AI × AEC · CURATED INSIGHT
Regional AI Clouds Expand NVIDIA Hardware and Deploy Agent Safety Frameworks
European cloud providers are deploying NVIDIA compute clusters with power-optimized scheduling, while adopting the new OpenShell architecture to restrict autonomous agent activities at both chip and software levels.
Service providers across Europe and adjacent regions are scaling AI facilities to handle training and low-latency inference. Recent updates include cluster-level energy coordination by operators such as Nscale, as well as the rollout of the OpenShell reference architecture unveiled at GTC Berlin, which integrates runtime policies with specialized processors to confine AI agent permissions.
Key takeaways
- In cluster trials using GB300 NVL72 hardware, Nscale reported that NVIDIA DSX MaxLPS raised inference output by approximately 50% and supported 37% more GPUs under an identical electrical capacity limit.
- Announced at GTC Berlin, NVIDIA's OpenShell architecture pairs runtime software with Vera CPUs and BlueField-4 DPUs executing Sentry to enforce operational boundaries on AI agent behavior.
Source：NVIDIA Blog Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.