Advancing AI Infrastructure for Agentic AI with NVIDIA DOCA In-Silicon Security

The AI era is driving a new class of infrastructure: AI factories that transform data into intelligence for autonomous AI agents operating at unprecedented scale. Powered by accelerated computing, AI factories enable enterprises to train, fine-tune, and deploy AI with greater speed and efficiency. 

This new class of infrastructure also introduces a fundamentally new attack surface spanning infrastructure, software supply chains, models, data, and autonomous agents with increasing authority to act. As agentic AI adoption accelerates, adversaries are increasingly targeting both AI infrastructure and the applications it powers, creating new risks to the confidentiality, integrity, and availability of mission-critical systems. 

Traditional security architectures were not designed for the scale, complexity, and performance demands of AI factories. Securing this new infrastructure requires security that is distributed, full-stack, and accelerated. 

This post explains how NVIDIA BlueField data processing units (DPUs) and NVIDIA DOCA secure AI factories through runtime detection, data access control, and accelerated network enforcement to protect AI infrastructure, workloads, agents, and data at AI scale.

How does in-silicon security change the traditional security model?

Purpose-built for AI infrastructure, NVIDIA BlueField DPUs combine high-performance networking, programmable compute, hardware acceleration, and advanced security capabilities into a single platform embedded into every AI factory compute node. Unlike traditional security approaches that rely on host system software, BlueField establishes a hardware-enforced, in-silicon, and workload-independent security layer.

Operating within its own trusted execution domain, BlueField isolates infrastructure and security services from the host system. Monitoring, policy enforcement, and telemetry operate even if the host or workloads become compromised. Because security functions remain isolated from the system they protect, attackers cannot tamper with or bypass BlueField-enforced infrastructure security policies.

This architecture fundamentally changes the traditional security model. Conventional endpoint protection shares the same trust boundaries and system resources as the environments it secures, making security software vulnerable to tampering, evasion, or disablement when a host is compromised. By offloading security processing to the BlueField silicon, it delivers resilient, full-stack protection without consuming host computing resources or competing with AI workloads—preserving peak infrastructure efficiency and AI performance.