Akamai Technologies, Inc. announced that it is collaborating with NVIDIA Corporation to bring an advanced security architecture into AI factories. The two companies announced an expansion of their security collaboration to bring Akamai Guardicore Segmentation to the NVIDIA Vera BlueField-4 STX storage architecture, powered by the NVIDIA DOCA software platform. The collaboration is designed to layer Zero Trust architecture into the AI factory itself, protecting the data, context memory, and autonomous agents that increasingly run modern enterprises.

This novel security integration will enable AI factory operators to enforce workload-aware segmentation, monitor agent behavior, and contain threats at the infrastructure layer, operating at the speed of accelerated computing without taxing the GPU, CPU, or storage cycles that AI workloads depend on. The expanded Akamai?NVIDIA integration, which builds on the architecture agreement that two companies introduced last February, is designed to remove that trade-off. Akamai Guardicore Segmentation provides the intelligence layer, continuously mapping how workloads, applications, and data interact across hybrid environments that include data centers, cloud infrastructure, Kubernetes clusters, and edge systems. Policies are defined by workload identity, application context, and runtime behavior, not by static network addresses.

Visibility extends across the full lifecycle of AI workloads, surfacing abnormal patterns and unauthorized access to sensitive data. NVIDIA Vera BlueField-4 STX, programmable through NVIDIA DOCA, provides the threat detection and enforcement layers in silicon. Security policies are applied in the data path at line speed, inside the infrastructure fabric rather than on the host.

Enforcement moves closer to the workload itself, so it won?t trip up the GPUs, CPUs, and storage processors that AI factories depend on. Together, the two layers establish identity-based Zero Trust as a property of the infrastructure rather than an additional product. The combined solution operates on the principle that intelligence must precede enforcement: Akamai Guardicore Segmentation continuously maps communication relationships across data centers, cloud, Kubernetes, and edge systems. Its agentless architecture observes AI workloads, including training pipelines, inference services, data ingestion systems, and orchestration platforms, without interfering with them.

Workload identity, application context, and runtime behavior are used to define explicit communication policies. A preprocessing node may access a dataset and a training service but nothing beyond that scope. Research environments are explicitly separated from production inference.

Pods can scale and services can evolve without weakening the policy boundary. NVIDIA DOCA applies those policies in BlueField-4 silicon, in the data path, at line speed. Security functions, including segmentation, telemetry, anomaly detection, and isolation of compromised systems, run inside the infrastructure fabric rather than on the host.

When a workload is compromised, the blast radius is limited to a small, identified segment of the environment. The rest of the AI factory continues operating uninterrupted. Akamai Guardicore Segmentation integrated with NVIDIA BlueField and NVIDIA DOCA is expected to be available in the second half of 2026 for implementing workload-aware segmentation in AI factories.

Akamai's integration with NVIDIA Vera BlueField-4 STX is expected to be available on storage and infrastructure partner platforms in the first half of 2027.