Skip to content
Neural Daily

BusinessNews

NVIDIA cloud partners expand AI capacity in Europe

At GTC Berlin, NVIDIA's European cloud partners showed new capacity, services built on open models and a reference architecture for agent safety. The common thread is sovereignty: compute should sit close to the data.

By: Mikola Máté Patrik · · 3 min read

NVIDIA illustration: a row of AI server racks in front of a globe with Europe, the Nordics and Central Asia highlighted
Image: NVIDIA · Press kit · Official blog hero image · Source

What changed?

On October 21 at GTC Berlin, NVIDIA summed up what its European cloud partners are building (NVIDIA blog). The update covers Europe, the Nordics, Africa and Central Asia. It is organized around three themes: sovereign infrastructure, open models and safe agents.

Infrastructure: Vera Rubin and better power use

  • Nscale uses the NVIDIA DSX platform to match power allocation to workloads. In tests on GB300 NVL72 systems, DSX MaxLPS enabled 37% more GPUs and nearly 50% higher aggregate inference throughput within the same provisioned power budget. Nscale also plans to deploy Vera Rubin for frontier model training, expanding across five European markets.
  • Crusoe is developing Spark, liquid-cooled modular AI factories designed for Vera Rubin, to bring low-latency inference closer to users.
  • Mistral is expanding its NVIDIA infrastructure, with Vera Rubin systems planned for delivery in 2027. According to NVIDIA, Mistral Large 4 was trained on 3,800 Grace Blackwell GPUs in Mistral-operated European data centers.
  • Aleria, Cassava Technologies, Firebird, G42, Radiant, SWI Compute and Vultr are also expanding NVIDIA-powered capacity in the region.

Open models in practice

  • Nebius has adopted NVIDIA Groq 3 LPX, connected to Vera Rubin NVL72 over Spectrum-X networking on Nebius Token Factory, for faster token generation in latency-sensitive agentic apps. It also put three NVIDIA retail agent blueprints (inventory, store operations, fine-tuning) into early access.
  • Baseten built a code-review routing model from Nemotron 3.5 Lightning for CodeRabbit. After SFT and RL, routing agreement rose from 75.8% to 80.7%, while estimated inference cost fell by 50% on the evaluated workload.
  • Fireworks and Era4 are bringing open models to enterprises in the U.K. and Europe, using Blackwell Ultra GPUs at Era4's Chesterfield facility. Target sectors are financial services, insurance and life sciences.
  • Verda plans to deploy Vera Rubin. It already runs Nemotron and Cosmos in production and uses the open source Dynamo framework for inference optimization.

Agent safety: OpenShell Reference Architecture

We already mentioned the OpenShell runtime in our Agent Toolkit story. The new reference architecture defines what an agent can access and do, from silicon to software. It has three layers: Vera Rubin NVL72 for inference, Vera CPUs with OpenShell for agent harnesses and tool use, and BlueField-4 DPUs running NVIDIA Sentry for in-silicon enforcement.

NVIDIA lists 19 supporting partners, including CoreWeave, Lambda, Nebius, Nscale and DigitalOcean.

Why it matters

This is an ecosystem roundup more than a product launch. The numbers are the partners' own measurements, with no independent verification. The direction is clear though: European companies get more options to train and serve models in-region, close to their data. For teams moving agents into production, the OpenShell architecture signals that permission control now reaches down to the hardware. Worth checking whether your own cloud provider is on the list.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles