← Back to stories Cybersecurity experts in hoodies analyzing encrypted data on computer screens in an indoor setting.
Photo by Tima Miroshnichenko on Pexels
凤凰科技 2026-03-18

Huang’s “Little Lobster” arrives: NVIDIA’s NemoClaw goes live — focused on security and ease of use

NemoClaw aims to make OpenClaw enterprise-ready

NVIDIA’s CEO Jensen Huang used his GTC 2026 keynote to roll out NemoClaw, a software toolbox designed to bolt enterprise-grade safety and privacy onto the open-source OpenClaw agent runtime. It has been reported that NemoClaw installs an “OpenShell” layer, sandboxes agents, and enforces policy-based security, network and privacy controls so companies can run agentic AI without exposing sensitive data or losing governance. Will this be enough to turn a viral open runtime into a safe production platform? That is the implicit bet.

Technical architecture and platform play

Huang framed NemoClaw inside a larger systems push: the Vera Rubin platform, seven new chips (including a new CPU and a Groq-derived LPU), and a “disaggregated inference” model that splits prefill/attention on Rubin-class GPUs from ultra-low‑latency decoding on LPUs. It has been reported that OpenClaw’s uptake has been explosive — NVIDIA claimed downloads surpassed Linux’s 30-year tally and that the project has over 250,000 GitHub stars — but Huang warned that OpenClaw lacked enterprise security until now. Nvidia says NemoClaw will lower deployment friction and, by expanding always-on agent workloads, boost long‑term demand for GPUs, CPUs and LPUs alike.

Partners, forecasts and geopolitical backdrop

NVIDIA listed partners across cloud and systems makers — including Alibaba (阿里巴巴), CoreWeave, Meta, Oracle (甲骨文), Dell, HP, Lenovo and AMD (超威) — and flagged automotive and robotics ties with BYD (比亚迪), Geely (吉利), Isuzu and others for L4 driving and industrial AI. It has been reported that Huang forecast Blackwell and Rubin families could generate roughly $1 trillion in revenue by the end of 2027, up from prior guidance. All of this comes against tightening U.S. export controls and a contested global semiconductor landscape; platform control over runtimes and runtimes’ security could become as strategically important as raw silicon in the coming years.

AIResearchSpaceRobotics
View original source →