GPU Cloud Provider · Redwood Shores, CA, USA
OCI (Oracle Cloud Infrastructure)
Oracle Cloud Infrastructure (OCI) Zettascale10 is positioned as one of the most advanced cloud computing platforms for high scale AI compute, featuring integration with OpenAI for massive AI superclusters. It leverages Oracle's Acceleron RoCE networking and NVIDIA AI infrastructure to handle multi-gigawatt AI workloads offering up to 16 zettaFLOPS of peak performance. OCI Zettascale10 boasts high GPU-to-GPU connectivity and reliability, aiming at reducing the power per unit of performance making it suitable for extremely large and complex AI deployments.
GPUs
1
Founded
2018
Countries
22
Data Centers
29
Uptime SLA
99.9%
Team Size
10,000+
GPU Marketplace
NVIDIA HGX Rubin NVL8On-Demand
Company Profile
Company TypeHyperscaler
Provider TypeHyperscaler
Founded2018
HeadquartersRedwood Shores, CA, USA
Legal EntityOracle Corporation
FundingPublic (NYSE: ORCL)
Total RaisedNot applicable (publicly traded)
Team Size10,000+
Infrastructure
GPU FleetNVIDIA H100 SXM5 80GB, NVIDIA H100 NVL PCIe, NVIDIA A100 80GB, NVIDIA V100, NVIDIA L40S, AMD MI300X
Total GPU CapacityNot disclosed publicly; OCI has announced plans for large-scale H100 clusters exceeding 16,000+ GPUs per cluster
Network FabricOracle Acceleron RoCE, 400G/800G throughput
Connectivity400Gbps GPU to GPU interconnects across racks
StorageNVMe SSDs
Data Center TierTier 3+ equivalent; OCI-owned and operated data centers across 40+ regions
Bare MetalYes — bare metal GPU instances available (BM.GPU series)
AvailabilityGeneral Availability from the second half of the next calendar year
EnterpriseGovernmentResearchStartupFinancial Services
Compute & Deployment
On-DemandYes
Spot / InterruptibleYes (preemptible instances available at reduced cost)
Reserved InstancesYes (1-year and 3-year reserved capacity commitments)
Bare MetalYes (bare metal GPU instances including A100, H100 clusters)
VM-BasedYes (GPU-accelerated VM shapes available)
Container-BasedYes (Docker and container-based workloads supported)
KubernetesYes (managed K8s via Oracle Container Engine for Kubernetes - OKE)
Serverless GPUNo
Spin-Up Time2-5 minutes (standard instances); bare metal GPU clusters may take 10-20 minutes
TerraformYes (official provider on HashiCorp Terraform Registry)
GPU Hardware
Latest GenH100 SXM, H100 PCIe, A100
Legacy SupportV100, P100
Multi-GPU NodesYes (up to 8x per node)
Max GPUs/Node8
NVLinkYes (NVLink on SXM nodes)
InfiniBandYes (RDMA over RoCE and InfiniBand, up to 400Gbps on H100 clusters)
PCIe vs SXMBoth PCIe and SXM
HGX PlatformYes (HGX H100 8-GPU)
Pricing Model
Per HourYes (primary billing unit)
Per MinuteNo
SubscriptionYes (monthly/annual Universal Credits and Committed Use agreements)
Reserved DiscountUp to 33% off with 1-year Universal Credit commitment, up to 50% with 3-year
Spot DiscountNo spot pricing for GPU instances (Preemptible instances available for compute, limited GPU availability)
Public PricingYes
Hidden FeesIP address charges apply; LB and networking bandwidth charges billed separately; support plans additional cost
Egress Charges10TB/month free per account, then $0.0085/GB (varies by region); free within OCI regions
Pay-as-you-goYes
Credit SystemYes (Universal Credits prepaid with committed spend discounts)
Performance & Scaling
Multi-Node TrainingYes (up to 1000+ nodes with RDMA/RoCE and NCCL)
Max Cluster Size4096 GPUs (via HPC clusters with BM.GPU.H100.8 bare metal shapes)
Elastic ScalingYes (add/remove nodes dynamically via instance pools and HPC cluster networks)
Auto ScalingYes (policy-based auto-scaling via Instance Pool autoscaling)
InfiniBandYes (RDMA over RoCE v2, up to 100Gbps per node; select H100 clusters use NDR InfiniBand 400Gbps)
NVSwitchYes (on SXM5-based H100 bare metal shapes BM.GPU.H100.8)
SLA99.9%
Perf IsolationYes (dedicated bare metal instances, no hypervisor sharing)
Noisy NeighborYes (bare metal, no sharing with other tenants)
Developer Experience
OnboardingSelf-service account creation with free tier; enterprise onboarding supported by Oracle sales and solution architects; provisioning in minutes via Console, CLI, or Terraform
FrameworksNot specifically mentioned but suitable for large-scale AI deployments
SDK LanguagesPython, Java, Go, Ruby, TypeScript, JavaScript, .NET, C++
CLI ToolingFull OCI CLI with instance management, file operations, and automation support; Terraform provider available
JupyterVia OCI Data Science service with managed JupyterLab environments; also accessible via SSH on compute instances
TemplatesPyTorch Training, TensorFlow, LLM Fine-tuning, Stable Diffusion, NVIDIA NeMo, Triton Inference Server, RAPIDS
Model MarketplaceOCI AI Services and Oracle Marketplace offer pre-built AI models; integration with NVIDIA NGC catalog
DocumentationComprehensive docs with tutorials, API reference, architecture guides, and hands-on labs via Oracle LiveLabs
API FeaturesNot specified
Security & Compliance
SecurityNot detailed
ComplianceData and AI sovereignty controls
NYSE: ORCL — Fortune 500 publicly traded companySOC 1/2/3 Type II certifiedISO 27001, FedRAMP High authorizedNVIDIA DGX-Ready Cloud partnerCSA STAR certifiedHIPAA, PCI DSS, GDPR compliantUsed by 50+ of Fortune 100 companies
Data Center Locations
Coverage
CountriesUnited States, United Kingdom, Germany, Netherlands, France, Spain, Italy, Sweden, Switzerland, United Arab Emirates, Saudi Arabia, India, Japan, South Korea, Australia, Singapore, Brazil, Canada, South Africa, Israel, Mexico, Chile
CitiesAshburn VA, Phoenix AZ, San Jose CA, Chicago IL, London, Frankfurt, Amsterdam, Paris, Madrid, Milan, Stockholm, Zurich, Dubai, Jeddah, Mumbai, Hyderabad, Tokyo, Osaka, Seoul, Sydney, Melbourne, Singapore, São Paulo, Toronto, Montreal, Johannesburg, Tel Aviv, Mexico City, Santiago
Multi-Region FailoverYes (automatic and manual failover with region pairs)
Latency TiersUltra-low (<1ms intra-DC), Standard cloud latency across regions, FastConnect for dedicated low-latency connectivity
North AmericaEuropeAsia-PacificMiddle EastLatin AmericaAfrica
Compliance Regions
EU Data ResidencyYes (Frankfurt, Amsterdam, Paris, Madrid, Milan, Stockholm, Zurich)
US Gov CloudYes (FedRAMP authorized, dedicated US Government Cloud regions)
India RegionYes (Mumbai, Hyderabad)
Datacenter Locations
Key Strengths
Highly competitive price-performance vs AWS/Azure/GCP on compute
Bare metal GPU instances with full hardware access
OCI Supercluster — RDMA-connected GPU clusters up to tens of thousands of GPUs
Strong enterprise and database integration (Oracle DB, Autonomous DB)
Multicloud agreements with Microsoft Azure and Google Cloud for seamless connectivity
Known Limitations
Smaller global ecosystem and third-party tooling compared to AWS/Azure
Fewer managed AI/ML services than AWS SageMaker or Azure ML
GPU availability in some regions can be constrained without advance reservation
Less mature developer community and fewer native AI startup integrations
Console UX considered less polished than competitors by some users
Additional Information
Support Options
Not specified
Community
Oracle Cloud Community forums, Oracle Developer community portal, GitHub (oracle org), active presence at Oracle CloudWorld and developer events; no large independent Discord community
Green Energy
Oracle has committed to 100% renewable energy for all global operations by 2025; reports progress annually
PUE Rating
1.2 (estimated); OCI reports energy efficiency improvements but specific PUE not universally published
Core Proposition
Oracle Cloud Infrastructure offers high-performance GPU compute with bare metal access, ultra-low latency networking via RDMA cluster networking, and deep integration with Oracle Database and enterprise applications at competitive pricing.
Notable Customers
NVIDIA
Uber
Zoom
Vodafone
Telefonica
FedEx
Pearson
Rakuten
Payment Methods
Credit CardWire TransferOracle Universal CreditsOCI Marketplace (AWS Marketplace integration available for some services)
Last updated March 2026. Information subject to change.