Advancing AI Infrastructure for Agentic AI with NVIDIA DOCA In-Silicon Security | NVIDIA Technical Blog - NVIDIA Developer

As the AI era advances, a new class of infrastructure—AI factories—is emerging, transforming data into intelligence for autonomous AI agents operating at unprecedented scale. However, this evolution i... Read original →

NVIDIA’s in-silicon security layer via BlueField and DOCA marks a paradigm shift—from software-based defenses to hardware-enforced isolation in AI infrastructure. This forces the entire stack, from 3nm EUV foundries (e.g., TSMC) to server OEMs, to redesign trusted execution environments amid surging autonomy in agentic AI data flows. Regulatory-wise, upcoming EU/US AI Acts demand full model lifecycle auditability; DOCA Argus’ runtime monitoring fits perfectly but raises compliance costs for non-NVIDIA ecosystems. AMD may accelerate DPU startup acquisitions, while Intel could counter with IPU+TDX integration. Within 18 months, DPUs with hardware-rooted data provenance will become mandatory in AI data centers—security is no longer bolted on but built in. Supply chains in Taiwan, China and Hong Kong, China must recognize: any AI infrastructure reliant on NVIDIA’s security stack risks embedded remote kill-switches under geopolitical stress.