Why Cloud Networking Needs a New Foundation
AI is entering a new phase. Training, inference, and agentic AI systems are becoming more capable, distributed, and central to how organizations operate.
These new workloads do not replace existing cloud-native, database, and enterprise applications. Those workloads will continue to run at scale and remain critical to customers. But AI adds new traffic patterns and performance expectations that challenge assumptions built into traditional cloud networking.
The result is an urgent need for a stronger network foundation—one that supports today’s diverse cloud traffic while preparing for the next generation of AI.
AMD and Oracle have partnered to meet this changing requirement by co-engineering AMD Pensando-based DPUs along with Oracle Acceleron SmartNICs to move networking, storage, security, and infrastructure functions closer to the data path and preserve host CPU capacity for customer applications.
AI introduces different network demands
Traditional cloud-native traffic is typically spread across a large and diverse set of clients, servers, and databases. That diversity helps distribute traffic across available network paths.
AI workloads can look very different. Training often moves large volumes of data among a relatively fixed set of graphics processing units (GPUs). These flows can be long-lived, bursty, and synchronized. Inference requires fast and consistent responses as AI models serve users and applications. Agentic AI systems add more distributed activity as agents autonomously plan, reason, iterate and coordinate across services, tools, and data sources.
These patterns can reduce flow diversity and make it harder to distribute traffic evenly across the network. A small number of congested paths can affect the performance of an entire training job, reduce utilization of expensive GPUs and impact user experience.
For this reason, as AI infrastructure scales in a multi-tenant environment, it needs more than raw bandwidth. It needs high throughput, predictable latency, strong isolation, efficient security, and the ability to evolve as workloads change.
Building a stronger foundation with AMD Pensando DPUs
AMD Pensando™ data processing units (DPUs) help power Oracle Acceleron SmartNICs for Oracle’s AMD compute and accelerator instances.
Oracle Acceleron SmartNICs combine host networking and cloud-control functions in a single architecture while maintaining strong isolation between customer and cloud-managed functions. This converged approach helps reduce hardware complexity and unnecessary data movement while improving operational efficiency.
For AI workloads, this is significant because infrastructure overhead can directly affect the resources available for training and inference. For traditional cloud workloads, it helps improve the efficiency of networking, storage, and security services without requiring customers to redesign their applications.
Accelerated Performance
AMD Pensando DPUs help accelerate networking, storage, and security functions in hardware. This reduces overhead in the data path and helps OCI deliver more consistent performance for demanding workloads on its AMD instances.
The same advantages support databases, cloud-native applications, and enterprise services. Database performance depends on efficient storage and network operations. Cloud-native applications rely on fast, secure communication among services. Enterprise workloads require strong isolation and encryption without consuming excessive host CPU capacity.
Oracle Acceleron SmartNIC architecture, powered by AMD Pensando DPUs, can deliver up to 2x improvement in network processing efficiency and increased storage performance1, depending on workload and configuration.
Enhanced Security Capabilities
As AI systems connect more applications, services, and data sources, security and isolation become even more important.
Oracle Acceleron SmartNICs use a security-first architecture that helps enforce isolation at the host level. The architecture supports OCI Zero Trust Packet Routing (ZPR), which applies access policies close to where traffic enters and leaves the host. This is important because unauthorized packets are flagged and contained early, preventing them from spreading to other parts of the network. It also enables near-zero-downtime security patching so that infrastructure remains protected without disrupting cloud services.
AI systems need to move and process data quickly but protect sensitive information and maintain clear boundaries between customers and services. This is achieved in the network powered by AMD Pensando-based Oracle Acceleron SmartNIC architecture.
Greater Programmability
AI workloads are evolving quickly. New models, training methods, inference patterns, and agentic applications will continue to create new requirements for the network.
AMD Pensando DPU technology gives OCI a programmable foundation for adapting to those changes. It enables OCI to evolve networking, storage, security, and observability services without adding unnecessary complexity to the host environment.
AMD Pensando-based Oracle Acceleron SmartNICs address these use cases and meet the challenge of extreme scale and flexibility across Acceleron-enabled OCI compute shapes.
Conclusion
AMD Pensando and Oracle Acceleron enable customers running AMD compute and accelerators on OCI to build a strong network foundation to meet evolving AI demands along with established cloud native enterprise workloads. The partnership between AMD and Oracle provides innovation and flexibility, enabling customers to future-proof their network designs.
Want to learn more? Dive deeper into Oracle Acceleron here, AMD Pensando DPUs here, and Oracle and AMD’s joint solution portfolio here.
