NVIDIA introduced new security capabilities for its NVIDIA BlueField-4 platform, positioning the new Vera BlueField-4 STX as a secure storage processing architecture for emerging agentic AI workloads. Announced at NVIDIA GTC Taipei, the platform combines accelerated storage processing with embedded security services designed to monitor and control how autonomous AI agents access enterprise data, context memory, and file-based systems.
NVIDIA said agentic AI systems—where software agents retrieve information, reason over context, and take action autonomously—are creating new infrastructure requirements around both performance and security. Vera BlueField-4 STX extends NVIDIA’s AI-native storage architecture by moving security enforcement directly into the DPU silicon. The company said this allows enterprises to inspect and govern interactions between agents, storage, and memory inline, without interrupting data movement. NVIDIA claims the platform can deliver runtime threat detection up to 1,000x faster than current agentless runtime approaches while enforcing file and network access policies at speeds up to 800Gbps.
The announcement expands the software footprint of NVIDIA DOCA with three key security components tailored for AI storage environments. DOCA Vault manages workload-level file access permissions; DOCA Argus provides visibility into AI agent behavior and workload activity; and DOCA Flow isolates traffic and protects sensitive data in multi-tenant AI environments. NVIDIA said the ecosystem around Vera BlueField-4 STX already includes major storage vendors, cybersecurity providers, OEMs, and systems integrators preparing platforms expected to ship in the second half of 2026.
- Security features added to Vera BlueField-4 STX
- DOCA Vault — zero-trust file access controls for AI workloads
- DOCA Argus — visibility into agent behavior and AI runtime activity
- DOCA Flow — network traffic isolation and policy enforcement for multi-tenant AI environments
- Performance claims
- Runtime threat detection up to 1,000x faster than agentless runtime security approaches
- Network and file policy enforcement up to 800Gbps
- Storage ecosystem partners
- Dell Technologies
- Hewlett Packard Enterprise
- IBM
- NetApp
- VAST Data
- WEKA
- DDN
- MinIO
- Security ecosystem partners
- Cisco
- CrowdStrike
- Palo Alto Networks
- Fortinet
- Check Point Software Technologies
- Zscaler
“Agentic AI turns enterprise data into a living, real-time system — and that system must be protected where data moves, where context is stored and where agents act,” said Jensen Huang, founder and CEO of NVIDIA. “With Vera BlueField-4 STX, NVIDIA and its ecosystem are building secure-by-design storage infrastructure that enforces trust in silicon at the speed of AI.”
🌐 Analysis: NVIDIA is extending the role of the DPU beyond networking and storage acceleration into runtime security enforcement for AI infrastructure. As enterprises move toward retrieval-augmented generation, autonomous agents, and long-context reasoning, the security boundary increasingly shifts closer to data and memory. BlueField-4 STX reflects that transition by treating storage not just as capacity, but as an active policy enforcement layer inside the AI data path.
The announcement also deepens NVIDIA’s vertical integration strategy across AI infrastructure—from GPUs and NVLink fabrics to networking, storage, and now embedded security. Competitors including AMD, Intel, and hyperscale cloud operators are also moving to secure AI infrastructure closer to the compute and memory layer, but NVIDIA is positioning BlueField and DOCA as a unified control plane spanning networking, storage, and security.