM eviction rates can be high during peak demand in popular regions
Additional Information
Support Options
["24/7 Support Desk","Community Forums","Dedicated Technical Account Manager"]
Community
Microsoft Tech Community forums, Azure Developer Discord, GitHub Discussions, Stack Overflow presence, Microsoft Q&A platform, and regional Azure user groups worldwide
Green Energy
Committed to 100% renewable energy by 2025; carbon negative by 2030; zero waste by 2030
PUE Rating
1.12 (global average reported)
Core Proposition
Deeply integrated Microsoft ecosystem (Azure AD, Office 365, GitHub, OpenAI partnership) with global hyperscale infrastructure and enterprise compliance coverage.
Notable Customers
OpenAI
GitHub
LinkedIn
Adobe
SAP
Walmart
Boeing
Payment Methods
Credit CardWire TransferInvoice/Purchase OrderAzure MarketplaceEnterprise AgreementCSP Partner
Last updated March 2026. Information subject to change.