Oracle Cloud Infrastructure is building a massive multi-vendor AI cloud for the next decade
Oracle Cloud Infrastructure is preparing for a major expansion of its AI cloud platform, bringing together accelerators and processors from NVIDIA, AMD, Intel, Google, Cerebras, SambaNova, and other technology providers. The goal is clear: give enterprises, developers, and AI companies access to powerful, flexible infrastructure for training, inference, agentic AI, and large-scale cloud workloads.
Announced during Oracle AI World 2026, the next phase of OCI focuses on building a more open and scalable AI infrastructure. Oracle says its cloud platform now includes more than 200 cloud services, with large-scale AI becoming one of its most important growth areas. The company also recently added 850MW of data center capacity, showing how aggressively it is expanding to meet demand from enterprise applications and frontier-scale AI systems.
At the center of this strategy is Oracle Acceleron, a high-performance networking fabric designed to connect huge clusters of compute resources across different accelerator platforms. As AI workloads become larger and more complex, cloud providers need more than just fast chips. They also need extremely fast networking, resilient data paths, power-efficient data centers, and the ability to scale across hundreds of thousands of processors and GPUs.
Oracle is positioning OCI as a cloud platform that can support all of that.
NVIDIA Vera Rubin systems are coming to OCI at multi-gigawatt scale
One of the biggest announcements is Oracle’s expanded partnership with NVIDIA. OCI plans to deploy Vera Rubin NVL72 systems at multi-gigawatt scale, targeting massive AI training, inference, and agentic AI applications. General availability is beginning now, giving customers access to one of NVIDIA’s most advanced AI platforms.
OCI will also deploy NVIDIA Vera CPU servers and NVL8 HGX systems, further expanding the range of NVIDIA-based infrastructure available on Oracle Cloud.
The NVIDIA-powered OCI Superclusters will combine NVL72 rack-scale systems, Oracle Acceleron networking, NVIDIA Spectrum-X Ethernet, Spectrum-6 switches, and BlueField-4 DPUs. These clusters are designed for extremely large AI workloads, including training and serving the largest modern models.
Each NVL72 rack connects 72 GPUs through NVIDIA NVLink scale-up networking. Oracle’s multiplanar network architecture is built to scale across hundreds of thousands of GPUs while maintaining resilience and performance. This is especially important for AI training jobs, where interruptions, bottlenecks, or network failures can create major delays and costs.
AMD EPYC CPUs and Instinct GPUs will power new OCI compute and AI systems
AMD is also playing a major role in Oracle’s AI infrastructure roadmap. OCI will use 6th Gen AMD EPYC processors to power its upcoming Compute E7 instances. These instances are aimed at demanding enterprise and AI workloads, including agent orchestration, concurrent agent sandboxes, retrieval services, in-memory caching, web services, API layers, and enterprise applications.
The E7 Acceleron Standard virtual machines will offer flexible compute and memory configurations, allowing customers to choose the right balance of performance and cost for their workloads.
Oracle also plans to deploy AMD’s Helios AI rack architecture, which will use the latest AMD Instinct MI455X GPUs. This continues AMD’s growing presence in AI data centers, where its Instinct GPU family is becoming an important option for high-performance AI training and inference.
OCI already uses a wide range of NVIDIA and AMD accelerators today. On the NVIDIA side, Oracle has deployed platforms such as P100, V100, A100, H100, H200, GB200, and GB300. On the AMD side, OCI has already supported Instinct MI300X and MI355X deployments. The new announcements show that Oracle is continuing to expand its multi-vendor AI strategy rather than relying on a single chip supplier.
Intel Diamond Rapids Xeon processors are planned for OCI X14
Oracle and Intel are also extending their collaboration through a multiyear plan to expand Intel Xeon-based computing on OCI. The next major step is the planned OCI X14 platform, which is expected to use Intel’s upcoming Diamond Rapids server processor, also referred to as Xeon 7.
Diamond Rapids is designed to boost AI and enterprise performance through a combination of general-purpose compute and dedicated AI acceleration. The platform is expected to support Intel AMX with FP8, FP16, and BF16 formats, along with Intel APX, or Advanced Performance Extensions.
This makes the upcoming OCI X14 platform important for enterprises that need strong performance for traditional applications, AI inference, and agentic AI workloads. While GPUs and specialized accelerators dominate large AI training, CPUs remain essential for many business-critical workloads, data processing tasks, application tiers, and inference use cases.
Google TPUs are coming to Oracle Cloud Infrastructure in 2027
In a major shift for the cloud AI market, Oracle also announced plans to bring Google TPUs to OCI. This would make Oracle the first hyperscale cloud provider outside of Google to deploy TPUs.
The rollout is expected to begin in 2027 and scale to gigawatt levels over multiple years. This gives OCI another major accelerator option for customers working on large-scale AI training and serving.
The latest TPU systems use on-die Inter-Chip Interconnect routing combined with Optical Circuit Switching. This allows the control plane to detect hardware or link degradation and dynamically route optical paths around faults in real time, without restarting jobs. For massive AI training runs, that kind of resilience can be extremely valuable.
Oracle says these systems can sustain more than 97% goodput across thousands of chips during large-scale training runs. The latest TPUs also bring more on-chip SRAM, higher HBM capacity, and a new low-diameter interconnect topology that reduces network hops by more than 50%. These improvements are designed to deliver up to 80% better performance per dollar for low-latency serving.
Cerebras and SambaNova will strengthen OCI’s AI inference options
Oracle is also adding more specialized AI inference systems to OCI through Cerebras and SambaNova. The first Cerebras CS3 deployments are planned for 2027, giving OCI customers another high-speed option for inference workloads that require predictable, ultra-low-latency performance.
As AI agents become more capable and more widely used, inference performance is becoming just as important as training performance. Businesses deploying AI assistants, autonomous workflows, real-time analytics, coding agents, customer support tools, and search systems need responses that are fast, reliable, and cost-efficient.
Oracle’s strategy is to offer customers more choice. Instead of forcing every workload onto the same hardware, OCI aims to provide different compute options for different needs. Some workloads may run best on NVIDIA GPUs, others on AMD GPUs, Intel Xeon CPUs, Google TPUs, Cerebras systems, or SambaNova platforms.
That flexibility could become a major advantage as AI adoption expands across industries.
Oracle Acceleron is the networking backbone of OCI’s AI cloud
A key part of Oracle’s AI infrastructure strategy is Acceleron, the company’s high-performance fabric for connecting compute, storage, and networking resources at massive scale.
AMD technology will also contribute to Acceleron through Pensando data processing units, or DPUs, supporting AMD CPUs and GPUs on OCI. Acceleron combines dedicated network fabrics, converged NICs, zero-trust routing, and multiple network paths to improve performance, security, and resiliency.
Oracle is now extending Acceleron with a High-Performance Fabric for front-end RDMA connectivity and more NVMe block-storage attachment options across bare metal and virtual machines. The company is also introducing the Oracle Collective Communications Library, which is designed to help distributed AI workloads use healthier backend paths.
Front-end multiplanar networking adds automatic path selection, failover, and congestion control. In practical terms, this means data can move faster, workloads can keep running even when a path fails, and customers can get more useful work from their deployed compute infrastructure.
This is especially important for large AI clusters. At massive scale, failures are inevitable. The real challenge is keeping jobs running smoothly when hardware, links, or routes degrade. Oracle’s networking approach is designed to reduce disruption and maintain high efficiency across huge AI deployments.
OCI is becoming one of the most open multi-vendor AI clouds
Oracle’s next-generation AI infrastructure is built around choice, scale, and specialization. The company is combining NVIDIA Vera Rubin systems, AMD EPYC CPUs and Instinct GPUs, Intel Diamond Rapids Xeon processors, Google TPUs, Cerebras CS3 systems, SambaNova inference technology, and its own Acceleron fabric into a broad cloud AI portfolio.
For customers, this means OCI could support everything from enterprise applications and CPU-heavy workloads to frontier model training, large-scale inference, retrieval systems, agentic AI, and ultra-low-latency AI services.
The wider message is that the AI cloud market is moving beyond a single type of accelerator. Training, inference, enterprise AI, agent orchestration, and real-time applications all have different performance and cost requirements. Oracle wants OCI to be a platform where customers can match each workload with the right compute architecture.
With new data center capacity, multi-gigawatt AI deployments, advanced networking, and partnerships across nearly every major AI hardware category, Oracle Cloud Infrastructure is positioning itself as a major force in the next wave of cloud-based artificial intelligence.
