NVIDIA AI Updates: September 2, 2026

1. NVIDIA and CrowdStrike Launch SafeMind at Fal.Con 2026

NVIDIA. Jensen Huang joined CrowdStrike CEO George Kurtz at Fal.Con 2026 to announce CrowdStrike SafeMind, an agentic cybersecurity system built by the CrowdStrike Cyber team on NVIDIA’s stack. The pitch is symmetry: attacks are already automated, so detection and response have to run as autonomous agents rather than analyst-paced workflows. The partnership extends NVIDIA’s push to make security one of the first production markets for long-horizon agents. Source

2. NVIDIA Publishes a Reference Build for Adaptive Security Agents on Nemotron

NVIDIA. The developer blog walks through building an adaptive agentic cybersecurity system on Nemotron models, covering how agents coordinate work and pursue objectives across long horizons rather than single-shot triage. It is a companion to the CrowdStrike announcement, aimed at security teams building in house instead of buying a platform. The post focuses on the orchestration and evaluation layers, which is where most agentic security prototypes stall. Source

3. NVIDIA Publishes GPU Sizing Guidance for Inference TCO

NVIDIA. The guide covers how to size GPU fleets for inference workloads without overprovisioning, connecting throughput and latency targets to total cost of ownership rather than peak benchmark numbers. It addresses a recurring problem for teams moving from prototype to production, where capacity is chosen from worst-case traffic assumptions and idles most of the time. Source