NVIDIA Expands Sovereign AI Cloud Infrastructure Across Europe with OpenShell Reference Architecture for Secure AI Agents

Image: NVIDIA · Source

NVIDIA and its European cloud partners are scaling sovereign AI infrastructure using liquid-cooled AI factories and the OpenShell Reference Architecture, enabling secure, efficient, and customizable AI models closer to regional data for safer AI agent deployments.

NVIDIA is advancing the AI cloud ecosystem in Europe and neighboring regions by partnering with leading cloud providers to deploy scalable, secure, and efficient AI infrastructure tailored for the continent’s unique data sovereignty and industrial expertise needs. This regional network supporting AI developers emphasizes local compute capacity, customizable open models, and full-stack security.

One of the key initiatives is the expansion of NVIDIA’s DSX AI Factory Platform, which enables near-zero water consumption for cooling and up to 30% power savings through technologies like DSX MaxLPS and DSX Flex. For example, Nscale demonstrated a 49% increase in inference throughput and 37% more GPU capacity within the same power budget using NVIDIA DSX on GB300 NVL72 systems. These AI factories utilize closed-loop liquid cooling integrated with dry coolers, significantly reducing environmental impact compared to traditional setups. Partners such as Crusoe, Aleria, Cassava Technologies, and Radiant are adopting these innovations across Europe, the Nordics, Africa, and Central Asia.

The infrastructure supports running open and customizable AI models, enabling training, fine-tuning, and inference at scale. Noteworthy is Nebius’s adoption of Groq 3 LPX chips integrated with NVIDIA’s Spectrum-X networking to accelerate token generation in latency-sensitive applications, along with their Physical AI Workbench for robotics and retail agent blueprints. Baseten’s specialization in code-review routing with NVIDIA Nemotron open models illustrates optimization of accuracy and inference cost savings. Further expansion includes Mistral’s forthcoming NVIDIA Vera Rubin infrastructure to train frontier multimodal models like Mistral Large 4, showcasing Europe’s competitive position in AI model development.

At the forefront of AI agent security, NVIDIA introduced the OpenShell Reference Architecture to ensure full-stack governance, containment, and real-time monitoring of AI agents on enterprise and edge systems. Combining hardware such as Vera Rubin for inference, Vera CPUs for agent control, and BlueField-4 DPUs running NVIDIA Sentry for in-silicon policy enforcement, the architecture enables clear operational boundaries for AI agents. The design integrates with leading operating systems and orchestration platforms, and is supported by key industry players including Anthropic, AWS, Microsoft, Google Cloud, and cybersecurity firms.

Enterprise adoption spans sectors like banking—where BBVA and Santander explore OpenShell for secure agent governance—to software platforms at SAP and ServiceNow, with latter’s personal AI agents leveraging OpenShell for runtime action control and rapid access revocation. Major system manufacturers Dell Technologies, Cisco, and HPE, alongside cloud providers such as CoreWeave and Oracle Cloud, are deploying OpenShell-ready infrastructure.

Together, these efforts demonstrate NVIDIA’s commitment to empowering the European AI ecosystem with sovereign, sustainable, and secure AI infrastructure. By providing developers and enterprises with regionally controlled, open, and performance-optimized platforms alongside stringent security frameworks, NVIDIA and its partners are fostering safer and more efficient AI innovation and deployment across Europe and beyond.