NVIDIA Cloud Partners Expand Across Europe With Sovereign, Open and Safe AI
NVIDIA says its Cloud Partners across Europe, the Nordics, Africa and Central Asia are expanding regional compute for training and inference. At GTC Berlin the company also announced the NVIDIA OpenShell Reference Architecture, which is meant to set clear limits on what AI agents can access and do.
- NVIDIA Cloud Partners are expanding AI capacity across Europe, the Nordics, Africa and Central Asia.
- Nscale reports 37% more GPUs and nearly 50% higher aggregate inference throughput within the same power budget in GB300 NVL72 tests.
- CodeRabbit raised routing agreement from 75.8% to 80.7% and cut estimated inference costs by 50% using a Baseten-built Nemotron 3.5 Lightning model.
- Mistral Large 4 was trained on 3,800 NVIDIA Grace Blackwell GPUs in European data centers.
- The OpenShell Reference Architecture, announced at GTC Berlin, is backed by 19 cloud partners.
Sovereign capacity and power efficiency
According to NVIDIA, demand from AI-native companies is driving the European AI cloud ecosystem forward. Developers want compute close to their data, open models they can customize, and infrastructure that is secure by design. Regional AI clouds are meant to supply that foundation for model builders, enterprises and industries.
Power is becoming a constraint as AI factories grow. Nscale is using the NVIDIA DSX Platform to match power allocation to workload needs. In tests on NVIDIA GB300 NVL72 systems, Nscale demonstrated that NVIDIA DSX MaxLPS enabled 37% more GPUs and nearly 50% higher aggregate inference throughput within the same provisioned power budget. Nscale also plans to deploy NVIDIA Vera Rubin infrastructure for frontier-model training and production workloads, with expansion across five European markets.
Crusoe is developing Spark, liquid-cooled modular AI factories designed for the NVIDIA Vera Rubin platform, to bring capacity closer to users for low-latency inference. Crusoe Cloud serves Base44, Gradium, LTX and Wonderful, among others, across EMEA. Aleria, Cassava Technologies, Firebird, G42, Radiant, SWI Compute and Vultr are also building or expanding NVIDIA-powered infrastructure in the regions.
Open models from training to inference
The cloud partners are also adding capacity and services for building with open models and libraries, covering training, fine-tuning for specialized tasks and inference at scale, including at the edge.
- Nebius has adopted NVIDIA Groq 3 LPX and connects it to Vera Rubin NVL72 on Nebius Token Factory using Spectrum-X Ethernet. NVIDIA says this will bring faster token generation to latency-sensitive and agentic applications. Nebius also offers a Physical AI Workbench for robotics firms such as Rhoda AI, RoboForce and Sereact, and has launched three NVIDIA retail agent blueprints into early access.
- Baseten built a code-review routing model with Nemotron 3.5 Lightning, used by CodeRabbit. Through supervised fine-tuning and reinforcement learning, CodeRabbit improved routing agreement from 75.8% to 80.7% while reducing estimated inference costs by 50% for the evaluated workload.
- Fireworks and Era4 are partnering to bring open models to enterprises in the U.K. and Europe, using liquid-cooled capacity powered by NVIDIA Blackwell Ultra GPUs at Era4's Chesterfield facility. Plans target financial services, insurance and life sciences.
- Mistral is expanding its NVIDIA infrastructure, including Vera Rubin planned for delivery in 2027. Mistral Large 4, its latest frontier multimodal model, was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral-operated European data centers.
- Verda plans to deploy Vera Rubin and already has NVIDIA Nemotron and NVIDIA Cosmos in production, using the NVIDIA Dynamo open source framework for inference optimization.
OpenShell Reference Architecture for agent boundaries
As developers connect models to tools and business systems, they need controls over what agents can access and do. NVIDIA announced the OpenShell Reference Architecture at GTC Berlin, with support from AI cloud partners. It builds on the NVIDIA OpenShell secure runtime software and is designed to give organizations full-stack control over agent deployments.
The architecture assigns roles across the hardware stack: Vera Rubin NVL72 for model inference, Vera CPUs with OpenShell for agent harnesses and tool use, and BlueField-4 DPUs running NVIDIA Sentry for in-silicon enforcement. NVIDIA says the pieces work together so agents can do what they are meant to without moving beyond the boundaries people set.
Supporting partners include Aleria, Baseten, Cassava, CoreWeave, Crusoe, DigitalOcean, Firebird, Fireworks, Firmus, G42, GMI Cloud, IREN, Lambda, Lightning AI, Nebius, Nscale, Radiant, Verda and Vultr.
What it means for builders
NVIDIA frames the announcements as three connected steps: expanded capacity gives builders more places to run workloads, model and inference services help them use it, and agent controls address what applications may do once in production. For European customers, the stated benefit is more choice and control over where workloads and data reside. Most figures, such as the Nscale and CodeRabbit results, come from NVIDIA's own account of partner tests and evaluations.
