AI infrastructure

Run the most demanding AI workloads faster, including generative AI training and inference, scientific computing, and recommendation engines, anywhere in our distributed cloud. Use Oracle Cloud Infrastructure (OCI) Supercluster to scale up to 131,072 GPUs.

Get the update on agentic AI
Delivering AI Anywhere with Oracle and NVIDIA (2:11)

NVIDIA AI Enterprise on the OCI Console

Soon, you can accelerate AI development anywhere with one click from the OCI Console.

Why run on OCI AI infrastructure?

Performance and value

Boost AI training with OCI’s unique GPU bare metal instances and ultrafast RDMA cluster networking that reduce latency to as little as 2.5 microseconds. Get up to 220% better pricing on GPU VMs than with other cloud providers.

HPC storage

Leverage OCI File Storage with high performance mount targets (HPMTs) and Lustre (coming soon) for terabytes per second of throughput. Use up to 61.44 TB of NVMe storage, the highest in the industry for GPU instances.

Sovereign AI

Oracle’s distributed cloud enables you to deploy AI infrastructure anywhere to help meet performance, security, and AI sovereignty requirements. Learn how Oracle and NVIDIA deliver sovereign AI anywhere.

Scalability of OCI Supercluster image, description below The image shows boxes that represent compute and storage, connected by lines for cluster networking. On the far left, there are four boxes of compute and two for storage for the smallest cluster with 16,000 NVIDIA H100 GPUs. To its right there are 8 boxes of compute and 4 boxes of storage for 32,000 NVIDIA A100 GPUs in a cluster. Next, there are 16 boxes of compute and 8 boxes of storage for 64,000 NVIDIA H200 GPUs. Finally, to the far right, there are 32 boxes of compute and 16 boxes of storage for 128,000 NVIDIA Blackwell and Grace Blackwell GPUs. This shows scalability of OCI Supercluster increasing by 8X from the smallest 16,000 GPU configuration on the far left to the largest 128,000 GPU configuration on the far right.

OCI Supercluster with NVIDIA Blackwell and Hopper GPUs

Up to 131,072 GPUs, 8X more scalability
Network fabric innovations enable OCI Supercluster to scale up to 131,072 NVIDIA B200 GPUs, more than 100,000 Blackwell GPUs in NVIDIA Grace Blackwell Superchips, and 65,536 NVIDIA H200 GPUs.

OCI AI infrastructure for all your needs

Whether you’re looking to perform inferencing or fine-tuning or train large scale-out models for generative AI, OCI offers industry-leading bare metal and virtual machine GPU cluster options powered by an ultrahigh-bandwidth network and high performance storage to fit your AI needs.

AI infrastructure products diagram, description below The image shows several products for AI Infrastructure starting on the bottom left with the smallest configurations then progressively increasing to medium scale and large scale configurations. The smallest configurations are with just 1 GPUs in a virtual machine and the largest configurations are for more than 100,000 GPUs in RDMA clusters.

Try-before-you-buy program for OCI Compute AMD with Instinct MI300X GPUs.

Explore OCI Supercluster for large-scale AI training

Massive scale-out clusters with NVIDIA Blackwell and Hopper


Supercharged compute
• Bare metal instances without any hypervisor overhead
• Accelerated by NVIDIA Blackwell (GB200 NVL72, HGX B200),
Hopper (H200, H100), and previous-generation GPUs
• Option to use AMD MI300X GPUs
• Data processing unit (DPU) for built-in hardware acceleration

Massive capacity and high-throughput storage
• Local storage: up to 61.44 TB of NVMe SSD capacity
• File storage: fully managed Lustre service (coming soon) and HPMT
with up to 80 Gb/sec of throughput
• Block storage: balanced, higher performance, and ultrahigh performance volumes with a performance SLA
• Object storage: distinct storage class tiers, bucket replication, and high capacity limits

Ultrafast networking
• Custom-designed RDMA over Converged Ethernet protocol (RoCE v2)
• 2.5 to 9.1 microseconds of latency for cluster networking
• Up to 3,200 Gb/sec of cluster network bandwidth
• Up to 200 Gb/sec of front-end network bandwidth

Compute for OCI Supercluster

OCI bare metal instances powered by NVIDIA GB200 NVL72, NVIDIA B200, NVIDIA H200, AMD MI300X, NVIDIA L40S, NVIDIA H100, and NVIDIA A100 GPUs let you run large AI models for use cases that include deep learning, conversational AI, and generative AI.

With OCI Supercluster, you can scale up to more than 100,000 GB200 Superchips, 131,072 B200 GPUs, 65,536 H200 GPUs, 32,768 A100 GPUs, 16,384 H100 GPUs, 16,384 MI300X GPUs, and 3,840 L40S GPUs per cluster.

Computer for Supercluster image
Enlarge+
The diagram shows GPU cluster nodes powered by NVIDIA A100 GPUs and networking with less than two-microsecond latency.

Networking for OCI Supercluster

High-speed RDMA cluster networking powered by NVIDIA ConnectX network interface cards with RDMA over Converged Ethernet version 2 lets you create large clusters of GPU instances with the same ultralow-latency networking and application scalability you expect on-premises.

You don’t pay extra for RDMA capability, block storage, or network bandwidth, and the first 10 TB of egress is free.

Networking for Supercluster
Enlarge+
The diagram shows A Supercluster RDMA network with eight NVIDIA A100 GPUs per node connected through a full-duplex network fabric with a total of 1.6 Tb/sec internode bandwidth.

Storage for OCI Supercluster

Through OCI Supercluster, customers can access local, block, object, and file storage for petabyte-scale computing. Among major cloud providers, OCI offers the highest capacity of high performance local NVMe storage for more frequent checkpointing during training runs, resulting in faster recovery from failures.

For massive data sets, OCI offers high performance file storage with Lustre and mount targets. HPC file systems, including BeeGFS, GlusterFS, and WEKA, can be used for AI training at scale without compromising performance.

Storage for Supercluster
Enlarge+
The diagram shows A GPU cluster connected to block storage and an Exabyte-scale file storage service through RDMA network fabric.

Zettascale OCI Superclusters

Watch OCI's top architects reveal how cluster networks power scalable generative AI. From a few GPUs to zettascale OCI Superclusters with more than 131,000 NVIDIA Blackwell GPUs, cluster networks deliver high speed, low latency, and a resilient network for your AI journey.

March 26, 2025

Announcing New AI Infrastructure Capabilities with NVIDIA Blackwell for Public, On-Premises, and Service Provider Clouds

Henry Jerez, Senior Vice President, Oracle
Christian Linacre, Vice President, Oracle

At Oracle Cloud Infrastructure (OCI), we’re committed to providing customers with the most advanced cloud infrastructure for AI workloads, wherever they need it.

Read the complete post

Featured blogs

Typical AI infrastructure use cases

Train AI models on OCI bare metal instances powered by GPUs, RDMA cluster networking, and OCI Data Science.


Deep learning training and inferencing diagram, description below
Train AI models on OCI bare metal instances powered by GPUs, RDMA cluster networking, and OCI Data Science.

Protecting the billions of financial transactions that happen every day requires enhanced AI tools that can analyze large amounts of historical customer data. AI models running on OCI Compute powered by NVIDIA GPUs along with model management tools such as OCI Data Science and other open source models help financial institutions mitigate fraud.


Fraud detection augmented by AI diagram, description below
AI models running on OCI Compute powered by NVIDIA GPUs along with model management tools such as OCI Data Science and other open source models help financial institutions mitigate fraud.

AI is often used to analyze various types of medical images (such as X-rays and MRIs) in a hospital. Trained models can help prioritize cases that need immediate review by a radiologist and report conclusive results on others.


AI-based medical image analysis diagram, description below
Trained models running on OCI Compute powered by GPUs can help analyze medical images and provide immediate conclusive results or prioritize images for further review.

Drug discovery is a time consuming and expensive process that can take many years and cost millions of dollars. By leveraging AI infrastructure and analytics, researchers can accelerate drug discovery. Additionally, OCI Compute powered by NVIDIA GPUs along with AI workflow management tools such as BioNeMo enables customers to curate and preprocess their data.


Using AI to accelerate drug discovery, description below
Leveraging AI Infrastructure and analytics, researchers can accelerate drug discovery and curate and preprocess their data.

AI infrastructure customer successes

Explore more customer stories

Get started with OCI AI infrastructure

Access AI subject matter experts

Get help with building your next AI solution or deploying your workload on OCI AI infrastructure.

  • They can answer questions such as

    • How do I get started with Oracle Cloud?
    • What kinds of AI workloads can I run on OCI?
    • What types of AI services does OCI offer?

See how to apply AI today

Enter a new era of productivity with generative AI solutions for your business. Learn how Oracle helps customers leverage AI embedded across the full technology stack.

  • What can you achieve with Oracle AI?

    • Fine-tune LLMs in OCI
    • Automate invoice processing
    • Build a chatbot with RAG
    • Summarize web content with generative AI
    • And so much more!

Additional resources

Learn more about RDMA cluster networking, GPU instances, bare metal servers, and more.

See how much you can save with OCI

Oracle Cloud pricing is simple, with consistent low pricing worldwide, supporting a wide range of use cases. To estimate your low rate, check out the cost estimator and configure the services to suit your needs.

Experience the difference

  • 1/4 the outbound bandwidth costs
  • 3X the compute price-performance
  • Same low price in every region
  • Low pricing without long term commitments