Databricks and NVIDIA expand collaboration for enterprise open-model customization and production deployment
Following·Day 5·2 sources
NVIDIA Announces European AI Infrastructure Push Across Hardware, Cloud, and Security
The gist
The announcements extend NVIDIA's reach across multiple layers of enterprise AI infrastructure in Europe, from hardware and inference software to secure deployment and model customization. European enterprises gain additional options for sovereign and on-premises AI deployment with maintained data control.
The full picture
NVIDIA published a cluster of announcements on October 20-21, 2026, covering inference speed gains on local hardware, new European cloud partnerships, expanded confidential computing options, Vera CPU adoption, and a Databricks collaboration for enterprise model customization.
On the inference side, new speculative decoding with MTP in llama.cpp delivers up to 1.4x faster end-to-end agentic tasks with Qwen 3.8 Flash Next on RTX PRO 6000 Blackwell, while new FlashInfer prefill kernels in vLLM improve Qwen3.8 27B agentic tasks by up to 1.3x on DGX Spark and GLM 5.3 Flash by up to 1.2x on DGX Station.
For European cloud, Fireworks and Era4 are partnering to provide open-model training and inference to UK and European enterprises in financial services, insurance, and life sciences, using Blackwell Ultra GPUs at Era4's Chesterfield facility. European AI cloud providers Nebius, Nscale, and Verda are building infrastructure on NVIDIA Vera CPUs, while server manufacturers including Bull, Dell Technologies, HPE, Lenovo, and Supermicro are building Vera-based systems for HPC and AI data centers. Data processing companies ClickHouse, Cloudera, Elastic, and SAP reported performance gains on Vera in early testing, and agent infrastructure providers Vercel, Modal, and E2B are building on Vera to power code execution environments.
On security, NVIDIA is expanding its Confidential Computing platform with a new Confidential Bare Metal option, bringing the total deployment options to three: Confidential Bare Metal, Confidential Containers for Kubernetes, and Confidential Virtual Machines. Support is expanding to Grace Blackwell NVL72, x86-based HGX Rubin NVL8, and DGX Rubin NVL8 systems. Cohere plans to integrate Confidential Bare Metal into Cohere Model Vault, and Google Cloud is extending Confidential Bare Metal to A4X Max instances powered by GB300 NVL72.
Databricks and NVIDIA are expanding their collaboration so enterprises can fine-tune open models on proprietary data and deploy them within the Databricks platform. The integration lets developers use NVIDIA NeMo, NGC containers, and Nemotron cookbooks on serverless NVIDIA GPUs, with enterprise data governance maintained throughout. Nemotron 3.5 Lightning and Super VL models are available on Databricks at launch.
How it developed
NVIDIA Vera CPU adoption by European cloud providers, server manufacturers, and agent infrastructure firms announced
Sources
NVIDIA Blog
- NVIDIA Accelerates llama.cpp and vLLM Inference for Local Agentic Workloads
- Fireworks and Era4 Partner to Bring Open Models to UK and European Enterprises
- NVIDIA Confidential Computing Expands, Enabling Frontier Model Builders to Securely Scale Across Global Infrastructure
- European Leaders Turn to NVIDIA Vera CPUs to Build Local Infrastructure
Related
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free