NVIDIA says cloud partners are expanding regional AI infrastructure and services, while supporting an architecture designed to set controls for AI agents.
NVIDIA outlined new and expanding cloud infrastructure across Europe, the Nordics, Africa and Central Asia on October 21, 2026, spanning model training, inference and applications that use AI agents. The company’s announcement names partners including Nscale, Crusoe, Nebius, Mistral and Verda, with projects at different stages of deployment and development.
Nscale reported tests on NVIDIA GB300 NVL72 systems in which its use of NVIDIA DSX MaxLPS delivered 37% more GPUs and nearly 50% higher aggregate inference throughput within the same provisioned power budget. The company plans to deploy NVIDIA Vera Rubin infrastructure for training and production workloads, with expansion across five European markets. Crusoe is developing liquid-cooled, modular AI facilities for the Vera Rubin platform to bring inference capacity closer to users.
Other efforts address model services and specialized workloads. Nebius has adopted NVIDIA Groq 3 LPX for its Token Factory, integrating it with Vera Rubin NVL72 systems through Spectrum-X Ethernet. Baseten and CodeRabbit reported that fine-tuning and reinforcement learning with Nemotron 3.5 Lightning raised code-review routing agreement from 75.8% to 80.7%, while cutting estimated inference costs by 50% for the evaluated workload. Mistral says it trained its Large 4 model on 3,800 NVIDIA Grace Blackwell GPUs in European data centers; its planned Vera Rubin infrastructure is expected to arrive in 2027.
The announcement also describes a security design for agent deployments. NVIDIA’s OpenShell Reference Architecture combines Vera Rubin NVL72 for inference, Vera CPUs running OpenShell for agent orchestration and tool use, and BlueField-4 DPUs running NVIDIA Sentry for hardware-level enforcement. NVIDIA says partners including Nebius, Nscale, Crusoe and CoreWeave support the architecture. The framework is intended to give organizations controls over agent access to files, systems and tools as regional cloud providers add capacity for increasingly complex AI workloads.