AI Factories

The latest push to secure AI factories is aimed at a problem that traditional tools have struggled to solve: how to stop fast-moving LLM-based attacks without slowing the system down. Akamai and NVIDIA are addressing that pressure with a new integration designed for autonomous AI workloads. The approach combines Akamai Guardicore Segmentation with NVIDIA Vera BlueField-4 STX to apply Zero Trust principles in real time across data, context memory, and AI compute processes. In practice, the goal is to keep security close to the workload while preserving the speed that high-performance AI environments require. Security that does not choke AI performance One of the biggest concerns in protecting AI factories is the risk that security layers will consume GPU and CPU cycles. Conventional defenses can slow operations, which is a serious trade-off in environments built for speed and automation. Akamai and NVIDIA are trying to avoid that bottleneck by moving security policies directly into the data path through NVIDIA DOCA software. This allows workload-based segmentation to run alongside AI processing rather than sitting outside it as a separate burden. From visibility to isolation The protection model starts with visibility across data center, cloud, and edge environments. Akamai Guardicore maps communication relationships without interrupting live operations, giving teams a clearer view of how workloads interact. From there, policies are defined using workload identity and application context instead of static network addresses. That makes access control more precise and better suited to AI factories, where workloads can shift quickly and connections are often highly dynamic. Protection enforced in silicon Once policies are set, NVIDIA DOCA enforces them directly in the BlueField-4 silicon at full speed. That design keeps protection inside the infrastructure layer and reduces the load on the main system. If one workload is compromised, the impact can be contained immediately within a small recognized segment. Other parts of the AI factory can continue operating normally, which helps stop a local incident from spreading across the wider computing environment. Why segmentation matters more now AI factories depend on many tightly connected components, so a weak point in one area can create a pathway to others. Without strong segmentation, an attack on one service can ripple through the rest of the operation and disrupt business processes. Zero Trust segmentation helps shrink that attack surface from the start. For autonomous AI systems that demand speed, consistency, and operational resilience, the model is designed to fit the way the infrastructure actually works. Availability timeline for the integration Akamai Guardicore Segmentation integrated with NVIDIA BlueField and DOCA is scheduled to be available in the second half of 2026. Full integration with NVIDIA Vera BlueField-4 STX on partner infrastructure and storage platforms is expected in the first half of 2027. For companies treating AI factories as critical assets, the combination of visibility, segmentation, and real-time enforcement points to a different security model. The emphasis is not only on blocking attacks, but also on keeping AI workloads moving quickly without opening broader gaps across the system. Source: www.medcom.id

Zero Trust Moves Into the Heart of AI Factories, Cutting LLM Attacks Without Slowing Speed