Accelerate telco AI with Red Hat OpenShift and NVIDIA DPUs

Telecommunications (telco) service providers globally are rapidly integrating AI to optimize performance, automate operations, and unlock new revenue streams. As AI becomes a core part of the telco business, networks must support intelligent, data-intensive workloads alongside traditional services, while maintaining performance, security, and operational simplicity across highly distributed environments. However, legacy infrastructures were not built for this—forcing trade-offs that slow AI adoption, increase costs, and limit the speed of innovation.

Red Hat and NVIDIA deliver an accelerated application platform for telco AI as part of the Red Hat® AI Factory with NVIDIA. The solution combines the cloud-native foundation of Red Hat OpenShift® with NVIDIA BlueField data processing units (DPUs) to create an integrated infrastructure in which compute, networking, security, and storage operate as a single, cohesive system.

Red Hat OpenShift provides the unified cloud-native platform layer, delivering consistent orchestration, lifecycle management, and policy enforcement across core, regional, and edge deployments. NVIDIA BlueField serves as the dedicated infrastructure processor for AI factories—offloading, accelerating, and isolating core datacenter services from the host computer processing unit (CPU), and directing available compute to AI workloads. Together, they offer service providers the scalable, security-centered, and operationally efficient foundation needed to deploy AI and bring new services to market at the speed the business demands.

Service provider cloud-native AI workloads generate massive amounts of communication data. Traditional software-defined networking (SDN) architectures running on host CPUs introduce latency and bandwidth ceilings that cannot satisfy the demands of modern AI workloads. As inference pipelines grow more complex, the gap between what general-purpose CPUs can deliver and what service providers need from AI continues to widen. That turns infrastructure into a bottleneck and slows the pace at which new services can be brought to market.

The joint solution addresses this bandwidth and latency bottleneck by moving packet processing off the host CPU and into dedicated hardware on NVIDIA BlueField DPUs. This is a fundamental shift in how network processing is handled, not an incremental improvement on existing architecture. By handling packet processing in purpose-built hardware, NVIDIA BlueField eliminates the throughput and latency constraints that have long limited traditional software-defined networks. Service providers achieve up to 800 Gb/s line speed throughput, giving their AI workloads the capacity to handle massive traffic volumes at the speeds modern AI pipelines require.

Critically, NVIDIA BlueField acceleration does not come at the cost of programmability. Service providers no longer face a trade-off between network flexibility and raw performance—routing policies and traffic priorities can still be adapted through software while the underlying acceleration is handled in hardware. Network behavior remains fully configurable; network throughput no longer relies on CPU processing power.

Red Hat OpenShift completes the picture with fully orchestrated networking that enforces consistent policy automatically across distributed deployments, from centralized AI factories to far-edge nodes. As deployments grow and new sites come online, configuration propagates without manual intervention, giving service providers the consistency and control they need to run telco AI reliably across every tier of their infrastructure.

On traditional infrastructure, a significant amount of CPU resources and system power is consumed by networking, security enforcement, and data movement tasks. These are necessary functions that compete directly with AI workloads for the same compute resources. As AI workload density increases, this overhead compounds, escalating infrastructure costs while limiting the capacity available for revenue-generating services.

NVIDIA BlueField acts as a dedicated processor for datacenter infrastructure, handling networking, security, and data movement tasks in hardware. By taking on these functions from the host CPU, NVIDIA BlueField makes the cores that were previously consumed by infrastructure processing available for AI inference, analytics, and other high-value workloads. Red Hat OpenShift services run directly on NVIDIA BlueField so the host CPU is never burdened with platform overhead—effectively augmenting the compute budget by reclaiming cycles previously consumed by infrastructure services.

Power efficiency gains follow naturally from the same principle. Purpose-built acceleration delivers infrastructure services at a fraction of the energy cost of general-purpose CPUs—a meaningful advantage at the node level that becomes even more substantial across a large distributed deployment. Service providers can run more AI and other workloads on the same hardware footprint, reduce energy costs per unit of compute, and scale capacity without a proportional increase in rack space or power draw.

For service providers managing tens of thousands of nodes across core and edge sites, these efficiency gains compound across the entire infrastructure footprint, reshaping the economics of AI deployment and creating room to scale without a corresponding rise in operational expenditure.

Service provider AI deployments need security enforcement at scale across highly distributed environments, but must avoid accompanying performance penalties that degrade service quality. The joint platform embeds security directly into the infrastructure layer through a defense-in-depth approach that operates across both hardware and software.

NVIDIA BlueField enforces strict workload separation using hardware roots of trust, helping prevent unauthorized lateral movement across the platform. NVIDIA DOCA Argus provides accelerated, real-time threat detection purpose-built for AI and other service provider workloads. It operates fully within the DPU, freeing up host CPU resources for higher-value tasks. Red Hat OpenShift and NVIDIA BlueField together provide a programmable security foundation that integrates with a broad ecosystem of carrier-class cybersecurity solutions for monitoring, policy enforcement, and compliance.

That ecosystem integration is a critical differentiator for service providers navigating complex regulatory and threat environments. Customers will not need to rebuild or replace their existing point-in-time security tooling., Red Hat OpenShift and NVIDIA BlueField deliver programmable interfaces that allow those existing investments in security operations, threat intelligence, and compliance tooling to extend naturally into AI and cloud-native infrastructure. Service providers can enforce consistent policy across bare-metal, virtualized, and containerized workloads, spanning core datacenters and distributed edge sites without maintaining separate suites of security tools for each environment.

Modern service provider AI architectures are inherently distributed. Inference workloads execute across centralized AI factories, regional datacenters, and far-edge sites, and each distinct environment requires consistent networking policies, security controls, and operational tooling. Achieving this consistency at scale has historically required significant manual configuration and custom integration work. As the number of edge sites grows and AI workloads become more sensitive to latency, the operational burden of managing this distributed fabric becomes a meaningful constraint on how quickly service providers can deploy and evolve their services.

The joint solution extends Red Hat OpenShift networking across core and edge environments through NVIDIA BlueField-powered distributed routing. NVIDIA BlueField advances Red Hat OpenShift networking by unifying L3 underlay and SDN overlay through NVIDIA DOCA distributed routing, delivering a single, consistent network fabric that spans multiple clusters and sites. Border Gateway Protocol (BGP)-based multipath routing provides high availability and supports unified coexistence with existing network services, eliminating the need for disruptive infrastructure changes during migration.

This architecture is designed to scale without a proportional increase in operational complexity. Because the network fabric is software defined and hardware accelerated at every node, adding new edge sites does not require manual reconfiguration of upstream infrastructure. Routing, policy, and security controls propagate automatically, allowing service providers to expand their distributed AI footprint at the pace the business demands rather than the pace their operations teams can manually support.

As service providers scale AI deployments and distributed infrastructure, the operational and cost implications become a primary constraint on the pace of transformation. Every new edge site, every additional AI workload, and every expansion of the service footprint carries with it a corresponding increase in energy consumption, operational overhead, and infrastructure complexity—unless the underlying platform can accommodate that growth efficiently. The joint Red Hat and NVIDIA solution addresses cost efficiency across 3 dimensions: energy, operations, and service agility.

Purpose-built acceleration shifts infrastructure processing to NVIDIA BlueField, reducing overall power consumption and improving performance per watt compared to general-purpose infrastructure. As AI factory deployments scale across hundreds or thousands of nodes, the cumulative energy savings become a substantial line item. Service providers are no longer paying the power cost of running networking, security, and data movement on general-purpose CPUs—those are tasks that NVIDIA BlueField handles at a fraction of the energy cost with purpose-built silicon. At datacenter scale, this translates directly into lower operating expenditure and a more sustainable infrastructure growth model.

On the operations side, Red Hat OpenShift simplifies networking and lifecycle management across distributed deployments. Automated policy enforcement, consistent configuration propagation, and centralized cluster management reduce the manual overhead that has historically made large-scale service provider infrastructure expensive to run. Operations teams spend less time on repetitive configuration tasks and more time on higher-value work. As the deployment footprint grows, the platform scales without a proportional increase in operational costs.

Service agility is the 3rd dimension—and is increasingly the one that is most competitively significant. Software-defined infrastructure helps reduce the time it takes service providers to roll out new services and adapt to changing network and application demands without disruptive infrastructure changes. New AI capabilities that once required lengthy integration and deployment cycles can be brought to market more rapidly, shortening time-to-revenue and increasing the return on AI investment. For service providers competing to differentiate on service quality and innovation speed, this agility is not only an operational benefit—it is a strategic advantage.

Experience the benefits of the Red Hat and NVIDIA integrated platform

Red Hat and NVIDIA unite to deliver the latest innovations, sharing an AI vision fueled by open source, cloud-native technologies. Read more about this partnership and see the list of joint offers.

Read the checklist “Accelerate telco AI transformation with NVIDIA and Red Hat,” which outlines the 5 key benefits¹ of deploying telco AI workloads on this, unified, accelerated platform from Red Hat and NVIDIA.